├── seq2seq_c ├── ops │ ├── __init__.py │ └── gen_beam_search_ops.py ├── python │ ├── __init__.py │ ├── __init__.pyc │ └── ops │ │ ├── loss.pyc │ │ ├── helper.pyc │ │ ├── __init__.pyc │ │ ├── decoder.pyc │ │ ├── basic_decoder.pyc │ │ ├── _beam_search_ops.so │ │ ├── beam_search_ops.pyc │ │ ├── attention_wrapper.pyc │ │ ├── beam_search_decoder.pyc │ │ ├── __init__.py │ │ ├── beam_search_ops.py │ │ ├── loss.py │ │ ├── basic_decoder.py │ │ ├── decoder.py │ │ └── helper.py ├── __init__.pyc └── __init__.py ├── README.md └── Trip ├── data ├── Osak_set.dat ├── Glas_set.dat ├── trajid.dat ├── TKY_split200_set.dat ├── Toro_set.dat ├── TKY_split400_set.dat ├── Melb_set.dat └── Edin_set.dat ├── origin_data ├── poi-Glas.csv ├── poi-TKY_split200.csv ├── poi-TKY_split400.csv ├── poi-Osak.csv ├── poi-Edin.csv ├── poi-Toro.csv └── poi-Melb.csv ├── metric.py ├── ops.py ├── AMSGrad.py └── gae_context.py /seq2seq_c/ops/__init__.py: -------------------------------------------------------------------------------- 1 | -------------------------------------------------------------------------------- /seq2seq_c/python/__init__.py: -------------------------------------------------------------------------------- 1 | -------------------------------------------------------------------------------- /seq2seq_c/__init__.pyc: -------------------------------------------------------------------------------- https://raw.githubusercontent.com/gcooq/DeepTrip/HEAD/seq2seq_c/__init__.pyc -------------------------------------------------------------------------------- /seq2seq_c/python/__init__.pyc: -------------------------------------------------------------------------------- https://raw.githubusercontent.com/gcooq/DeepTrip/HEAD/seq2seq_c/python/__init__.pyc -------------------------------------------------------------------------------- /seq2seq_c/python/ops/loss.pyc: -------------------------------------------------------------------------------- https://raw.githubusercontent.com/gcooq/DeepTrip/HEAD/seq2seq_c/python/ops/loss.pyc -------------------------------------------------------------------------------- /seq2seq_c/python/ops/helper.pyc: -------------------------------------------------------------------------------- https://raw.githubusercontent.com/gcooq/DeepTrip/HEAD/seq2seq_c/python/ops/helper.pyc -------------------------------------------------------------------------------- /seq2seq_c/python/ops/__init__.pyc: -------------------------------------------------------------------------------- https://raw.githubusercontent.com/gcooq/DeepTrip/HEAD/seq2seq_c/python/ops/__init__.pyc -------------------------------------------------------------------------------- /seq2seq_c/python/ops/decoder.pyc: -------------------------------------------------------------------------------- https://raw.githubusercontent.com/gcooq/DeepTrip/HEAD/seq2seq_c/python/ops/decoder.pyc -------------------------------------------------------------------------------- /seq2seq_c/python/ops/basic_decoder.pyc: -------------------------------------------------------------------------------- https://raw.githubusercontent.com/gcooq/DeepTrip/HEAD/seq2seq_c/python/ops/basic_decoder.pyc -------------------------------------------------------------------------------- /seq2seq_c/python/ops/_beam_search_ops.so: -------------------------------------------------------------------------------- https://raw.githubusercontent.com/gcooq/DeepTrip/HEAD/seq2seq_c/python/ops/_beam_search_ops.so -------------------------------------------------------------------------------- /seq2seq_c/python/ops/beam_search_ops.pyc: -------------------------------------------------------------------------------- https://raw.githubusercontent.com/gcooq/DeepTrip/HEAD/seq2seq_c/python/ops/beam_search_ops.pyc -------------------------------------------------------------------------------- /seq2seq_c/python/ops/attention_wrapper.pyc: -------------------------------------------------------------------------------- https://raw.githubusercontent.com/gcooq/DeepTrip/HEAD/seq2seq_c/python/ops/attention_wrapper.pyc -------------------------------------------------------------------------------- /seq2seq_c/python/ops/beam_search_decoder.pyc: -------------------------------------------------------------------------------- https://raw.githubusercontent.com/gcooq/DeepTrip/HEAD/seq2seq_c/python/ops/beam_search_decoder.pyc -------------------------------------------------------------------------------- /seq2seq_c/python/ops/__init__.py: -------------------------------------------------------------------------------- 1 | # Copyright 2017 The TensorFlow Authors. All Rights Reserved. 2 | # 3 | # Licensed under the Apache License, Version 2.0 (the "License"); 4 | # you may not use this file except in compliance with the License. 5 | # You may obtain a copy of the License at 6 | # 7 | # http://www.apache.org/licenses/LICENSE-2.0 8 | # 9 | # Unless required by applicable law or agreed to in writing, software 10 | # distributed under the License is distributed on an "AS IS" BASIS, 11 | # WITHOUT WARRANTIES OR CONDITIONS OF ANY KIND, either express or implied. 12 | # See the License for the specific language governing permissions and 13 | # limitations under the License. 14 | # ============================================================================== 15 | 16 | from __future__ import absolute_import 17 | from __future__ import division 18 | from __future__ import print_function 19 | -------------------------------------------------------------------------------- /README.md: -------------------------------------------------------------------------------- 1 | # DeepTrip
2 | source code for DeepTrip 3 | This is the implementation of the deep learning based DeepTrip model. 4 | 5 | # Environment 6 | * python >= 2.7 7 | * Tensorflow 1.8 or ++ 8 | 9 | # Dataset 10 | We use four different cities extrated from real-world Location-based Social Network data as follows. 11 | * Flickr: 12 | * Foursquare(Tokyo): 13 | * (remark) Please do not use these datasets for commercial purpose. For academic uses, please cite the paper. For Foursquare dataset, we use the same method as in [35] to construct the social networks. Thanks for their help.(see the reference [2,35] in the paper). 14 | # Usage 15 | * python gae_context.py 16 | 17 | # Reference 18 | Hope such an implementation could help you on your projects. Any comments and feedback are appreciated. 19 | 20 | * Gao, Qiang, Goce Trajcevski, Fan Zhou, Kunpeng Zhang, Ting Zhong, and Fengli Zhang. "DeepTrip: Adversarially Understanding Human Mobility for Trip Recommendation." In Proceedings of the 27th ACM SIGSPATIAL International Conference on Advances in Geographic Information Systems, pp. 444-447. 2019. 21 | -------------------------------------------------------------------------------- /seq2seq_c/python/ops/beam_search_ops.py: -------------------------------------------------------------------------------- 1 | # Copyright 2017 The TensorFlow Authors. All Rights Reserved. 2 | # 3 | # Licensed under the Apache License, Version 2.0 (the "License"); 4 | # you may not use this file except in compliance with the License. 5 | # You may obtain a copy of the License at 6 | # 7 | # http://www.apache.org/licenses/LICENSE-2.0 8 | # 9 | # Unless required by applicable law or agreed to in writing, software 10 | # distributed under the License is distributed on an "AS IS" BASIS, 11 | # WITHOUT WARRANTIES OR CONDITIONS OF ANY KIND, either express or implied. 12 | # See the License for the specific language governing permissions and 13 | # limitations under the License. 14 | # ============================================================================== 15 | """Beam Search helper ops.""" 16 | from __future__ import absolute_import 17 | from __future__ import division 18 | from __future__ import print_function 19 | 20 | from tensorflow.contrib.seq2seq.ops import gen_beam_search_ops 21 | from tensorflow.contrib.util import loader 22 | from tensorflow.python.platform import resource_loader 23 | 24 | _beam_search_ops_so = loader.load_op_library( 25 | resource_loader.get_path_to_datafile("_beam_search_ops.so")) 26 | 27 | gather_tree = gen_beam_search_ops.gather_tree 28 | -------------------------------------------------------------------------------- /Trip/data/Osak_set.dat: -------------------------------------------------------------------------------- 1 | 95157838@N00-1085 1 19 12 2 | 95157838@N00-1084 5 4 7 3 | 95157838@N00-1086 9 0 7 4 | 22669489@N00-86 6 1 0 3 7 5 | 43749456@N05-441 6 4 1 12 6 | 31484513@N00-263 8 19 7 7 | 97972855@N00-1101 4 1 12 7 8 | 85736811@N00-935 1 3 7 9 | 91041933@N00-989 5 4 20 10 | 71459982@N00-736 8 0 3 11 | 27921677@N00-181 5 0 3 12 | 38235150@N00-367 6 9 1 13 | 7183194@N02-744 16 19 12 0 7 10 14 | 51035737977@N01-574 4 1 0 15 | 8909846@N06-970 4 0 3 16 | 11354044@N05-8 9 21 12 7 17 | 22087824@N00-83 1 29 0 18 | 77526889@N00-820 6 9 4 18 0 19 | 53289105@N00-584 1 19 3 7 17 20 | 21147679@N08-69 0 7 17 21 | 14003952@N04-43 5 4 0 22 | 87807876@N00-955 16 12 0 3 7 23 | 63282110@N00-680 12 3 10 24 | 47957339@N00-487 5 8 11 25 | 12452841@N00-23 0 3 10 26 | 35391423@N00-324 6 9 5 1 27 | 64667184@N02-697 5 0 10 28 | 26678013@N00-171 6 9 11 29 | 12452841@N00-24 16 19 0 3 17 30 | 91609485@N00-1001 6 9 11 31 | 75595126@N00-801 4 13 7 32 | 68089733@N00-713 6 9 0 33 | 68089733@N00-712 6 9 5 34 | 44926895@N07-462 1 0 7 15 35 | 56944727@N00-626 4 1 0 36 | 49462908@N00-526 5 4 0 17 10 37 | 29989965@N00-204 6 9 1 0 7 38 | 94387617@N00-1069 0 3 17 39 | 10340578@N06-3 16 3 7 40 | 10307040@N08-2 4 1 3 7 41 | 82414749@N00-897 5 4 0 42 | 10340578@N06-4 6 9 1 43 | 87807876@N00-945 0 3 11 44 | 36330824413@N01-338 5 1 3 45 | 88016824@N00-963 6 9 1 46 | 94387617@N00-1028 5 3 11 47 | 9643030@N08-1094 5 1 12 7 48 | -------------------------------------------------------------------------------- /Trip/origin_data/poi-Glas.csv: -------------------------------------------------------------------------------- 1 | poiID,poiCat,poiLon,poiLat 2 | 1,Transport,-4.258101169014088,55.85792013480796 3 | 2,Transport,-4.250727638646297,55.861495672488964 4 | 4,Transport,-4.595632,55.509388 5 | 5,Transport,-4.331517,55.868896500000005 6 | 6,Transport,-4.269904625000001,55.85613520833331 7 | 7,Education,-4.242159871999993,55.861145432 8 | 8,Education,-4.249761639639642,55.86707233333342 9 | 9,Education,-4.285801843195263,55.872184307692386 10 | 10,Education,-4.288902204592891,55.871993964508796 11 | 11,Education,-4.264269125992071,55.866234078372784 12 | 12,Education,-4.25708622437137,55.86211197872341 13 | 13,Religion,-4.2344129376146915,55.86270032844042 14 | 14,Religion,-4.250646,55.853465421052626 15 | 15,Shopping,-4.292544210306379,55.87441117270186 16 | 16,Shopping,-4.255375836084915,55.86053728183968 17 | 17,Museum,-4.30770575,55.830818249999986 18 | 18,Park,-4.2056362195121935,55.849497292682905 19 | 19,Structure,-4.288323547169811,55.85990364420503 20 | 20,Structure,-4.25317754353561,55.86441777836411 21 | 21,Museum,-4.25158601294498,55.859934364077695 22 | 22,Park,-4.238215758620696,55.85119285775855 23 | 24,Structure,-4.309061802631581,55.853225565789465 24 | 25,Museum,-4.294561920863311,55.86907381294964 25 | 26,Museum,-4.290447339401815,55.86847013524039 26 | 27,Structure,-4.317774888888886,55.82766903030305 27 | 28,Structure,-4.261300118451011,55.864777521639816 28 | 29,Structure,-4.268838855855862,55.86140168468457 29 | -------------------------------------------------------------------------------- /Trip/origin_data/poi-TKY_split200.csv: -------------------------------------------------------------------------------- 1 | poiID,poiCat,poiLon,poiLat 2 | 5,Transport,139.7730182486,35.6985961953 3 | 8,Store,139.7746217251,35.699015417 4 | 11,Transport,139.7760584402,35.7134117258 5 | 21,Transport,139.7660815716,35.6813787223 6 | 28,Transport,139.770963192,35.7277712438 7 | 35,Transport,139.7002792266,35.6908026942 8 | 59,Transport,139.7020304203,35.658196312 9 | 71,Transport,139.7581327377,35.6666920514 10 | 77,Transport,139.6596515179,35.5757297602 11 | 104,Transport,139.5443956554,35.6518891133 12 | 212,Transport,139.56082955,35.7027062994 13 | 253,Transport,139.7013260345,35.658521727 14 | 260,Transport,139.7404289246,35.6301457297 15 | 261,Transport,139.7279191017,35.5884342009 16 | 263,Transport,139.7160503268,35.5624778239 17 | 284,Transport,139.7109555728,35.7298646365 18 | 297,Transport,139.7567421198,35.6556181358 19 | 336,Transport,139.7347489277,35.6071430623 20 | 351,Transport,139.7208380699,35.7780624397 21 | 428,Transport,139.7285527736,35.6196991729 22 | 449,Transport,139.6969020367,35.531326929 23 | 480,Transport,139.765222,35.699583 24 | 603,Transport,139.6657991409,35.7058285501 25 | 640,Transport,139.7038972378,35.7129024404 26 | 651,Transport,139.747568965,35.6457318132 27 | 1,Transport,139.8051452637,35.7496469435 28 | 765,Transport,139.7387123108,35.6282272324 29 | 875,Transport,139.5799040794,35.7032018165 30 | 1005,Transport,139.7380418533,35.6281616973 31 | 2151,Transport,139.5075917244,35.6037536336 32 | -------------------------------------------------------------------------------- /Trip/origin_data/poi-TKY_split400.csv: -------------------------------------------------------------------------------- 1 | poiID,poiCat,poiLon,poiLat 2 | 5,Transport,139.7730182486,35.6985961953 3 | 8,Store,139.7746217251,35.699015417 4 | 11,Transport,139.7760584402,35.7134117258 5 | 21,Transport,139.7660815716,35.6813787223 6 | 28,Transport,139.770963192,35.7277712438 7 | 35,Transport,139.7002792266,35.6908026942 8 | 59,Transport,139.7020304203,35.658196312 9 | 71,Transport,139.7581327377,35.6666920514 10 | 77,Transport,139.6596515179,35.5757297602 11 | 104,Transport,139.5443956554,35.6518891133 12 | 212,Transport,139.56082955,35.7027062994 13 | 253,Transport,139.7013260345,35.658521727 14 | 260,Transport,139.7404289246,35.6301457297 15 | 261,Transport,139.7279191017,35.5884342009 16 | 263,Transport,139.7160503268,35.5624778239 17 | 284,Transport,139.7109555728,35.7298646365 18 | 297,Transport,139.7567421198,35.6556181358 19 | 336,Transport,139.7347489277,35.6071430623 20 | 351,Transport,139.7208380699,35.7780624397 21 | 428,Transport,139.7285527736,35.6196991729 22 | 449,Transport,139.6969020367,35.531326929 23 | 480,Transport,139.765222,35.699583 24 | 603,Transport,139.6657991409,35.7058285501 25 | 640,Transport,139.7038972378,35.7129024404 26 | 651,Transport,139.747568965,35.6457318132 27 | 1,Transport,139.8051452637,35.7496469435 28 | 765,Transport,139.7387123108,35.6282272324 29 | 875,Transport,139.5799040794,35.7032018165 30 | 1005,Transport,139.7380418533,35.6281616973 31 | 2151,Transport,139.5075917244,35.6037536336 32 | -------------------------------------------------------------------------------- /Trip/origin_data/poi-Osak.csv: -------------------------------------------------------------------------------- 1 | poiID,poiCat,poiLon,poiLat 2 | 1,Amusement,135.4289382878105,34.65479175959366 3 | 2,Amusement,135.4306267944665,34.65614490909093 4 | 3,Amusement,135.50855763940035,34.65242308514182 5 | 4,Amusement,135.4336271953132,34.66475830078134 6 | 5,Amusement,135.4998940285709,34.703680330158626 7 | 6,Amusement,135.49013254999943,34.705322627272615 8 | 7,Park,135.50696077664924,34.69266442131975 9 | 8,Park,135.52582424865906,34.68653437337396 10 | 9,Park,135.48880599999998,34.612893 11 | 10,Park,135.51043665534092,34.65104219417449 12 | 11,Park,135.49385408196719,34.68545975409837 13 | 12,Park,135.52128527777776,34.61132941666667 14 | 15,Historical,135.5164392663755,34.65409199563325 15 | 16,Historical,135.49256448076923,34.612657000000006 16 | 17,Historical,135.52994,34.677756666666674 17 | 18,Historical,135.51246823934412,34.695621363934485 18 | 19,Entertainment,135.4796296190476,34.66778380952381 19 | 20,Entertainment,135.5009323832682,34.66742997665371 20 | 21,Entertainment,135.500556111111,34.67239257239053 21 | 22,Entertainment,135.5061961355934,34.652166481840126 22 | 23,Entertainment,135.50561435294117,34.660004392156864 23 | 24,Entertainment,135.49807975999988,34.67208495200003 24 | 25,Entertainment,135.48224743939394,34.72006954545454 25 | 26,Entertainment,139.76896139828057,35.675154257879626 26 | 27,Entertainment,135.501092,34.65772466666667 27 | 28,Entertainment,135.50244797297296,34.70325674324324 28 | 29,Entertainment,135.50583799999998,34.644641 29 | -------------------------------------------------------------------------------- /Trip/origin_data/poi-Edin.csv: -------------------------------------------------------------------------------- 1 | poiID,poiCat,poiLon,poiLat 2 | 1,Historical,-3.199862034580875,55.94884683716422 3 | 2,Historical,-3.1724675937052966,55.95276297997137 4 | 3,Museum,-3.190347155095568,55.9467227646501 5 | 4,Structure,-3.1751214683972875,55.95223434537238 6 | 5,Structure,-3.3887200833333337,55.999739000000005 7 | 6,Structure,-3.404049090909091,56.0012820909091 8 | 7,Park,-3.1618879074074004,55.94412091798946 9 | 8,Historical,-3.193193944356141,55.952538583465206 10 | 9,Cultural,-3.190637230008985,55.94980869047352 11 | 10,Park,-3.1829573739644994,55.95515047337272 12 | 11,Historical,-3.181937191191187,55.95460454054049 13 | 12,Cultural,-3.2044906274834797,55.94680475745124 14 | 13,Cultural,-3.1869824308822845,55.94635503896903 15 | 14,Structure,-3.209618379562047,55.94610982238432 16 | 15,Structure,-3.196984674391582,55.95182393800676 17 | 16,Museum,-3.195737522458634,55.951008898345115 18 | 17,Structure,-3.1902347191097684,55.948767777436686 19 | 18,Park,-3.202453709104388,55.95054091561817 20 | 19,Structure,-3.1892391019884583,55.95262047466306 21 | 20,Park,-3.207734793416563,55.96588335187255 22 | 21,Structure,-3.172436005988025,55.977925371257456 23 | 22,Structure,-3.2076172933754026,55.95175426182978 24 | 23,Structure,-3.1847238000000044,55.95370758124997 25 | 24,Entertainment,-3.1742341271393735,55.95074926405848 26 | 25,Historical,-3.2392285999999997,55.9184182 27 | 27,Museum,-3.2271948343195236,55.95087610650889 28 | 28,Structure,-3.1868621630188576,55.948245795471635 29 | 29,Structure,-3.1952324337146703,55.94908885935436 30 | -------------------------------------------------------------------------------- /Trip/origin_data/poi-Toro.csv: -------------------------------------------------------------------------------- 1 | poiID,poiCat,poiLon,poiLat 2 | 1,Sport,-79.379243379063,43.64318250142281 3 | 2,Sport,-79.41863385714308,43.63277161740573 4 | 3,Sport,-79.3800452078488,43.66217527180223 5 | 4,Sport,-79.38928974410585,43.641297368457444 6 | 6,Cultural,-79.39239594093709,43.65366205295312 7 | 7,Cultural,-79.37732735752459,43.647150975158695 8 | 8,Cultural,-79.3853489798394,43.64238526267252 9 | 9,Cultural,-79.33916950000024,43.71644713669081 10 | 10,Cultural,-79.36110748265887,43.66706691907516 11 | 11,Cultural,-79.3944584713709,43.66718257115105 12 | 12,Cultural,-79.18221148856527,43.82008013305617 13 | 13,Cultural,-79.40936380912937,43.67815669398353 14 | 14,Amusement,-79.41700068794327,43.633924191489356 15 | 15,Amusement,-79.3735729115044,43.61983586725666 16 | 16,Amusement,-79.38706547902999,43.642849137630805 17 | 17,Amusement,-79.41601125371233,43.63256282673258 18 | 18,Amusement,-79.45080761538462,43.6371833846154 19 | 19,Beach,-79.3782267477478,43.62169749549551 20 | 20,Beach,-79.46238176404492,43.64655716853935 21 | 21,Beach,-79.38045327151268,43.65627400696254 22 | 22,Beach,-79.38370190118991,43.652478209270036 23 | 23,Shopping,-79.3798840401936,43.653868152197106 24 | 24,Shopping,-79.38232002042837,43.63862138326839 25 | 25,Shopping,-79.4011678236331,43.65474784303318 26 | 26,Shopping,-79.45268321153846,43.725760711538456 27 | 27,Shopping,-79.3909345777776,43.67010353968252 28 | 28,Structure,-79.38118417735065,43.65218102670948 29 | 29,Structure,-79.39126472274569,43.66213848048437 30 | 30,Structure,-79.3805837906372,43.64565054031221 31 | -------------------------------------------------------------------------------- /Trip/metric.py: -------------------------------------------------------------------------------- 1 | import numpy as np 2 | 3 | def calc_F1(traj_act, traj_rec, noloop=False): 4 | '''Compute recall, precision and F1 for recommended trajectories''' 5 | assert (isinstance(noloop, bool)) 6 | assert (len(traj_act) > 0) 7 | assert (len(traj_rec) > 0) 8 | 9 | if noloop == True: 10 | intersize = len(set(traj_act) & set(traj_rec)) 11 | else: 12 | match_tags = np.zeros(len(traj_act), dtype=np.bool) 13 | for poi in traj_rec: 14 | for j in range(len(traj_act)): 15 | if match_tags[j] == False and poi == traj_act[j]: 16 | match_tags[j] = True 17 | break 18 | intersize = np.nonzero(match_tags)[0].shape[0] 19 | 20 | recall = intersize * 1.0 / len(traj_act) 21 | precision = intersize * 1.0 / len(traj_rec) 22 | Denominator=recall+precision 23 | if Denominator==0: 24 | Denominator=1 25 | F1 = 2 * precision * recall * 1.0 / Denominator 26 | return F1 27 | 28 | 29 | # cpdef float calc_pairsF1(y, y_hat): 30 | def calc_pairsF1(y, y_hat): 31 | assert (len(y) > 0) 32 | # assert (len(y) == len(set(y))) # no loops in y 33 | # cdef int n, nr, nc, poi1, poi2, i, j 34 | # cdef double n0, n0r 35 | n = len(y) 36 | nr = len(y_hat) 37 | n0 = n * (n - 1) / 2 38 | n0r = nr * (nr - 1) / 2 39 | 40 | # y determines the correct visiting order 41 | order_dict = dict() 42 | for i in range(n): 43 | order_dict[y[i]] = i 44 | 45 | nc = 0 46 | for i in range(nr): 47 | poi1 = y_hat[i] 48 | for j in range(i + 1, nr): 49 | poi2 = y_hat[j] 50 | if poi1 in order_dict and poi2 in order_dict and poi1 != poi2: 51 | if order_dict[poi1] < order_dict[poi2]: nc += 1 52 | 53 | 54 | precision = (1.0 * nc) / (1.0 * n0r) 55 | recall = (1.0 * nc) / (1.0 * n0) 56 | if nc == 0: 57 | F1 = 0 58 | else: 59 | F1 = 2. * precision * recall / (precision + recall) 60 | return float(F1) 61 | 62 | def calc_pairsF12(y, y_hat): 63 | f1=0 64 | for i in range(len(y)): 65 | if (y[i]==y_hat[i]): 66 | f1+=1 67 | return float(float(f1)/float(len(y))) #float type 68 | -------------------------------------------------------------------------------- /seq2seq_c/__init__.py: -------------------------------------------------------------------------------- 1 | # Copyright 2015 The TensorFlow Authors. All Rights Reserved. 2 | # 3 | # Licensed under the Apache License, Version 2.0 (the "License"); 4 | # you may not use this file except in compliance with the License. 5 | # You may obtain a copy of the License at 6 | # 7 | # http://www.apache.org/licenses/LICENSE-2.0 8 | # 9 | # Unless required by applicable law or agreed to in writing, software 10 | # distributed under the License is distributed on an "AS IS" BASIS, 11 | # WITHOUT WARRANTIES OR CONDITIONS OF ANY KIND, either express or implied. 12 | # See the License for the specific language governing permissions and 13 | # limitations under the License. 14 | # ============================================================================== 15 | 16 | """Ops for building neural network seq2seq decoders and losses. 17 | 18 | See the @{$python/contrib.seq2seq} guide. 19 | """ 20 | 21 | from __future__ import absolute_import 22 | from __future__ import division 23 | from __future__ import print_function 24 | 25 | # pylint: disable=unused-import,wildcard-import,line-too-long 26 | from seq2seq_c.python.ops.attention_wrapper import * 27 | from seq2seq_c.python.ops.basic_decoder import * 28 | from seq2seq_c.python.ops.beam_search_decoder import * 29 | from seq2seq_c.python.ops.beam_search_ops import * 30 | from seq2seq_c.python.ops.decoder import * 31 | from seq2seq_c.python.ops.helper import * 32 | from seq2seq_c.python.ops.loss import * 33 | from tensorflow.python.util.all_util import remove_undocumented 34 | # pylint: enable=unused-import,widcard-import,line-too-long 35 | 36 | _allowed_symbols = [ 37 | "sequence_loss", 38 | "Decoder", 39 | "dynamic_decode", 40 | "BasicDecoder", 41 | "BasicDecoderOutput", 42 | "BeamSearchDecoder", 43 | "BeamSearchDecoderOutput", 44 | "BeamSearchDecoderState", 45 | "Helper", 46 | "CustomHelper", 47 | "FinalBeamSearchDecoderOutput", 48 | "gather_tree", 49 | "GreedyEmbeddingHelper", 50 | "GreedyEmbeddingHelper2", 51 | "InferenceHelper", 52 | "SampleEmbeddingHelper", 53 | "ScheduledEmbeddingTrainingHelper", 54 | "ScheduledOutputTrainingHelper", 55 | "TrainingHelper", 56 | "BahdanauAttention", 57 | "LuongAttention", 58 | "hardmax", 59 | "AttentionWrapperState", 60 | "AttentionWrapper", 61 | "AttentionMechanism", 62 | "tile_batch", 63 | "safe_cumprod", 64 | "monotonic_attention", 65 | "monotonic_probability_fn", 66 | "BahdanauMonotonicAttention", 67 | "LuongMonotonicAttention", 68 | ] 69 | 70 | 71 | remove_undocumented(__name__, _allowed_symbols) 72 | -------------------------------------------------------------------------------- /Trip/data/Glas_set.dat: -------------------------------------------------------------------------------- 1 | 55329285@N05-1400 3 13 2 2 | 72746018@N00-1716 2 6 8 3 | 72746018@N00-1713 6 1 8 4 4 | 43002463@N00-1162 13 2 7 5 | 7797604@N05-1845 2 8 7 6 | 28370443@N08-539 0 16 6 9 11 1 4 10 7 | 24616128@N00-479 0 11 8 8 | 33278177@N00-790 3 19 1 4 9 | 84272941@N00-1976 0 1 4 10 | 89182227@N00-2048 13 2 5 11 | 30024546@N04-671 0 9 8 12 | 20483509@N00-238 0 13 7 13 | 22087304@N07-390 1 4 7 14 | 11452054@N00-55 0 11 10 15 | 16312054@N06-154 3 9 1 8 4 16 | 11037560@N04-39 0 1 4 17 | 74937968@N00-1774 3 1 8 18 | 9828412@N05-2204 6 7 10 19 | 56846157@N00-1426 2 6 5 1 4 20 | 11037560@N04-38 0 11 4 21 | 32714545@N05-781 11 22 1 22 | 32693718@N07-780 3 0 1 23 | 51207064@N00-1324 0 9 8 24 | 30024546@N04-666 9 8 10 25 | 33563858@N00-796 11 8 10 26 | 30024546@N04-588 2 9 14 27 | 62798459@N00-1495 24 9 1 28 | 17358831@N00-178 6 1 10 29 | 91685781@N00-2081 0 1 4 30 | 70938076@N00-1676 0 1 8 31 | 51168133@N00-1305 0 2 11 1 4 32 | 112042056@N06-49 0 2 4 7 33 | 17643132@N00-184 0 11 1 4 34 | 35761150@N04-977 0 1 17 35 | 35761150@N04-976 3 13 2 5 1 4 7 36 | 82619366@N00-1910 0 2 9 37 | 82619366@N00-1911 13 2 5 7 38 | 55329285@N05-1383 13 2 5 39 | 92367195@N00-2106 23 7 20 40 | 55898913@N04-1416 3 0 19 1 7 41 | 35034349172@N01-900 2 5 14 42 | 82103247@N00-1897 0 1 4 43 | 16516058@N03-159 3 0 9 1 4 44 | 82103247@N00-1899 1 8 4 45 | 82103247@N00-1898 1 8 4 46 | 56889878@N03-1429 0 9 10 47 | 14578360@N00-119 13 2 5 48 | 7512717@N06-1779 0 9 1 8 4 49 | 21047061@N04-248 3 0 1 50 | 85397137@N00-1992 3 1 8 17 51 | 63651050@N00-1507 3 6 11 52 | 22087304@N07-386 0 13 2 4 53 | 41418909@N06-1124 0 1 4 54 | 34427470616@N01-814 3 1 10 55 | 34427470616@N01-815 3 13 8 10 56 | 30985799@N00-698 0 8 7 57 | 35468159247@N01-958 0 13 2 14 58 | 10482493@N05-31 0 1 8 4 59 | 9286726@N03-2143 3 0 11 4 60 | 7512717@N06-1780 3 6 1 61 | 7868489@N05-1852 0 11 10 62 | 57879122@N00-1444 0 1 4 10 63 | 8503026@N04-1985 3 0 13 2 7 64 | 73429256@N00-1733 3 11 1 65 | 41894174966@N01-1128 3 0 4 66 | 12982579@N00-90 6 8 14 7 10 67 | 38166049@N00-1084 11 8 7 68 | 16497759@N07-157 6 12 10 69 | 45072550@N03-1223 0 13 11 1 4 70 | 52686644@N04-1336 3 9 1 8 10 71 | 57474170@N05-1441 0 12 8 4 72 | 30671962@N05-683 0 25 4 73 | 29797746@N08-576 3 6 9 14 7 10 17 74 | 7587786@N05-1797 3 0 4 75 | 70284683@N06-1560 24 1 4 76 | 51207064@N00-1317 0 1 4 77 | 30511496@N00-681 0 6 11 78 | 14086506@N02-107 3 0 9 22 1 4 79 | 8286908@N02-1951 0 2 11 80 | 38826189@N00-1085 3 13 6 1 81 | 7512717@N06-1778 0 1 8 4 82 | 49253583@N02-1294 6 9 10 83 | 9511818@N03-2156 3 1 4 84 | 44232489@N02-1216 3 0 8 85 | 30713600@N00-685 3 0 1 86 | 23363966@N02-454 0 12 7 87 | 31438571@N07-733 0 1 4 88 | 48001787@N06-1250 0 8 7 89 | 31803571@N02-762 0 11 15 90 | 82103247@N00-1901 3 0 1 4 91 | 22751315@N05-419 3 0 4 92 | 25396215@N00-484 0 6 4 93 | 65244239@N00-1518 12 14 7 94 | 22087304@N07-401 23 7 20 95 | 35468159247@N01-965 2 5 12 96 | 98833136@N00-2214 0 11 4 97 | 35468152754@N01-923 3 0 1 4 98 | 37622685@N02-1076 13 2 7 99 | 44124372821@N01-1211 0 11 7 10 100 | 7883724@N05-1853 0 11 4 101 | 85217508@N00-1988 0 6 11 1 4 20 102 | 70544918@N00-1633 9 5 4 103 | 31564588@N06-760 0 1 4 104 | 35034349172@N01-899 2 5 14 7 105 | 35034349172@N01-898 2 14 7 106 | 22087304@N07-402 5 14 7 107 | 30024546@N04-644 6 9 7 108 | 49998283@N00-1298 0 6 8 10 109 | 72579404@N00-1712 0 1 8 110 | 49998283@N00-1297 9 11 10 111 | 17424601@N00-179 3 0 4 112 | 34995267@N00-827 9 10 17 113 | -------------------------------------------------------------------------------- /Trip/origin_data/poi-Melb.csv: -------------------------------------------------------------------------------- 1 | poiID,poiCat,poiLat,poiLon 2 | 0,City precincts,-37.821670000000005,144.96778 3 | 1,City precincts,-37.817,144.946 4 | 2,City precincts,-37.8119,144.97299999999998 5 | 3,City precincts,-37.79972,144.96694 6 | 4,City precincts,-37.80778,144.96333 7 | 5,City precincts,-37.829,144.957 8 | 6,City precincts,-37.81667,144.98333 9 | 7,City precincts,-37.7963,144.9614 10 | 8,Shopping,-37.81583,144.96444 11 | 9,Shopping,-37.8139,144.96452 12 | 10,Shopping,-37.817563899999996,144.99032780000002 13 | 11,Shopping,-37.79833,144.978472 14 | 12,Shopping,-37.84083,144.99528 15 | 13,Shopping,-37.816583,144.9655 16 | 14,Shopping,-37.8168,144.9622 17 | 15,Shopping,-37.8175,144.96583 18 | 16,Shopping,-37.8096,144.961 19 | 17,Shopping,-37.81694,144.96556 20 | 18,Shopping,-37.8151,144.9641 21 | 19,Shopping,-37.810559999999995,144.96 22 | 20,Shopping,-37.847414,144.993503 23 | 21,Shopping,-37.806965999999996,144.956693 24 | 22,Shopping,-37.81,144.9652 25 | 23,Shopping,-37.8145,144.9641 26 | 24,Shopping,-37.8064,144.963 27 | 25,Entertainment,-37.822324,144.96009899999999 28 | 26,Entertainment,-37.868036,144.976369 29 | 27,Entertainment,-37.821044,144.95801699999998 30 | 28,Entertainment,-37.784762,144.95209499999999 31 | 29,Public galleries,-37.826496999999996,144.96701399999998 32 | 30,Public galleries,-37.7975,144.964111 33 | 31,Public galleries,-37.822595,144.968634 34 | 32,Institutions,-37.81384,144.963028 35 | 33,Institutions,-37.827939,144.976939 36 | 34,Institutions,-37.8297,144.9755 37 | 35,Institutions,-37.815145,144.966777 38 | 36,Institutions,-37.80806,144.96528 39 | 37,Institutions,-37.8087,144.966 40 | 38,Institutions,-37.812153,144.956794 41 | 39,Institutions,-37.813153,144.974121 42 | 40,Institutions,-37.811054999999996,144.97329 43 | 41,Institutions,-37.809801,144.964787 44 | 42,Institutions,-37.814132,144.957932 45 | 43,Institutions,-37.827391,144.970447 46 | 44,Structures,-37.82028,144.96833 47 | 45,Structures,-37.821670000000005,144.96444 48 | 46,Structures,-37.8256,144.9541 49 | 47,Structures,-37.8110723,144.9368763 50 | 48,Structures,-37.804728000000004,144.971225 51 | 49,Structures,-37.830434000000004,144.97325800000002 52 | 50,Structures,-37.816853,144.967384 53 | 51,Structures,-37.81,144.97611 54 | 52,Sports stadiums,-37.82528,144.98389 55 | 53,Sports stadiums,-37.816390000000006,144.9475 56 | 54,Sports stadiums,-37.790279999999996,144.9125 57 | 55,Sports stadiums,-37.821112,144.97741000000002 58 | 56,Sports stadiums,-37.82278,144.98167 59 | 57,Sports stadiums,-37.82,144.98333 60 | 58,Sports stadiums,-37.84972,144.96833 61 | 59,Sports stadiums,-37.84667,144.96611000000001 62 | 60,Sports stadiums,-37.84028,144.965 63 | 61,Sports stadiums,-37.8123777,144.93556740000002 64 | 62,Sports stadiums,-37.82472,144.98111 65 | 63,Sports stadiums,-37.821670000000005,144.97833 66 | 64,Sports stadiums,-37.97,145.03 67 | 65,Sports stadiums,-37.78389,144.96167 68 | 66,Parks and spaces,-37.846,144.97 69 | 67,Parks and spaces,-37.820370000000004,144.971938 70 | 68,Parks and spaces,-37.818329999999996,144.9725 71 | 69,Parks and spaces,-37.80611,144.97028 72 | 70,Parks and spaces,-37.816,144.967 73 | 71,Parks and spaces,-37.817797999999996,144.96871399999998 74 | 72,Parks and spaces,-37.8125641,144.9803925 75 | 73,Parks and spaces,-37.8105,144.9544 76 | 74,Parks and spaces,-37.8282598,144.97758149999999 77 | 75,Parks and spaces,-37.8218,144.9716 78 | 76,Parks and spaces,-37.8334,144.98033 79 | 77,Parks and spaces,-37.790183299999995,144.9511667 80 | 78,Parks and spaces,-37.814158,144.976194 81 | 79,Parks and spaces,-37.818043700000004,144.9852312 82 | 80,Transport,-37.8194778,144.932125 83 | 81,Transport,-37.819744,144.968516 84 | 82,Transport,-37.818078,144.96681 85 | 83,Transport,-37.67333,144.84333 86 | 84,Transport,-37.81,144.96278 87 | 85,Transport,-37.8184,144.9524 88 | 86,Transport,-37.829440000000005,144.89806000000002 89 | 87,Transport,-37.85194,144.90833 90 | -------------------------------------------------------------------------------- /seq2seq_c/python/ops/loss.py: -------------------------------------------------------------------------------- 1 | # Copyright 2016 The TensorFlow Authors. All Rights Reserved. 2 | # 3 | # Licensed under the Apache License, Version 2.0 (the "License"); 4 | # you may not use this file except in compliance with the License. 5 | # You may obtain a copy of the License at 6 | # 7 | # http://www.apache.org/licenses/LICENSE-2.0 8 | # 9 | # Unless required by applicable law or agreed to in writing, software 10 | # distributed under the License is distributed on an "AS IS" BASIS, 11 | # WITHOUT WARRANTIES OR CONDITIONS OF ANY KIND, either express or implied. 12 | # See the License for the specific language governing permissions and 13 | # limitations under the License. 14 | # ============================================================================== 15 | """Seq2seq loss operations for use in sequence models. 16 | """ 17 | 18 | from __future__ import absolute_import 19 | from __future__ import division 20 | from __future__ import print_function 21 | 22 | from tensorflow.python.framework import ops 23 | from tensorflow.python.ops import array_ops 24 | from tensorflow.python.ops import math_ops 25 | from tensorflow.python.ops import nn_ops 26 | 27 | __all__ = ["sequence_loss"] 28 | 29 | 30 | def sequence_loss(logits, 31 | targets, 32 | weights, 33 | average_across_timesteps=True, 34 | average_across_batch=True, 35 | softmax_loss_function=None, 36 | name=None): 37 | """Weighted cross-entropy loss for a sequence of logits. 38 | 39 | Depending on the values of `average_across_timesteps` and 40 | `average_across_batch`, the return Tensor will have rank 0, 1, or 2 as these 41 | arguments reduce the cross-entropy at each target, which has shape 42 | `[batch_size, sequence_length]`, over their respective dimensions. For 43 | example, if `average_across_timesteps` is `True` and `average_across_batch` 44 | is `False`, then the return Tensor will have shape `[batch_size]`. 45 | 46 | Args: 47 | logits: A Tensor of shape 48 | `[batch_size, sequence_length, num_decoder_symbols]` and dtype float. 49 | The logits correspond to the prediction across all classes at each 50 | timestep. 51 | targets: A Tensor of shape `[batch_size, sequence_length]` and dtype 52 | int. The target represents the true class at each timestep. 53 | weights: A Tensor of shape `[batch_size, sequence_length]` and dtype 54 | float. `weights` constitutes the weighting of each prediction in the 55 | sequence. When using `weights` as masking, set all valid timesteps to 1 56 | and all padded timesteps to 0, e.g. a mask returned by `tf.sequence_mask`. 57 | average_across_timesteps: If set, sum the cost across the sequence 58 | dimension and divide the cost by the total label weight across timesteps. 59 | average_across_batch: If set, sum the cost across the batch dimension and 60 | divide the returned cost by the batch size. 61 | softmax_loss_function: Function (labels, logits) -> loss-batch 62 | to be used instead of the standard softmax (the default if this is None). 63 | **Note that to avoid confusion, it is required for the function to accept 64 | named arguments.** 65 | name: Optional name for this operation, defaults to "sequence_loss". 66 | 67 | Returns: 68 | A float Tensor of rank 0, 1, or 2 depending on the 69 | `average_across_timesteps` and `average_across_batch` arguments. By default, 70 | it has rank 0 (scalar) and is the weighted average cross-entropy 71 | (log-perplexity) per symbol. 72 | 73 | Raises: 74 | ValueError: logits does not have 3 dimensions or targets does not have 2 75 | dimensions or weights does not have 2 dimensions. 76 | """ 77 | if len(logits.get_shape()) != 3: 78 | raise ValueError("Logits must be a " 79 | "[batch_size x sequence_length x logits] tensor") 80 | if len(targets.get_shape()) != 2: 81 | raise ValueError("Targets must be a [batch_size x sequence_length] " 82 | "tensor") 83 | if len(weights.get_shape()) != 2: 84 | raise ValueError("Weights must be a [batch_size x sequence_length] " 85 | "tensor") 86 | with ops.name_scope(name, "sequence_loss", [logits, targets, weights]): 87 | num_classes = array_ops.shape(logits)[2] 88 | logits_flat = array_ops.reshape(logits, [-1, num_classes]) 89 | targets = array_ops.reshape(targets, [-1]) 90 | if softmax_loss_function is None: 91 | crossent = nn_ops.sparse_softmax_cross_entropy_with_logits( 92 | labels=targets, logits=logits_flat) 93 | else: 94 | crossent = softmax_loss_function(labels=targets, logits=logits_flat) 95 | crossent *= array_ops.reshape(weights, [-1]) 96 | if average_across_timesteps and average_across_batch: 97 | crossent = math_ops.reduce_sum(crossent) 98 | total_size = math_ops.reduce_sum(weights) 99 | total_size += 1e-12 # to avoid division by 0 for all-0 weights 100 | crossent /= total_size 101 | else: 102 | batch_size = array_ops.shape(logits)[0] 103 | sequence_length = array_ops.shape(logits)[1] 104 | crossent = array_ops.reshape(crossent, [batch_size, sequence_length]) 105 | if average_across_timesteps and not average_across_batch: 106 | crossent = math_ops.reduce_sum(crossent, axis=[1]) 107 | total_size = math_ops.reduce_sum(weights, axis=[1]) 108 | total_size += 1e-12 # to avoid division by 0 for all-0 weights 109 | crossent /= total_size 110 | if not average_across_timesteps and average_across_batch: 111 | crossent = math_ops.reduce_sum(crossent, axis=[0]) 112 | total_size = math_ops.reduce_sum(weights, axis=[0]) 113 | total_size += 1e-12 # to avoid division by 0 for all-0 weights 114 | crossent /= total_size 115 | return crossent 116 | -------------------------------------------------------------------------------- /seq2seq_c/ops/gen_beam_search_ops.py: -------------------------------------------------------------------------------- 1 | """Python wrappers around TensorFlow ops. 2 | 3 | This file is MACHINE GENERATED! Do not edit. 4 | Original C++ source file: beam_search_ops.cc 5 | """ 6 | 7 | import collections as _collections 8 | import six as _six 9 | 10 | from tensorflow.python import pywrap_tensorflow as _pywrap_tensorflow 11 | from tensorflow.python.eager import context as _context 12 | from tensorflow.python.eager import core as _core 13 | from tensorflow.python.eager import execute as _execute 14 | from tensorflow.python.framework import dtypes as _dtypes 15 | from tensorflow.python.framework import errors as _errors 16 | from tensorflow.python.framework import tensor_shape as _tensor_shape 17 | 18 | from tensorflow.core.framework import op_def_pb2 as _op_def_pb2 19 | # Needed to trigger the call to _set_call_cpp_shape_fn. 20 | from tensorflow.python.framework import common_shapes as _common_shapes 21 | from tensorflow.python.framework import op_def_registry as _op_def_registry 22 | from tensorflow.python.framework import ops as _ops 23 | from tensorflow.python.framework import op_def_library as _op_def_library 24 | from tensorflow.python.util.tf_export import tf_export 25 | 26 | 27 | @tf_export('gather_tree') 28 | def gather_tree(step_ids, parent_ids, max_sequence_lengths, end_token, name=None): 29 | r"""Calculates the full beams from the per-step ids and parent beam ids. 30 | 31 | On CPU, if an out of bound parent id is found, an error is returned. 32 | On GPU, if an out of bound parent id is found, a -1 is stored in the 33 | corresponding output value and the execution for that beam returns early. 34 | 35 | For a given beam, past the time step containing the first decoded `end_token` 36 | all values are filled in with `end_token`. 37 | 38 | TODO(ebrevdo): fill in the remainder of this docstring. 39 | 40 | Args: 41 | step_ids: A `Tensor`. Must be one of the following types: `int32`. 42 | `[max_time, batch_size, beam_width]`. 43 | parent_ids: A `Tensor`. Must have the same type as `step_ids`. 44 | `[max_time, batch_size, beam_width]`. 45 | max_sequence_lengths: A `Tensor` of type `int32`. `[batch_size]`. 46 | end_token: A `Tensor`. Must have the same type as `step_ids`. `[]`. 47 | name: A name for the operation (optional). 48 | 49 | Returns: 50 | A `Tensor`. Has the same type as `step_ids`. 51 | `[max_time, batch_size, beam_width]`. 52 | """ 53 | _ctx = _context._context 54 | if _ctx is None or not _ctx._eager_context.is_eager: 55 | _, _, _op = _op_def_lib._apply_op_helper( 56 | "GatherTree", step_ids=step_ids, parent_ids=parent_ids, 57 | max_sequence_lengths=max_sequence_lengths, end_token=end_token, 58 | name=name) 59 | _result = _op.outputs[:] 60 | _inputs_flat = _op.inputs 61 | _attrs = ("T", _op.get_attr("T")) 62 | _execute.record_gradient( 63 | "GatherTree", _inputs_flat, _attrs, _result, name) 64 | _result, = _result 65 | return _result 66 | 67 | else: 68 | try: 69 | _result = _pywrap_tensorflow.TFE_Py_FastPathExecute( 70 | _ctx._context_handle, _ctx._eager_context.device_name, "GatherTree", 71 | name, _ctx._post_execution_callbacks, step_ids, parent_ids, 72 | max_sequence_lengths, end_token) 73 | return _result 74 | except _core._FallbackException: 75 | return gather_tree_eager_fallback( 76 | step_ids, parent_ids, max_sequence_lengths, end_token, name=name, 77 | ctx=_ctx) 78 | except _core._NotOkStatusException as e: 79 | if name is not None: 80 | message = e.message + " name: " + name 81 | else: 82 | message = e.message 83 | _six.raise_from(_core._status_to_exception(e.code, message), None) 84 | 85 | 86 | def gather_tree_eager_fallback(step_ids, parent_ids, max_sequence_lengths, end_token, name=None, ctx=None): 87 | r"""This is the slowpath function for Eager mode. 88 | This is for function gather_tree 89 | """ 90 | _ctx = ctx if ctx else _context.context() 91 | _attr_T, _inputs_T = _execute.args_to_matching_eager([step_ids, parent_ids, end_token], _ctx) 92 | (step_ids, parent_ids, end_token) = _inputs_T 93 | max_sequence_lengths = _ops.convert_to_tensor(max_sequence_lengths, _dtypes.int32) 94 | _inputs_flat = [step_ids, parent_ids, max_sequence_lengths, end_token] 95 | _attrs = ("T", _attr_T) 96 | _result = _execute.execute(b"GatherTree", 1, inputs=_inputs_flat, 97 | attrs=_attrs, ctx=_ctx, name=name) 98 | _execute.record_gradient( 99 | "GatherTree", _inputs_flat, _attrs, _result, name) 100 | _result, = _result 101 | return _result 102 | 103 | _ops.RegisterShape("GatherTree")(None) 104 | 105 | def _InitOpDefLibrary(op_list_proto_bytes): 106 | op_list = _op_def_pb2.OpList() 107 | op_list.ParseFromString(op_list_proto_bytes) 108 | _op_def_registry.register_op_list(op_list) 109 | op_def_lib = _op_def_library.OpDefLibrary() 110 | op_def_lib.add_op_list(op_list) 111 | return op_def_lib 112 | # op { 113 | # name: "GatherTree" 114 | # input_arg { 115 | # name: "step_ids" 116 | # type_attr: "T" 117 | # } 118 | # input_arg { 119 | # name: "parent_ids" 120 | # type_attr: "T" 121 | # } 122 | # input_arg { 123 | # name: "max_sequence_lengths" 124 | # type: DT_INT32 125 | # } 126 | # input_arg { 127 | # name: "end_token" 128 | # type_attr: "T" 129 | # } 130 | # output_arg { 131 | # name: "beams" 132 | # type_attr: "T" 133 | # } 134 | # attr { 135 | # name: "T" 136 | # type: "type" 137 | # allowed_values { 138 | # list { 139 | # type: DT_INT32 140 | # } 141 | # } 142 | # } 143 | # } 144 | _op_def_lib = _InitOpDefLibrary(b"\nt\n\nGatherTree\022\r\n\010step_ids\"\001T\022\017\n\nparent_ids\"\001T\022\030\n\024max_sequence_lengths\030\003\022\016\n\tend_token\"\001T\032\n\n\005beams\"\001T\"\020\n\001T\022\004type:\005\n\0032\001\003") 145 | -------------------------------------------------------------------------------- /Trip/ops.py: -------------------------------------------------------------------------------- 1 | import tensorflow as tf 2 | 3 | def fully_connected(input, output_shape, initializer, scope="fc", is_last=False, is_decoder=False): 4 | with tf.variable_scope(scope): 5 | input_shape = input.get_shape()[-1].value 6 | W = tf.get_variable("weight", [input_shape, output_shape], initializer=initializer) 7 | b = tf.get_variable("bias", [output_shape], initializer=initializer) 8 | fc = tf.add(tf.matmul(input, W), b) 9 | fc = normalize(fc) 10 | 11 | if not is_last: 12 | if not is_decoder: 13 | output = tf.nn.relu(fc) 14 | else: 15 | output = lrelu(fc) 16 | else: 17 | output = fc 18 | 19 | return output 20 | 21 | def normalize(inputs, 22 | type="bn", 23 | decay=.99, 24 | is_training=True, 25 | activation_fn=None, 26 | scope="normalize"): 27 | '''Applies {batch|layer} normalization. 28 | 29 | Args: 30 | inputs: A tensor with 2 or more dimensions, where the first dimension has 31 | `batch_size`. If type is `bn`, the normalization is over all but 32 | the last dimension. Or if type is `ln`, the normalization is over 33 | the last dimension. Note that this is different from the native 34 | `tf.contrib.layers.batch_norm`. For this I recommend you change 35 | a line in ``tensorflow/contrib/layers/python/layers/layer.py` 36 | as follows. 37 | Before: mean, variance = nn.moments(inputs, axis, keep_dims=True) 38 | After: mean, variance = nn.moments(inputs, [-1], keep_dims=True) 39 | type: A string. Either "bn" or "ln". 40 | decay: Decay for the moving average. Reasonable values for `decay` are close 41 | to 1.0, typically in the multiple-nines range: 0.999, 0.99, 0.9, etc. 42 | Lower `decay` value (recommend trying `decay`=0.9) if model experiences 43 | reasonably good training performance but poor validation and/or test 44 | performance. 45 | is_training: Whether or not the layer is in training mode. W 46 | activation_fn: Activation function. 47 | scope: Optional scope for `variable_scope`. 48 | 49 | Returns: 50 | A tensor with the same shape and data dtype as `inputs`. 51 | ''' 52 | if type == "bn": 53 | inputs_shape = inputs.get_shape() 54 | inputs_rank = inputs_shape.ndims 55 | 56 | # use fused batch norm if inputs_rank in [2, 3, 4] as it is much faster. 57 | # pay attention to the fact that fused_batch_norm requires shape to be rank 4 of NHWC. 58 | if inputs_rank in [2, 3, 4]: 59 | if inputs_rank == 2: 60 | inputs = tf.expand_dims(inputs, axis=1) 61 | inputs = tf.expand_dims(inputs, axis=2) 62 | elif inputs_rank == 3: 63 | inputs = tf.expand_dims(inputs, axis=1) 64 | 65 | outputs = tf.contrib.layers.batch_norm(inputs=inputs, 66 | decay=decay, 67 | center=True, 68 | scale=True, 69 | activation_fn=activation_fn, 70 | updates_collections=None, 71 | is_training=is_training, 72 | scope=scope, 73 | zero_debias_moving_mean=True, 74 | fused=True) 75 | # restore original shape 76 | if inputs_rank == 2: 77 | outputs = tf.squeeze(outputs, axis=[1, 2]) 78 | elif inputs_rank == 3: 79 | outputs = tf.squeeze(outputs, axis=1) 80 | else: # fallback to naive batch norm 81 | outputs = tf.contrib.layers.batch_norm(inputs=inputs, 82 | decay=decay, 83 | center=True, 84 | scale=True, 85 | activation_fn=activation_fn, 86 | updates_collections=None, 87 | is_training=is_training, 88 | scope=scope, 89 | fused=False) 90 | elif type == "ln": 91 | outputs = tf.contrib.layers.layer_norm(inputs=inputs, 92 | center=True, 93 | scale=True, 94 | activation_fn=activation_fn, 95 | scope=scope) 96 | elif type == "in": 97 | with tf.variable_scope(scope): 98 | batch, steps, channels = inputs.get_shape().as_list() 99 | var_shape = [channels] 100 | mu, sigma_sq = tf.nn.moments(inputs, [1], keep_dims=True) 101 | shift = tf.Variable(tf.zeros(var_shape)) 102 | scale = tf.Variable(tf.ones(var_shape)) 103 | epsilon = 1e-8 104 | normalized = (inputs - mu) / (sigma_sq + epsilon) ** (.5) 105 | outputs = scale * normalized + shift 106 | if activation_fn: 107 | outputs = activation_fn(outputs) 108 | else: 109 | raise ValueError("Currently we support `bn` or `ln` only.") 110 | 111 | return outputs 112 | 113 | 114 | 115 | def gaussian_noise_layer(input_layer, std): 116 | noise = tf.random_normal(shape=tf.shape(input_layer), mean=0.0, stddev=std, dtype=tf.float32) 117 | return input_layer + noise 118 | 119 | def lrelu(x, leak=0.2, name="lrelu"): 120 | with tf.variable_scope(name): 121 | f1 = 0.5 * (1 + leak) 122 | f2 = 0.5 * (1 - leak) 123 | return f1 * x + f2 * abs(x) -------------------------------------------------------------------------------- /seq2seq_c/python/ops/basic_decoder.py: -------------------------------------------------------------------------------- 1 | # Copyright 2016 The TensorFlow Authors. All Rights Reserved. 2 | # 3 | # Licensed under the Apache License, Version 2.0 (the "License"); 4 | # you may not use this file except in compliance with the License. 5 | # You may obtain a copy of the License at 6 | # 7 | # http://www.apache.org/licenses/LICENSE-2.0 8 | # 9 | # Unless required by applicable law or agreed to in writing, software 10 | # distributed under the License is distributed on an "AS IS" BASIS, 11 | # WITHOUT WARRANTIES OR CONDITIONS OF ANY KIND, either express or implied. 12 | # See the License for the specific language governing permissions and 13 | # limitations under the License. 14 | # ============================================================================== 15 | """A class of Decoders that may sample to generate the next input. 16 | """ 17 | 18 | from __future__ import absolute_import 19 | from __future__ import division 20 | from __future__ import print_function 21 | 22 | import collections 23 | import tensorflow as tf 24 | from seq2seq_c.python.ops import decoder 25 | from seq2seq_c.python.ops import helper as helper_py 26 | from tensorflow.python.framework import ops 27 | from tensorflow.python.framework import tensor_shape 28 | from tensorflow.python.layers import base as layers_base 29 | from tensorflow.python.ops import rnn_cell_impl 30 | from tensorflow.python.util import nest 31 | 32 | 33 | __all__ = [ 34 | "BasicDecoderOutput", 35 | "BasicDecoder", 36 | ] 37 | 38 | 39 | class BasicDecoderOutput( 40 | collections.namedtuple("BasicDecoderOutput", ("rnn_output", "sample_id"))): 41 | pass 42 | 43 | 44 | class BasicDecoder(decoder.Decoder): 45 | """Basic sampling decoder.""" 46 | 47 | def __init__(self, cell, helper, initial_state, lantent_tensor,output_layer=None): 48 | """Initialize BasicDecoder. 49 | 50 | Args: 51 | cell: An `RNNCell` instance. 52 | helper: A `Helper` instance. 53 | initial_state: A (possibly nested tuple of...) tensors and TensorArrays. 54 | The initial state of the RNNCell. 55 | output_layer: (Optional) An instance of `tf.layers.Layer`, i.e., 56 | `tf.layers.Dense`. Optional layer to apply to the RNN output prior 57 | to storing the result or sampling. 58 | 59 | Raises: 60 | TypeError: if `cell`, `helper` or `output_layer` have an incorrect type. 61 | """ 62 | rnn_cell_impl.assert_like_rnncell("cell", cell) 63 | if not isinstance(helper, helper_py.Helper): 64 | raise TypeError("helper must be a Helper, received: %s" % type(helper)) 65 | if (output_layer is not None 66 | and not isinstance(output_layer, layers_base.Layer)): 67 | raise TypeError( 68 | "output_layer must be a Layer, received: %s" % type(output_layer)) 69 | self._cell = cell 70 | self._helper = helper 71 | self._initial_state = initial_state 72 | self._output_layer = output_layer 73 | self._lantent_tensor=lantent_tensor 74 | 75 | @property 76 | def batch_size(self): 77 | return self._helper.batch_size 78 | 79 | def _rnn_output_size(self): 80 | size = self._cell.output_size 81 | if self._output_layer is None: 82 | return size 83 | else: 84 | # To use layer's compute_output_shape, we need to convert the 85 | # RNNCell's output_size entries into shapes with an unknown 86 | # batch size. We then pass this through the layer's 87 | # compute_output_shape and read off all but the first (batch) 88 | # dimensions to get the output size of the rnn with the layer 89 | # applied to the top. 90 | output_shape_with_unknown_batch = nest.map_structure( 91 | lambda s: tensor_shape.TensorShape([None]).concatenate(s), 92 | size) 93 | layer_output_shape = self._output_layer.compute_output_shape( 94 | output_shape_with_unknown_batch) 95 | return nest.map_structure(lambda s: s[1:], layer_output_shape) 96 | 97 | @property 98 | def output_size(self): 99 | # Return the cell output and the id 100 | return BasicDecoderOutput( 101 | rnn_output=self._rnn_output_size(), 102 | sample_id=self._helper.sample_ids_shape) 103 | 104 | @property 105 | def output_dtype(self): 106 | # Assume the dtype of the cell is the output_size structure 107 | # containing the input_state's first component's dtype. 108 | # Return that structure and the sample_ids_dtype from the helper. 109 | dtype = nest.flatten(self._initial_state)[0].dtype 110 | return BasicDecoderOutput( 111 | nest.map_structure(lambda _: dtype, self._rnn_output_size()), 112 | self._helper.sample_ids_dtype) 113 | 114 | def initialize(self, name=None): 115 | """Initialize the decoder. 116 | 117 | Args: 118 | name: Name scope for any created operations. 119 | 120 | Returns: 121 | `(finished, first_inputs, initial_state)`. 122 | """ 123 | return self._helper.initialize() + (self._initial_state,) 124 | 125 | def step(self, time, inputs, state, name=None): 126 | """Perform a decoding step. 127 | 128 | Args: 129 | time: scalar `int32` tensor. 130 | inputs: A (structure of) input tensors. 131 | state: A (structure of) state tensors and TensorArrays. 132 | name: Name scope for any created operations. 133 | 134 | Returns: 135 | `(outputs, next_state, next_inputs, finished)`. 136 | """ 137 | with ops.name_scope(name, "BasicDecoderStep", (time, inputs, state)): 138 | cell_outputs, cell_state = self._cell(inputs, state) 139 | #add tensor to each output 140 | cell_outputs = tf.concat([cell_outputs, self._lantent_tensor], 1) 141 | 142 | if self._output_layer is not None: 143 | cell_outputs = self._output_layer(cell_outputs) 144 | sample_ids = self._helper.sample( 145 | time=time, outputs=cell_outputs, state=cell_state) 146 | (finished, next_inputs, next_state) = self._helper.next_inputs( 147 | time=time, 148 | outputs=cell_outputs, 149 | state=cell_state, 150 | sample_ids=sample_ids) 151 | outputs = BasicDecoderOutput(cell_outputs, sample_ids) 152 | return (outputs, next_state, next_inputs, finished) 153 | -------------------------------------------------------------------------------- /Trip/data/trajid.dat: -------------------------------------------------------------------------------- 1 | 4 2 | 13 3 | 32 4 | 33 5 | 34 6 | 39 7 | 41 8 | 42 9 | 44 10 | 46 11 | 67 12 | 74 13 | 81 14 | 82 15 | 98 16 | 99 17 | 101 18 | 122 19 | 127 20 | 142 21 | 150 22 | 164 23 | 185 24 | 193 25 | 194 26 | 202 27 | 203 28 | 219 29 | 222 30 | 229 31 | 240 32 | 260 33 | 265 34 | 266 35 | 267 36 | 274 37 | 275 38 | 282 39 | 284 40 | 311 41 | 316 42 | 317 43 | 349 44 | 365 45 | 366 46 | 368 47 | 370 48 | 371 49 | 386 50 | 387 51 | 398 52 | 411 53 | 438 54 | 446 55 | 449 56 | 450 57 | 464 58 | 481 59 | 507 60 | 512 61 | 513 62 | 514 63 | 528 64 | 529 65 | 540 66 | 542 67 | 543 68 | 581 69 | 644 70 | 652 71 | 672 72 | 680 73 | 682 74 | 690 75 | 691 76 | 694 77 | 695 78 | 701 79 | 702 80 | 705 81 | 734 82 | 739 83 | 758 84 | 800 85 | 802 86 | 807 87 | 808 88 | 826 89 | 837 90 | 838 91 | 867 92 | 881 93 | 882 94 | 907 95 | 913 96 | 924 97 | 931 98 | 937 99 | 944 100 | 945 101 | 948 102 | 956 103 | 969 104 | 970 105 | 981 106 | 986 107 | 1017 108 | 1022 109 | 1027 110 | 1029 111 | 1030 112 | 1042 113 | 1045 114 | 1046 115 | 1051 116 | 1055 117 | 1074 118 | 1075 119 | 1077 120 | 1080 121 | 1083 122 | 1087 123 | 1094 124 | 1095 125 | 1097 126 | 1103 127 | 1107 128 | 1110 129 | 1111 130 | 1120 131 | 1121 132 | 1122 133 | 1128 134 | 1146 135 | 1158 136 | 1160 137 | 1173 138 | 1178 139 | 1179 140 | 1180 141 | 1181 142 | 1186 143 | 1193 144 | 1201 145 | 1203 146 | 1208 147 | 1216 148 | 1219 149 | 1220 150 | 1226 151 | 1262 152 | 1264 153 | 1270 154 | 1271 155 | 1273 156 | 1281 157 | 1282 158 | 1286 159 | 1287 160 | 1290 161 | 1294 162 | 1302 163 | 1308 164 | 1317 165 | 1320 166 | 1323 167 | 1338 168 | 1339 169 | 1345 170 | 1349 171 | 1351 172 | 1367 173 | 1392 174 | 1394 175 | 1423 176 | 1427 177 | 1428 178 | 1432 179 | 1452 180 | 1471 181 | 1473 182 | 1483 183 | 1487 184 | 1490 185 | 1495 186 | 1507 187 | 1508 188 | 1522 189 | 1526 190 | 1529 191 | 1533 192 | 1535 193 | 1536 194 | 1537 195 | 1538 196 | 1542 197 | 1555 198 | 1560 199 | 1567 200 | 1568 201 | 1575 202 | 1577 203 | 1607 204 | 1626 205 | 1627 206 | 1639 207 | 1651 208 | 1652 209 | 1656 210 | 1669 211 | 1680 212 | 1683 213 | 1692 214 | 1700 215 | 1703 216 | 1714 217 | 1717 218 | 1747 219 | 1761 220 | 1765 221 | 1774 222 | 1799 223 | 1811 224 | 1821 225 | 1839 226 | 1850 227 | 1862 228 | 1874 229 | 1875 230 | 1876 231 | 1884 232 | 1953 233 | 1965 234 | 1966 235 | 1970 236 | 1989 237 | 1996 238 | 2003 239 | 2007 240 | 2052 241 | 2054 242 | 2055 243 | 2060 244 | 2065 245 | 2072 246 | 2078 247 | 2091 248 | 2092 249 | 2093 250 | 2095 251 | 2108 252 | 2109 253 | 2137 254 | 2138 255 | 2144 256 | 2151 257 | 2175 258 | 2185 259 | 2209 260 | 2217 261 | 2230 262 | 2231 263 | 2233 264 | 2237 265 | 2246 266 | 2249 267 | 2254 268 | 2261 269 | 2262 270 | 2266 271 | 2267 272 | 2268 273 | 2270 274 | 2271 275 | 2272 276 | 2273 277 | 2280 278 | 2286 279 | 2307 280 | 2312 281 | 2318 282 | 2319 283 | 2320 284 | 2333 285 | 2334 286 | 2335 287 | 2339 288 | 2348 289 | 2370 290 | 2371 291 | 2379 292 | 2380 293 | 2388 294 | 2401 295 | 2402 296 | 2404 297 | 2405 298 | 2421 299 | 2427 300 | 2428 301 | 2429 302 | 2436 303 | 2461 304 | 2463 305 | 2491 306 | 2492 307 | 2493 308 | 2495 309 | 2496 310 | 2498 311 | 2503 312 | 2508 313 | 2514 314 | 2515 315 | 2516 316 | 2528 317 | 2572 318 | 2578 319 | 2583 320 | 2587 321 | 2600 322 | 2602 323 | 2605 324 | 2608 325 | 2617 326 | 2618 327 | 2625 328 | 2633 329 | 2639 330 | 2645 331 | 2647 332 | 2648 333 | 2651 334 | 2673 335 | 2674 336 | 2683 337 | 2684 338 | 2688 339 | 2718 340 | 2720 341 | 2721 342 | 2722 343 | 2723 344 | 2727 345 | 2749 346 | 2752 347 | 2757 348 | 2769 349 | 2791 350 | 2794 351 | 2795 352 | 2820 353 | 2840 354 | 2845 355 | 2850 356 | 2853 357 | 2879 358 | 2891 359 | 2893 360 | 2896 361 | 2907 362 | 2924 363 | 2942 364 | 2947 365 | 2949 366 | 2950 367 | 2953 368 | 2962 369 | 2963 370 | 2967 371 | 2968 372 | 2969 373 | 2972 374 | 2978 375 | 2979 376 | 2980 377 | 2981 378 | 2983 379 | 2998 380 | 3025 381 | 3032 382 | 3038 383 | 3046 384 | 3047 385 | 3049 386 | 3056 387 | 3057 388 | 3059 389 | 3064 390 | 3070 391 | 3076 392 | 3080 393 | 3085 394 | 3096 395 | 3099 396 | 3113 397 | 3115 398 | 3122 399 | 3135 400 | 3136 401 | 3138 402 | 3140 403 | 3145 404 | 3146 405 | 3147 406 | 3154 407 | 3155 408 | 3180 409 | 3181 410 | 3183 411 | 3199 412 | 3214 413 | 3216 414 | 3225 415 | 3233 416 | 3245 417 | 3257 418 | 3282 419 | 3284 420 | 3285 421 | 3288 422 | 3301 423 | 3302 424 | 3303 425 | 3308 426 | 3323 427 | 3324 428 | 3329 429 | 3332 430 | 3335 431 | 3340 432 | 3344 433 | 3364 434 | 3371 435 | 3372 436 | 3381 437 | 3388 438 | 3389 439 | 3394 440 | 3399 441 | 3411 442 | 3413 443 | 3414 444 | 3416 445 | 3417 446 | 3423 447 | 3424 448 | 3425 449 | 3466 450 | 3505 451 | 3530 452 | 3565 453 | 3572 454 | 3586 455 | 3597 456 | 3601 457 | 3604 458 | 3609 459 | 3611 460 | 3616 461 | 3623 462 | 3625 463 | 3638 464 | 3644 465 | 3650 466 | 3651 467 | 3654 468 | 3656 469 | 3657 470 | 3665 471 | 3667 472 | 3677 473 | 3681 474 | 3684 475 | 3693 476 | 3696 477 | 3699 478 | 3700 479 | 3701 480 | 3707 481 | 3709 482 | 3711 483 | 3719 484 | 3734 485 | 3748 486 | 3758 487 | 3776 488 | 3789 489 | 3796 490 | 3807 491 | 3810 492 | 3811 493 | 3814 494 | 3825 495 | 3839 496 | 3840 497 | 3842 498 | 3862 499 | 3882 500 | 3889 501 | 3895 502 | 3896 503 | 3920 504 | 3928 505 | 3930 506 | 3948 507 | 3966 508 | 3968 509 | 3985 510 | 3991 511 | 4010 512 | 4011 513 | 4012 514 | 4020 515 | 4031 516 | 4043 517 | 4044 518 | 4046 519 | 4051 520 | 4053 521 | 4065 522 | 4066 523 | 4071 524 | 4080 525 | 4083 526 | 4087 527 | 4095 528 | 4108 529 | 4117 530 | 4120 531 | 4155 532 | 4157 533 | 4160 534 | 4161 535 | 4162 536 | 4164 537 | 4165 538 | 4168 539 | 4169 540 | 4173 541 | 4174 542 | 4175 543 | 4176 544 | 4180 545 | 4181 546 | 4196 547 | 4197 548 | 4210 549 | 4212 550 | 4213 551 | 4219 552 | 4220 553 | 4221 554 | 4227 555 | 4229 556 | 4232 557 | 4235 558 | 4236 559 | 4239 560 | 4242 561 | 4277 562 | 4287 563 | 4292 564 | 4293 565 | 4295 566 | 4297 567 | 4316 568 | 4318 569 | 4335 570 | 4346 571 | 4352 572 | 4353 573 | 4355 574 | 4356 575 | 4357 576 | 4358 577 | 4362 578 | 4369 579 | 4421 580 | 4455 581 | 4456 582 | 4457 583 | 4468 584 | 4469 585 | 4470 586 | 4471 587 | 4473 588 | 4474 589 | 4475 590 | 4490 591 | 4491 592 | 4493 593 | 4496 594 | 4557 595 | 4568 596 | 4574 597 | 4575 598 | 4591 599 | 4592 600 | 4593 601 | 4702 602 | 4718 603 | 4719 604 | 4721 605 | 4722 606 | 4727 607 | 4728 608 | 4744 609 | 4745 610 | 4746 611 | 4748 612 | 4750 613 | 4799 614 | 4804 615 | 4805 616 | 4812 617 | 4817 618 | 4818 619 | 4823 620 | 4858 621 | 4894 622 | 4922 623 | 4923 624 | 4937 625 | 4941 626 | 4951 627 | 4952 628 | 4955 629 | 4969 630 | 4989 631 | 5001 632 | 5016 633 | 5017 634 | 5027 635 | -------------------------------------------------------------------------------- /Trip/AMSGrad.py: -------------------------------------------------------------------------------- 1 | """AMSGrad for TensorFlow.""" 2 | 3 | from tensorflow.python.eager import context 4 | from tensorflow.python.framework import ops 5 | from tensorflow.python.ops import control_flow_ops 6 | from tensorflow.python.ops import math_ops 7 | from tensorflow.python.ops import resource_variable_ops 8 | from tensorflow.python.ops import state_ops 9 | from tensorflow.python.ops import variable_scope 10 | from tensorflow.python.training import optimizer 11 | 12 | 13 | class AMSGrad(optimizer.Optimizer): 14 | def __init__(self, learning_rate=0.01, beta1=0.9, beta2=0.99, epsilon=1e-8, use_locking=False, name="AMSGrad"): 15 | super(AMSGrad, self).__init__(use_locking, name) 16 | self._lr = learning_rate 17 | self._beta1 = beta1 18 | self._beta2 = beta2 19 | self._epsilon = epsilon 20 | 21 | self._lr_t = None 22 | self._beta1_t = None 23 | self._beta2_t = None 24 | self._epsilon_t = None 25 | 26 | self._beta1_power = None 27 | self._beta2_power = None 28 | 29 | def _create_slots(self, var_list): 30 | first_var = min(var_list, key=lambda x: x.name) 31 | 32 | create_new = self._beta1_power is None 33 | if not create_new and context.in_graph_mode(): 34 | create_new = (self._beta1_power.graph is not first_var.graph) 35 | 36 | if create_new: 37 | with ops.colocate_with(first_var): 38 | self._beta1_power = variable_scope.variable(self._beta1, name="beta1_power", trainable=False) 39 | self._beta2_power = variable_scope.variable(self._beta2, name="beta2_power", trainable=False) 40 | # Create slots for the first and second moments. 41 | for v in var_list : 42 | self._zeros_slot(v, "m", self._name) 43 | self._zeros_slot(v, "v", self._name) 44 | self._zeros_slot(v, "vhat", self._name) 45 | 46 | def _prepare(self): 47 | self._lr_t = ops.convert_to_tensor(self._lr) 48 | self._beta1_t = ops.convert_to_tensor(self._beta1) 49 | self._beta2_t = ops.convert_to_tensor(self._beta2) 50 | self._epsilon_t = ops.convert_to_tensor(self._epsilon) 51 | 52 | def _apply_dense(self, grad, var): 53 | beta1_power = math_ops.cast(self._beta1_power, var.dtype.base_dtype) 54 | beta2_power = math_ops.cast(self._beta2_power, var.dtype.base_dtype) 55 | lr_t = math_ops.cast(self._lr_t, var.dtype.base_dtype) 56 | beta1_t = math_ops.cast(self._beta1_t, var.dtype.base_dtype) 57 | beta2_t = math_ops.cast(self._beta2_t, var.dtype.base_dtype) 58 | epsilon_t = math_ops.cast(self._epsilon_t, var.dtype.base_dtype) 59 | 60 | lr = (lr_t * math_ops.sqrt(1 - beta2_power) / (1 - beta1_power)) 61 | 62 | # m_t = beta1 * m + (1 - beta1) * g_t 63 | m = self.get_slot(var, "m") 64 | m_scaled_g_values = grad * (1 - beta1_t) 65 | m_t = state_ops.assign(m, beta1_t * m + m_scaled_g_values, use_locking=self._use_locking) 66 | 67 | # v_t = beta2 * v + (1 - beta2) * (g_t * g_t) 68 | v = self.get_slot(var, "v") 69 | v_scaled_g_values = (grad * grad) * (1 - beta2_t) 70 | v_t = state_ops.assign(v, beta2_t * v + v_scaled_g_values, use_locking=self._use_locking) 71 | 72 | # amsgrad 73 | vhat = self.get_slot(var, "vhat") 74 | vhat_t = state_ops.assign(vhat, math_ops.maximum(v_t, vhat)) 75 | v_sqrt = math_ops.sqrt(vhat_t) 76 | 77 | var_update = state_ops.assign_sub(var, lr * m_t / (v_sqrt + epsilon_t), use_locking=self._use_locking) 78 | return control_flow_ops.group(*[var_update, m_t, v_t, vhat_t]) 79 | 80 | def _resource_apply_dense(self, grad, var): 81 | var = var.handle 82 | beta1_power = math_ops.cast(self._beta1_power, grad.dtype.base_dtype) 83 | beta2_power = math_ops.cast(self._beta2_power, grad.dtype.base_dtype) 84 | lr_t = math_ops.cast(self._lr_t, grad.dtype.base_dtype) 85 | beta1_t = math_ops.cast(self._beta1_t, grad.dtype.base_dtype) 86 | beta2_t = math_ops.cast(self._beta2_t, grad.dtype.base_dtype) 87 | epsilon_t = math_ops.cast(self._epsilon_t, grad.dtype.base_dtype) 88 | 89 | lr = (lr_t * math_ops.sqrt(1 - beta2_power) / (1 - beta1_power)) 90 | 91 | # m_t = beta1 * m + (1 - beta1) * g_t 92 | m = self.get_slot(var, "m").handle 93 | m_scaled_g_values = grad * (1 - beta1_t) 94 | m_t = state_ops.assign(m, beta1_t * m + m_scaled_g_values, use_locking=self._use_locking) 95 | 96 | # v_t = beta2 * v + (1 - beta2) * (g_t * g_t) 97 | v = self.get_slot(var, "v").handle 98 | v_scaled_g_values = (grad * grad) * (1 - beta2_t) 99 | v_t = state_ops.assign(v, beta2_t * v + v_scaled_g_values, use_locking=self._use_locking) 100 | 101 | # amsgrad 102 | vhat = self.get_slot(var, "vhat").handle 103 | vhat_t = state_ops.assign(vhat, math_ops.maximum(v_t, vhat)) 104 | v_sqrt = math_ops.sqrt(vhat_t) 105 | 106 | var_update = state_ops.assign_sub(var, lr * m_t / (v_sqrt + epsilon_t), use_locking=self._use_locking) 107 | return control_flow_ops.group(*[var_update, m_t, v_t, vhat_t]) 108 | 109 | def _apply_sparse_shared(self, grad, var, indices, scatter_add): 110 | beta1_power = math_ops.cast(self._beta1_power, var.dtype.base_dtype) 111 | beta2_power = math_ops.cast(self._beta2_power, var.dtype.base_dtype) 112 | lr_t = math_ops.cast(self._lr_t, var.dtype.base_dtype) 113 | beta1_t = math_ops.cast(self._beta1_t, var.dtype.base_dtype) 114 | beta2_t = math_ops.cast(self._beta2_t, var.dtype.base_dtype) 115 | epsilon_t = math_ops.cast(self._epsilon_t, var.dtype.base_dtype) 116 | 117 | lr = (lr_t * math_ops.sqrt(1 - beta2_power) / (1 - beta1_power)) 118 | 119 | # m_t = beta1 * m + (1 - beta1) * g_t 120 | m = self.get_slot(var, "m") 121 | m_scaled_g_values = grad * (1 - beta1_t) 122 | m_t = state_ops.assign(m, m * beta1_t, use_locking=self._use_locking) 123 | with ops.control_dependencies([m_t]): 124 | m_t = scatter_add(m, indices, m_scaled_g_values) 125 | 126 | # v_t = beta2 * v + (1 - beta2) * (g_t * g_t) 127 | v = self.get_slot(var, "v") 128 | v_scaled_g_values = (grad * grad) * (1 - beta2_t) 129 | v_t = state_ops.assign(v, v * beta2_t, use_locking=self._use_locking) 130 | with ops.control_dependencies([v_t]): 131 | v_t = scatter_add(v, indices, v_scaled_g_values) 132 | 133 | # amsgrad 134 | vhat = self.get_slot(var, "vhat") 135 | vhat_t = state_ops.assign(vhat, math_ops.maximum(v_t, vhat)) 136 | v_sqrt = math_ops.sqrt(vhat_t) 137 | var_update = state_ops.assign_sub(var, lr * m_t / (v_sqrt + epsilon_t), use_locking=self._use_locking) 138 | return control_flow_ops.group(*[var_update, m_t, v_t, vhat_t]) 139 | 140 | def _apply_sparse(self, grad, var): 141 | return self._apply_sparse_shared( 142 | grad.values, var, grad.indices, 143 | lambda x, i, v: state_ops.scatter_add( # pylint: disable=g-long-lambda 144 | x, i, v, use_locking=self._use_locking)) 145 | 146 | def _resource_scatter_add(self, x, i, v): 147 | with ops.control_dependencies( 148 | [resource_variable_ops.resource_scatter_add(x.handle, i, v)]): 149 | return x.value() 150 | 151 | def _resource_apply_sparse(self, grad, var, indices): 152 | return self._apply_sparse_shared( 153 | grad, var, indices, self._resource_scatter_add) 154 | 155 | def _finish(self, update_ops, name_scope): 156 | # Update the power accumulators. 157 | with ops.control_dependencies(update_ops): 158 | with ops.colocate_with(self._beta1_power): 159 | update_beta1 = self._beta1_power.assign( 160 | self._beta1_power * self._beta1_t, 161 | use_locking=self._use_locking) 162 | update_beta2 = self._beta2_power.assign( 163 | self._beta2_power * self._beta2_t, 164 | use_locking=self._use_locking) 165 | return control_flow_ops.group(*update_ops + [update_beta1, update_beta2], 166 | name=name_scope) 167 | -------------------------------------------------------------------------------- /Trip/data/TKY_split200_set.dat: -------------------------------------------------------------------------------- 1 | 187-21480 1 0 16 2 | 187-21486 1 0 16 3 | 151-17418 13 6 4 4 | 151-17411 6 1 1 6 5 | 162-18806 21 17 17 21 6 | 192-22008 15 24 11 7 | 192-22004 5 12 3 8 | 55-6316 0 8 0 9 | 135-15393 3 9 9 3 10 | 117-13773 7 1 1 11 | 29-3680 2 17 2 12 | 117-13779 10 17 17 13 | 8-788 4 1 12 14 | 118-13805 22 2 2 1 15 | 118-13806 1 4 19 0 8 23 22 16 | 65-7834 3 0 5 17 | 163-18882 0 22 16 18 | 134-15360 11 1 0 19 | 113-13048 9 0 0 20 | 117-13783 1 1 10 21 | 51-5874 19 0 13 22 | 36-4372 3 3 13 23 | 159-18400 4 1 3 24 | 43-4908 15 10 1 25 | 63-7408 0 8 8 4 26 | 14-1494 1 1 25 27 | 196-22454 6 1 19 28 | 30-3739 13 0 0 29 | 195-22360 1 2 14 30 | 196-22452 3 19 4 6 31 | 12-1275 1 1 11 1 32 | 113-12997 0 2 9 33 | 12-1277 25 1 11 34 | 30-3731 13 3 20 20 13 35 | 12-1270 6 1 11 36 | 12-1273 1 7 11 37 | 113-12991 21 7 4 0 38 | 18-2243 1 11 8 39 | 68-8190 0 21 4 1 40 | 29-3545 3 3 0 41 | 35-4137 0 19 0 42 | 26-3077 3 3 2 43 | 102-11773 15 0 0 1 11 15 44 | 26-3075 3 2 3 45 | 26-3072 3 26 25 26 0 0 46 | 130-14983 22 0 13 47 | 196-22535 0 1 4 48 | 196-22532 19 1 6 49 | 161-18553 0 0 10 50 | 30-3748 13 3 0 51 | 30-3747 2 21 4 52 | 30-3741 13 0 0 53 | 83-9524 3 13 13 3 54 | 196-22433 6 1 19 55 | 187-21600 3 11 3 56 | 35-4142 0 2 14 57 | 21-2519 1 11 1 58 | 13-1352 8 20 20 59 | 43-4884 24 10 1 60 | 21-2620 1 1 13 61 | 124-14432 6 6 11 62 | 196-22508 1 1 21 63 | 147-17030 20 1 1 13 64 | 113-13009 9 10 0 65 | 45-5035 1 11 25 66 | 113-13004 0 2 9 67 | 126-14577 10 1 11 68 | 151-17465 0 19 1 69 | 151-17463 6 1 6 70 | 151-17462 13 6 1 11 1 6 71 | 36-4335 2 0 0 72 | 146-16892 6 1 1 12 73 | 13-1420 8 20 20 74 | 124-14404 3 3 13 75 | 150-17361 15 1 11 19 76 | 118-13836 4 1 1 0 77 | 118-13830 0 20 0 1 1 1 4 78 | 68-8146 2 3 3 79 | 65-7867 5 12 3 80 | 18-2048 1 11 8 81 | 18-1965 4 0 8 82 | 119-13923 27 27 27 83 | 113-12975 0 10 9 84 | 96-11178 12 0 2 85 | 113-12979 9 0 0 86 | 76-8967 4 7 7 4 87 | 76-8963 1 1 4 88 | 146-16808 1 11 7 89 | 65-7783 3 3 5 90 | 65-7788 4 4 5 91 | 86-9746 9 5 5 92 | 86-9748 9 21 21 93 | 196-22403 6 1 19 94 | 196-22407 6 1 19 95 | 198-22859 17 2 13 96 | 196-22656 19 1 6 97 | 26-3042 3 2 2 98 | 196-22654 19 1 6 99 | 63-7541 25 1 11 100 | 1-14 1 11 11 1 6 101 | 1-17 6 1 1 102 | 1-13 26 4 0 103 | 141-16263 14 8 8 104 | 96-11088 2 0 0 2 105 | 196-22663 6 1 19 106 | 49-5556 14 27 27 107 | 196-22544 19 1 6 108 | 196-22540 19 1 6 109 | 196-22548 11 1 6 110 | 12-1313 6 1 0 111 | 12-1317 1 4 24 28 5 112 | 12-1315 1 24 1 113 | 26-3038 3 3 2 114 | 187-21492 1 19 0 27 115 | 187-21496 6 0 16 116 | 102-11766 0 21 15 117 | 47-5121 9 28 28 118 | 187-21556 1 0 16 119 | 187-21653 1 0 16 120 | 76-9006 1 1 2 121 | 76-9008 4 0 1 122 | 8-772 12 21 4 123 | 187-21446 1 0 16 124 | 55-6365 0 19 19 4 125 | 117-13764 10 1 1 126 | 79-9121 0 1 3 127 | 117-13760 7 1 1 10 128 | 36-4391 4 1 6 129 | 105-12125 0 19 1 130 | 105-12123 10 2 0 0 131 | 105-12120 0 1 11 1 0 132 | 65-7820 3 0 2 2 5 133 | 98-11333 0 2 14 134 | 29-3603 2 3 3 135 | 116-13686 21 0 0 136 | 187-21449 1 19 0 137 | 164-19004 3 3 4 138 | 147-16959 23 23 0 139 | 1-5 1 1 6 140 | 5-457 1 1 2 141 | 135-15520 3 7 7 9 142 | 168-19371 8 8 0 143 | 65-7896 3 0 2 10 144 | 196-22442 6 1 19 145 | 124-14403 3 3 13 146 | 12-1266 6 1 7 2 1 147 | 65-7951 5 10 3 148 | 65-7952 3 1 5 149 | 12-1280 4 1 11 6 150 | 63-7500 1 4 8 8 0 21 151 | 99-11541 8 4 6 152 | 21-2593 1 11 1 153 | 102-11764 11 1 15 154 | 33-3980 5 1 5 155 | 196-22506 1 1 6 156 | 196-22509 19 1 4 157 | 58-6792 7 28 5 158 | 49-5519 12 27 27 159 | 161-18546 2 0 18 160 | 161-18540 10 2 27 161 | 135-15459 3 9 7 3 162 | 26-2994 3 6 25 6 163 | 96-11031 2 0 0 164 | 30-3754 13 3 20 165 | 38-4513 10 4 1 166 | 55-6339 2 2 10 167 | 38-4519 10 4 1 168 | 113-13124 9 10 0 169 | 196-22595 1 6 6 170 | 196-22597 19 1 6 171 | 43-4894 1 3 1 172 | 196-22592 19 4 1 6 173 | 100-11596 19 1 1 174 | 164-18991 19 3 3 175 | 187-21468 1 0 16 176 | 47-5173 0 9 28 177 | 187-21462 1 0 16 178 | 187-21461 16 0 1 19 179 | 163-18855 12 12 5 180 | 187-21501 1 0 16 181 | 45-5026 3 11 1 182 | 187-21509 1 0 16 183 | 151-17473 1 19 0 2 184 | 113-13010 0 4 19 0 185 | 187-21667 1 0 16 186 | 196-22424 1 1 6 187 | 31-3794 21 4 6 188 | 31-3791 7 4 6 189 | 13-1434 0 0 8 190 | 13-1436 0 20 20 191 | 146-16881 1 11 1 192 | 146-16884 2 0 19 1 193 | 117-13717 10 1 1 194 | 78-9073 0 0 1 195 | 117-13714 5 1 1 196 | 13-1439 4 19 1 197 | 126-14605 1 1 6 198 | 196-22486 6 1 19 199 | 126-14602 0 0 13 200 | 196-22485 6 1 1 4 6 201 | 124-14441 13 6 11 4 4 202 | 68-8159 26 4 0 203 | 68-8157 9 1 1 9 26 204 | 59-6857 20 8 22 205 | 68-8155 9 1 11 206 | 68-8153 0 19 1 1 6 207 | 68-8150 9 4 1 1 208 | 113-13066 9 0 0 209 | 30-3733 10 4 13 210 | 29-3672 3 2 1 3 211 | 113-12994 9 10 0 212 | 29-3677 3 21 21 3 213 | 43-4920 15 1 11 214 | 96-11160 2 20 2 0 215 | 128-14716 11 0 1 216 | 5-462 10 29 29 28 217 | 96-11168 2 0 0 2 218 | 12-1272 6 1 11 7 1 11 219 | 196-22467 11 1 6 220 | 30-3716 24 1 6 221 | 33-4012 5 23 1 222 | 196-22435 6 1 19 223 | 86-9771 4 8 23 224 | 18-2222 1 11 8 225 | 14-1567 25 1 11 226 | 63-7470 2 2 21 227 | 196-22558 1 6 13 228 | 194-22311 26 6 1 229 | 194-22313 1 9 9 230 | 199-22883 13 3 1 11 0 231 | 173-20014 0 22 22 0 232 | 98-11405 20 0 0 233 | 187-21540 1 0 16 234 | 12-1295 29 29 7 1 1 6 6 235 | 187-21542 1 0 16 236 | 113-13111 0 2 2 237 | 18-2127 1 11 8 238 | 80-9239 4 3 20 239 | 192-22025 3 2 3 240 | 105-12113 0 3 6 6 241 | 65-7694 7 5 5 242 | 105-12114 6 3 0 243 | 163-18866 16 0 0 16 244 | 187-21531 1 0 16 245 | 65-7814 5 21 2 5 246 | 187-21537 1 0 16 247 | 36-4312 17 17 18 2 248 | 164-19014 3 2 0 249 | 161-18703 10 2 18 250 | 150-17281 15 7 15 251 | 78-9083 2 1 0 252 | 78-9080 2 2 1 253 | 65-7751 4 12 5 254 | 75-8898 3 0 14 255 | 21-2449 0 3 13 256 | 13-1373 8 0 2 14 257 | 44-4923 2 1 1 258 | 196-22477 6 1 19 259 | 196-22476 6 1 19 260 | 168-19362 11 1 18 261 | 68-8160 1 1 9 262 | 65-7961 3 19 4 5 263 | 65-7964 5 10 3 264 | 13-1382 8 0 0 8 265 | 18-2062 20 3 3 0 266 | 117-13695 7 7 7 267 | 196-22394 6 19 1 268 | 96-11052 2 0 0 2 269 | 147-16945 0 0 0 270 | 63-7514 11 1 4 271 | 102-11756 24 4 1 11 272 | 141-16237 14 19 1 273 | 49-5527 4 22 22 22 274 | 49-5526 1 11 1 0 275 | 21-2619 0 1 1 276 | 196-22512 19 1 6 277 | 21-2614 1 1 9 9 1 11 278 | 26-3083 4 4 18 279 | 187-21668 1 0 16 280 | 124-14502 13 3 1 1 281 | 96-11122 12 16 16 12 282 | 63-7560 2 0 8 0 283 | 194-22320 25 26 26 284 | 161-18522 10 2 27 285 | 187-21478 1 0 16 286 | 187-21475 1 0 16 287 | 187-21476 0 1 3 288 | 1-2 1 11 1 6 289 | 187-21470 11 1 19 0 290 | 1-7 6 1 11 11 1 6 291 | 187-21572 1 0 16 292 | 55-6394 17 7 10 293 | 151-17401 0 1 6 294 | 192-22010 5 4 1 11 1 295 | 18-2110 1 11 8 296 | 13-1409 20 23 23 8 297 | 58-6744 5 10 7 298 | 128-14744 19 0 0 19 299 | 65-7844 5 10 0 20 300 | 65-7840 3 1 5 301 | 196-22629 19 1 6 302 | 59-6840 6 1 0 303 | 113-13077 9 10 0 304 | 113-13074 9 2 0 305 | 113-13075 2 0 0 306 | 18-2184 1 11 8 307 | 151-17497 23 8 0 308 | 117-13794 7 6 26 309 | 117-13793 1 1 10 310 | 36-4340 28 28 13 311 | 18-2026 1 11 8 312 | 151-17498 0 1 6 313 | 96-11155 23 2 2 314 | 9-1057 6 1 7 9 9 10 315 | 159-18413 0 19 1 316 | 5-478 1 1 1 317 | 135-15421 3 9 9 318 | 13-1320 6 1 1 319 | 13-1323 0 0 20 320 | 146-16829 18 1 28 321 | 13-1328 8 0 0 8 322 | 196-22422 6 1 19 323 | 30-3729 0 20 3 13 324 | 196-22426 20 20 3 13 325 | 30-3726 0 3 13 326 | 113-12981 0 2 2 327 | 113-12982 9 0 0 328 | 63-7462 4 1 11 329 | 124-14399 9 7 4 1 330 | 21-2577 1 1 4 331 | 57-6566 4 0 1 332 | 100-11639 1 11 1 333 | 196-22473 1 6 25 334 | 196-22640 6 1 6 335 | 196-22528 19 1 6 336 | 63-7489 4 1 11 1 4 337 | 58-6757 5 7 1 338 | 35-4191 8 0 2 339 | 76-8948 1 0 18 340 | 191-21905 11 7 7 341 | 146-16910 12 4 1 342 | 96-11204 2 0 2 343 | 191-21908 22 8 0 11 344 | 124-14398 1 4 7 9 345 | 143-16516 19 1 11 346 | 151-17527 0 19 0 0 347 | 18-2096 1 11 8 348 | 196-22633 19 1 6 349 | 196-22634 6 19 1 350 | 196-22580 6 1 4 351 | 65-7916 4 10 5 352 | 116-13467 21 18 18 353 | 147-17041 21 21 0 354 | 100-11630 23 19 11 1 355 | 187-21528 1 0 16 356 | 187-21529 1 0 16 357 | 187-21522 1 0 16 358 | 10-1112 2 1 11 6 359 | 151-17450 0 0 0 360 | 151-17457 0 19 1 361 | 8-888 12 3 1 362 | 179-20532 5 28 28 5 363 | 18-1984 24 8 7 364 | 113-13039 9 0 0 365 | 65-7721 5 7 4 5 366 | 12-1288 6 1 19 367 | 78-9076 0 0 3 368 | 146-16860 1 11 1 4 12 369 | 78-9072 1 1 0 370 | 126-14628 4 19 1 11 371 | 126-14629 1 6 9 372 | 13-1361 8 20 0 373 | 196-22466 11 1 6 374 | 118-13843 20 0 1 375 | 68-8176 13 2 3 26 376 | 107-12395 20 10 5 5 10 377 | 68-8178 18 4 4 378 | 13-1444 8 0 0 379 | 35-4228 10 28 5 380 | 13-1446 10 2 14 2 381 | 8-839 12 4 1 4 382 | 102-11744 11 1 4 383 | 102-11745 4 8 23 384 | 76-8979 4 1 4 385 | 29-3576 2 17 2 386 | 49-5532 0 0 27 387 | 49-5536 11 1 0 0 388 | 146-16810 1 10 15 389 | 196-22418 6 1 19 390 | 102-11759 9 1 11 1 0 391 | 196-22411 6 1 19 392 | 196-22413 6 1 19 393 | 124-14510 1 4 9 394 | 198-22866 4 6 13 395 | 193-22193 3 3 2 2 396 | 63-7458 21 8 0 21 397 | 9-987 1 1 4 398 | 49-5547 14 2 27 27 399 | 196-22575 19 1 6 400 | 22-2708 25 3 20 401 | 49-5548 27 2 14 402 | 12-1303 1 1 11 403 | 29-3609 2 17 2 404 | -------------------------------------------------------------------------------- /Trip/data/Toro_set.dat: -------------------------------------------------------------------------------- 1 | 20741443@N00-745 9 0 2 2 | 20741443@N00-744 0 8 2 3 | 64834051@N00-4182 10 14 4 4 | 11191102@N07-155 10 4 1 5 | 11191102@N07-157 5 3 13 6 | 11090433@N05-139 15 1 3 2 7 | 86292040@N00-5377 11 14 4 3 2 8 | 9911655@N08-5959 0 8 2 9 | 23908938@N02-1121 5 17 4 10 | 69525052@N06-4345 0 1 2 11 | 91032493@N00-5556 6 4 1 8 2 12 | 39038071@N00-2431 9 1 7 13 | 20741443@N00-884 0 3 8 2 14 | 84987970@N00-5237 15 1 2 15 | 20741443@N00-883 0 3 2 16 | 20741443@N00-880 6 1 2 17 | 23119895@N00-1078 6 14 4 18 | 99127884@N00-5963 0 3 8 19 | 67694608@N00-4284 10 4 1 8 20 | 67194633@N00-4276 4 1 16 21 | 14391210@N00-379 9 1 2 22 | 26736723@N00-1335 10 14 4 23 | 43139087@N00-2703 6 12 2 24 | 60597745@N00-3932 10 1 8 25 | 9025385@N07-5539 6 4 12 2 26 | 32827327@N03-1786 9 5 27 12 7 2 27 | 75450299@N07-4580 0 1 8 28 | 66922282@N00-4271 1 3 8 29 | 32934461@N00-1792 14 4 0 2 30 | 51260381@N05-3556 6 7 2 31 | 50542255@N00-3469 6 4 8 32 | 84518681@N00-5073 10 14 4 33 | 57519914@N00-3767 6 14 4 1 8 34 | 84987970@N00-5337 15 0 1 3 8 35 | 84987970@N00-5256 0 1 3 36 | 65438265@N00-4198 6 4 1 2 37 | 84987970@N00-5228 15 0 3 38 | 34211328@N00-1931 0 1 3 39 | 82845843@N04-5022 9 5 13 40 | 23908938@N02-1118 6 4 8 2 41 | 23908938@N02-1117 10 1 2 42 | 25802393@N02-1281 10 6 14 4 3 2 43 | 98715075@N00-5932 0 1 3 44 | 99771506@N00-5981 19 6 5 17 0 3 16 2 45 | 20342758@N00-681 10 14 4 3 12 8 2 46 | 23269969@N00-1082 9 14 4 2 47 | 20741443@N00-898 14 4 2 48 | 15205252@N00-472 19 24 21 49 | 61377999@N03-3967 9 0 1 50 | 62373848@N00-4004 10 4 8 51 | 89613559@N07-5530 6 14 4 52 | 96684313@N00-5856 6 14 2 53 | 35468159247@N01-2196 9 4 0 12 2 54 | 44359705@N07-2896 10 4 0 2 55 | 70554294@N00-4408 4 1 8 16 2 56 | 20741443@N00-729 6 0 1 3 8 57 | 34211328@N00-1934 6 14 4 58 | 11349639@N00-169 10 4 2 59 | 63706803@N00-4132 1 3 16 60 | 81471618@N06-4972 0 3 8 61 | 47616704@N00-3052 10 9 6 4 1 12 2 62 | 20741443@N00-809 6 0 3 8 2 63 | 20741443@N00-807 6 1 2 64 | 81471618@N06-4973 5 23 22 65 | 98222757@N02-5913 6 3 12 8 66 | 24616128@N00-1194 14 3 16 67 | 20741443@N00-730 6 3 8 68 | 20741443@N00-731 0 1 3 69 | 44925192@N00-2916 10 6 4 0 2 70 | 94064020@N00-5709 9 5 0 8 71 | 20741443@N00-929 6 0 3 8 72 | 20741443@N00-928 0 13 16 73 | 84987970@N00-5328 0 1 3 74 | 9985167@N04-6011 0 1 8 75 | 84987970@N00-5324 15 1 3 8 76 | 22996605@N05-1066 6 14 4 2 77 | 37517876@N07-2340 0 1 12 8 78 | 20741443@N00-920 6 0 2 79 | 20741443@N00-923 6 0 2 80 | 97675182@N00-5890 19 24 21 81 | 42907325@N00-2686 10 4 8 82 | 28288718@N03-1424 3 8 2 83 | 84035351@N00-5029 0 3 2 84 | 28288718@N03-1423 0 8 2 85 | 33473816@N00-1835 14 1 2 86 | 28288718@N03-1421 0 13 2 87 | 91255327@N00-5575 15 9 0 1 88 | 84987970@N00-5215 15 0 1 89 | 16326397@N02-515 5 4 3 90 | 55929357@N00-3707 1 3 8 91 | 88276719@N00-5497 6 4 2 92 | 84987970@N00-5325 23 22 12 93 | 34128007@N04-1912 11 10 14 4 16 94 | 84987970@N00-5250 9 4 0 1 3 2 95 | 32944866@N04-1796 5 1 16 96 | 58919362@N00-3856 6 0 3 2 97 | 78433558@N00-4805 11 6 8 98 | 32781576@N04-1782 1 7 8 99 | 42536354@N00-2666 9 0 7 100 | 37517876@N07-2338 11 6 2 101 | 34211328@N00-1968 6 0 1 3 2 102 | 25561968@N00-1275 10 14 4 103 | 37517876@N07-2331 5 3 13 104 | 26240579@N05-1312 4 0 8 2 105 | 16048448@N00-510 6 4 0 1 3 8 16 2 106 | 26344918@N04-1321 11 6 1 8 107 | 37517876@N07-2334 11 9 6 0 108 | 62486370@N00-4005 19 24 21 109 | 25133522@N00-1259 14 17 0 12 110 | 20741443@N00-722 6 1 3 111 | 20741443@N00-813 14 3 8 2 112 | 13273343@N00-288 14 4 2 113 | 22111831@N05-1031 6 1 3 114 | 20741443@N00-814 6 3 8 2 115 | 84987970@N00-5254 9 6 0 1 3 8 2 116 | 20741443@N00-816 3 8 2 117 | 20741443@N00-778 6 8 2 118 | 18412989@N00-623 0 3 12 119 | 39460517@N03-2472 0 1 3 8 120 | 10627620@N06-100 0 3 2 121 | 18412989@N00-624 11 6 3 122 | 46769540@N00-3008 9 1 8 123 | 27895744@N00-1378 11 14 4 124 | 99127884@N00-5964 11 9 6 5 23 1 8 16 2 125 | 35107151@N04-2110 9 6 1 126 | 33547369@N00-1877 0 1 3 127 | 84987970@N00-5301 6 1 8 2 128 | 37804979@N00-2377 4 0 1 8 129 | 16926694@N06-544 6 14 2 130 | 11191102@N07-162 6 1 2 131 | 24854893@N00-1237 5 12 7 2 132 | 20741443@N00-790 0 3 8 133 | 13644648@N03-298 11 10 9 6 14 5 4 1 3 12 8 16 2 134 | 59525924@N05-3881 11 10 18 135 | 31114347@N05-1642 14 12 2 136 | 49877040@N03-3438 10 4 7 137 | 20741443@N00-843 4 3 2 138 | 87473264@N00-5475 17 4 12 139 | 10627620@N06-99 15 3 13 140 | 84711541@N00-5099 6 0 1 3 2 141 | 27008470@N00-1344 14 1 2 142 | 99436246@N00-5971 4 1 16 143 | 72778314@N00-4493 10 0 1 3 7 8 144 | 39587684@N00-2499 1 3 8 145 | 43919105@N02-2803 1 3 8 146 | 34961066@N00-2048 5 0 3 147 | 35034348378@N01-2072 11 14 4 0 1 8 2 148 | 41721440@N00-2592 0 3 8 149 | 35034347485@N01-2069 9 0 1 3 8 150 | 49894734@N00-3454 0 1 3 151 | 39587684@N00-2497 11 6 7 152 | 39587684@N00-2496 6 14 2 153 | 37663331@N00-2349 0 3 8 154 | 39357749@N00-2440 6 4 2 155 | 20741443@N00-716 11 6 0 156 | 98715075@N00-5927 11 10 2 157 | 13528098@N00-294 14 4 2 158 | 20741443@N00-867 6 14 2 159 | 20741443@N00-712 6 3 2 160 | 54058873@N00-3636 15 14 0 1 3 8 161 | 34608255@N08-2036 6 5 1 162 | 14878709@N00-453 4 3 8 163 | 34619038@N00-2039 9 1 8 164 | 71482738@N00-4450 14 5 4 7 2 165 | 48678427@N06-3149 9 0 1 8 16 166 | 8250978@N04-4997 6 14 1 8 2 167 | 26195569@N00-1310 9 5 13 168 | 27031977@N04-1345 1 3 8 169 | 35468159247@N01-2198 0 1 3 170 | 37996606796@N01-2401 14 4 12 171 | 50642338@N00-3481 5 13 2 172 | 30739892@N03-1620 0 1 8 173 | 33182025@N00-1802 6 22 0 3 174 | 24854893@N00-1225 4 21 22 175 | 34314322@N00-1993 4 1 8 16 2 176 | 35237095252@N01-2158 10 6 4 2 177 | 9985167@N04-6018 0 3 8 178 | 20741443@N00-783 6 0 1 3 2 179 | 47501960@N00-3040 0 3 13 180 | 20741443@N00-786 6 0 3 8 181 | 20741443@N00-788 5 0 1 3 8 182 | 60597745@N00-3935 10 17 4 183 | 39460517@N03-2471 0 1 8 184 | 77112637@N00-4701 0 1 3 185 | 41894165897@N01-2604 6 5 16 186 | 17373191@N00-549 11 14 0 1 8 187 | 11241023@N03-166 6 1 8 188 | 9449875@N03-5751 19 24 21 189 | 20741443@N00-979 6 1 3 16 190 | 31349854@N00-1653 1 3 7 191 | 92487715@N03-5640 6 4 1 192 | 24128704@N08-1171 11 14 12 193 | 61239510@N00-3960 14 12 2 194 | 84987970@N00-5278 9 0 3 8 195 | 20741443@N00-708 6 0 3 8 2 196 | 9025385@N07-5541 11 10 6 4 2 197 | 34211328@N00-1940 9 5 1 3 16 198 | 75450299@N07-4578 9 0 1 7 8 199 | 20741443@N00-870 15 0 8 200 | 69754957@N00-4351 11 19 15 10 6 5 4 3 2 201 | 96083563@N00-5842 6 5 4 202 | 84987970@N00-5183 15 0 1 7 203 | 69754957@N00-4354 15 0 16 2 204 | 34211328@N00-1949 14 17 4 205 | 20741443@N00-879 6 0 3 206 | 84987970@N00-5277 0 1 3 13 8 16 207 | 30624156@N00-1598 6 1 13 8 2 208 | 60142746@N00-3898 4 3 2 209 | 29619840@N00-1536 10 14 4 210 | 91008793@N00-5555 10 9 14 5 4 211 | 94828981@N08-5798 0 3 8 212 | 35764233@N00-2209 4 0 1 213 | 35764233@N00-2207 11 17 0 1 3 214 | 30624156@N00-1591 6 14 4 2 215 | 30624156@N00-1592 0 1 3 8 216 | 14391210@N00-380 1 3 8 2 217 | 48453097@N00-3133 0 7 8 218 | 58919362@N00-3851 1 12 2 219 | 13672753@N00-302 10 14 4 1 220 | 9985167@N04-6028 15 1 3 8 2 221 | 84987970@N00-5287 0 1 8 222 | 9449875@N03-5747 0 3 8 223 | 44124367252@N01-2838 10 9 4 224 | 84987970@N00-5282 5 1 12 16 225 | 39460517@N03-2466 0 1 3 8 226 | 28157992@N03-1404 11 23 22 227 | 67952341@N00-4308 10 14 4 228 | 34211328@N00-1969 6 8 16 229 | 39460517@N03-2461 19 24 21 230 | 10502709@N05-58 6 5 13 231 | 34211328@N00-1933 19 24 21 232 | 51035555243@N01-3493 0 1 3 2 233 | 13907834@N00-316 5 0 13 234 | 23987663@N00-1128 18 0 1 8 235 | 34211328@N00-1937 19 24 21 236 | 13907834@N00-315 10 14 4 237 | 34211328@N00-1938 5 24 21 13 238 | 93241698@N00-5649 4 1 3 239 | 93241698@N00-5648 1 3 8 240 | 61377999@N03-3968 10 4 1 241 | 61377999@N03-3969 6 5 7 242 | 36357702@N08-2252 10 6 14 4 1 8 2 243 | 51035681908@N01-3504 10 5 4 0 1 7 244 | 84987970@N00-5253 11 0 3 245 | 20741443@N00-963 3 8 2 246 | 20741443@N00-962 0 3 8 247 | 20741443@N00-965 6 0 2 248 | 84987970@N00-5276 10 9 6 0 8 2 249 | 20741443@N00-969 15 0 8 250 | 20741443@N00-775 0 1 2 251 | 84987970@N00-5207 19 10 21 252 | 72236905@N00-4474 14 5 13 253 | 9161595@N03-5625 9 4 0 1 254 | 22979982@N06-1063 3 7 16 255 | 78015320@N00-4782 9 14 4 1 8 256 | 20456447@N03-686 11 6 14 4 0 1 2 257 | 20456447@N03-687 10 6 14 4 258 | 20456447@N03-688 15 9 1 16 259 | 68803445@N00-4326 10 14 4 260 | 30855862@N07-1627 11 6 4 261 | 23987663@N00-1123 5 0 1 3 262 | 43334562@N00-2795 6 0 3 263 | 84987970@N00-5294 19 24 21 264 | 84800532@N02-5132 5 4 1 265 | 78015320@N00-4783 9 5 17 266 | 20075574@N00-675 9 1 7 267 | 20741443@N00-988 6 3 12 8 268 | 53874606@N00-3626 11 14 4 269 | 29352917@N00-1523 4 1 3 8 270 | 29352917@N00-1524 4 16 2 271 | 20741443@N00-980 6 12 8 272 | 20344209@N00-683 17 4 16 273 | 14878709@N00-454 14 4 0 1 3 13 8 274 | 10502709@N05-67 1 3 8 275 | 47554402@N00-3045 0 3 8 276 | 23987663@N00-1135 1 3 8 277 | 8070489@N02-4914 1 3 8 278 | 51035610516@N01-3501 10 14 4 0 2 279 | 44654079@N05-2908 9 1 8 280 | 24395354@N02-1179 9 5 26 281 | 25475928@N04-1270 0 3 8 2 282 | 51035729697@N01-3514 9 6 2 283 | 98185064@N00-5912 10 4 8 284 | 20741443@N00-911 0 3 16 285 | 43139087@N00-2705 6 0 1 3 8 286 | 25846522@N00-1283 11 1 3 8 287 | 84987970@N00-5214 15 6 1 3 16 288 | 29352917@N00-1525 9 3 7 289 | 20741443@N00-763 0 3 8 290 | 24854893@N00-1247 4 12 7 291 | 20741443@N00-852 11 6 2 292 | 20741443@N00-753 6 3 8 2 293 | 27168489@N00-1358 14 4 0 1 3 8 294 | 7776449@N06-4758 9 0 3 7 13 295 | 23339848@N00-1089 5 17 2 296 | 47501960@N00-3037 15 3 8 297 | 9911655@N08-5947 14 7 16 298 | 32044903@N04-1726 6 13 8 299 | 18948547@N00-634 1 3 8 300 | 33547369@N00-1875 1 3 8 301 | 33906853@N00-1894 11 1 16 302 | 7776449@N06-4756 5 1 16 303 | 36553196@N08-2275 9 4 1 2 304 | 61609579@N00-3975 0 1 3 8 305 | 48784629@N00-3155 10 14 4 306 | 71482738@N00-4451 9 0 1 8 307 | 93897327@N00-5707 6 5 4 12 308 | 84987970@N00-5263 15 6 0 1 3 8 2 309 | 48085290@N00-3090 6 4 2 310 | 94725783@N00-5759 0 1 3 7 311 | 73416633@N00-4506 9 0 3 312 | 70289230@N00-4374 3 13 16 313 | 16693950@N00-523 4 1 3 314 | 85801042@N00-5369 15 10 6 14 4 0 3 8 315 | 10502709@N05-71 0 1 3 8 316 | 16693950@N00-525 6 14 4 0 1 3 8 2 317 | 16693950@N00-524 23 22 7 318 | 20741443@N00-821 1 3 8 319 | 20741443@N00-822 6 3 8 2 320 | 27947906@N08-1385 17 4 12 321 | 20741443@N00-825 1 3 8 322 | 34211328@N00-1918 10 12 2 323 | 75396048@N00-4570 0 12 2 324 | 20741443@N00-905 6 5 0 1 3 8 2 325 | 20741443@N00-904 1 3 2 326 | 49503002894@N01-3401 15 14 1 327 | 71206023@N00-4420 14 4 12 328 | 49503002894@N01-3403 14 4 1 329 | 49503002894@N01-3402 14 22 1 330 | 18412989@N00-619 4 1 3 331 | 84987970@N00-5343 0 1 2 332 | 9416709@N08-5715 14 4 1 333 | 33473816@N00-1865 6 13 2 334 | 84987970@N00-5180 0 1 3 8 335 | 8110030@N05-4961 11 10 6 4 22 0 3 12 8 2 336 | -------------------------------------------------------------------------------- /seq2seq_c/python/ops/decoder.py: -------------------------------------------------------------------------------- 1 | # Copyright 2016 The TensorFlow Authors. All Rights Reserved. 2 | # 3 | # Licensed under the Apache License, Version 2.0 (the "License"); 4 | # you may not use this file except in compliance with the License. 5 | # You may obtain a copy of the License at 6 | # 7 | # http://www.apache.org/licenses/LICENSE-2.0 8 | # 9 | # Unless required by applicable law or agreed to in writing, software 10 | # distributed under the License is distributed on an "AS IS" BASIS, 11 | # WITHOUT WARRANTIES OR CONDITIONS OF ANY KIND, either express or implied. 12 | # See the License for the specific language governing permissions and 13 | # limitations under the License. 14 | # ============================================================================== 15 | """Seq2seq layer operations for use in neural networks.""" 16 | 17 | from __future__ import absolute_import 18 | from __future__ import division 19 | from __future__ import print_function 20 | 21 | import abc 22 | import six 23 | 24 | from tensorflow.python.framework import constant_op 25 | from tensorflow.python.framework import dtypes 26 | from tensorflow.python.framework import ops 27 | from tensorflow.python.framework import tensor_shape 28 | from tensorflow.python.framework import tensor_util 29 | from tensorflow.python.ops import array_ops 30 | from tensorflow.python.ops import control_flow_ops 31 | from tensorflow.python.ops import math_ops 32 | from tensorflow.python.ops import rnn 33 | from tensorflow.python.ops import rnn_cell_impl 34 | from tensorflow.python.ops import tensor_array_ops 35 | from tensorflow.python.ops import variable_scope 36 | from tensorflow.python.util import nest 37 | 38 | 39 | __all__ = ["Decoder", "dynamic_decode"] 40 | 41 | 42 | _transpose_batch_time = rnn._transpose_batch_time # pylint: disable=protected-access 43 | _zero_state_tensors = rnn_cell_impl._zero_state_tensors # pylint: disable=protected-access 44 | 45 | 46 | @six.add_metaclass(abc.ABCMeta) 47 | class Decoder(object): 48 | """An RNN Decoder abstract interface object. 49 | 50 | Concepts used by this interface: 51 | - `inputs`: (structure of) tensors and TensorArrays that is passed as input to 52 | the RNNCell composing the decoder, at each time step. 53 | - `state`: (structure of) tensors and TensorArrays that is passed to the 54 | RNNCell instance as the state. 55 | - `finished`: boolean tensor telling whether each sequence in the batch is 56 | finished. 57 | - `outputs`: Instance of BasicDecoderOutput. Result of the decoding, at each 58 | time step. 59 | """ 60 | 61 | @property 62 | def batch_size(self): 63 | """The batch size of input values.""" 64 | raise NotImplementedError 65 | 66 | @property 67 | def output_size(self): 68 | """A (possibly nested tuple of...) integer[s] or `TensorShape` object[s].""" 69 | raise NotImplementedError 70 | 71 | @property 72 | def output_dtype(self): 73 | """A (possibly nested tuple of...) dtype[s].""" 74 | raise NotImplementedError 75 | 76 | @abc.abstractmethod 77 | def initialize(self, name=None): 78 | """Called before any decoding iterations. 79 | 80 | This methods must compute initial input values and initial state. 81 | 82 | Args: 83 | name: Name scope for any created operations. 84 | 85 | Returns: 86 | `(finished, initial_inputs, initial_state)`: initial values of 87 | 'finished' flags, inputs and state. 88 | """ 89 | raise NotImplementedError 90 | 91 | @abc.abstractmethod 92 | def step(self, time, inputs, state, name=None): 93 | """Called per step of decoding (but only once for dynamic decoding). 94 | 95 | Args: 96 | time: Scalar `int32` tensor. Current step number. 97 | inputs: RNNCell input (possibly nested tuple of) tensor[s] for this time 98 | step. 99 | state: RNNCell state (possibly nested tuple of) tensor[s] from previous 100 | time step. 101 | name: Name scope for any created operations. 102 | 103 | Returns: 104 | `(outputs, next_state, next_inputs, finished)`: `outputs` is an object 105 | containing the decoder output, `next_state` is a (structure of) state 106 | tensors and TensorArrays, `next_inputs` is the tensor that should be used 107 | as input for the next step, `finished` is a boolean tensor telling whether 108 | the sequence is complete, for each sequence in the batch. 109 | """ 110 | raise NotImplementedError 111 | 112 | def finalize(self, outputs, final_state, sequence_lengths): 113 | raise NotImplementedError 114 | 115 | @property 116 | def tracks_own_finished(self): 117 | """Describes whether the Decoder keeps track of finished states. 118 | 119 | Most decoders will emit a true/false `finished` value independently 120 | at each time step. In this case, the `dynamic_decode` function keeps track 121 | of which batch entries are already finished, and performs a logical OR to 122 | insert new batches to the finished set. 123 | 124 | Some decoders, however, shuffle batches / beams between time steps and 125 | `dynamic_decode` will mix up the finished state across these entries because 126 | it does not track the reshuffle across time steps. In this case, it is 127 | up to the decoder to declare that it will keep track of its own finished 128 | state by setting this property to `True`. 129 | 130 | Returns: 131 | Python bool. 132 | """ 133 | return False 134 | 135 | 136 | def _create_zero_outputs(size, dtype, batch_size): 137 | """Create a zero outputs Tensor structure.""" 138 | def _create(s, d): 139 | return _zero_state_tensors(s, batch_size, d) 140 | 141 | return nest.map_structure(_create, size, dtype) 142 | 143 | 144 | def dynamic_decode(decoder, 145 | output_time_major=False, 146 | impute_finished=False, 147 | maximum_iterations=None, 148 | parallel_iterations=32, 149 | swap_memory=False, 150 | scope=None): 151 | """Perform dynamic decoding with `decoder`. 152 | 153 | Calls initialize() once and step() repeatedly on the Decoder object. 154 | 155 | Args: 156 | decoder: A `Decoder` instance. 157 | output_time_major: Python boolean. Default: `False` (batch major). If 158 | `True`, outputs are returned as time major tensors (this mode is faster). 159 | Otherwise, outputs are returned as batch major tensors (this adds extra 160 | time to the computation). 161 | impute_finished: Python boolean. If `True`, then states for batch 162 | entries which are marked as finished get copied through and the 163 | corresponding outputs get zeroed out. This causes some slowdown at 164 | each time step, but ensures that the final state and outputs have 165 | the correct values and that backprop ignores time steps that were 166 | marked as finished. 167 | maximum_iterations: `int32` scalar, maximum allowed number of decoding 168 | steps. Default is `None` (decode until the decoder is fully done). 169 | parallel_iterations: Argument passed to `tf.while_loop`. 170 | swap_memory: Argument passed to `tf.while_loop`. 171 | scope: Optional variable scope to use. 172 | 173 | Returns: 174 | `(final_outputs, final_state, final_sequence_lengths)`. 175 | 176 | Raises: 177 | TypeError: if `decoder` is not an instance of `Decoder`. 178 | ValueError: if `maximum_iterations` is provided but is not a scalar. 179 | """ 180 | if not isinstance(decoder, Decoder): 181 | raise TypeError("Expected decoder to be type Decoder, but saw: %s" % 182 | type(decoder)) 183 | 184 | with variable_scope.variable_scope(scope, "decoder") as varscope: 185 | # Properly cache variable values inside the while_loop 186 | if varscope.caching_device is None: 187 | varscope.set_caching_device(lambda op: op.device) 188 | 189 | if maximum_iterations is not None: 190 | maximum_iterations = ops.convert_to_tensor( 191 | maximum_iterations, dtype=dtypes.int32, name="maximum_iterations") 192 | if maximum_iterations.get_shape().ndims != 0: 193 | raise ValueError("maximum_iterations must be a scalar") 194 | 195 | initial_finished, initial_inputs, initial_state = decoder.initialize() 196 | 197 | zero_outputs = _create_zero_outputs(decoder.output_size, 198 | decoder.output_dtype, 199 | decoder.batch_size) 200 | 201 | if maximum_iterations is not None: 202 | initial_finished = math_ops.logical_or( 203 | initial_finished, 0 >= maximum_iterations) 204 | initial_sequence_lengths = array_ops.zeros_like( 205 | initial_finished, dtype=dtypes.int32) 206 | initial_time = constant_op.constant(0, dtype=dtypes.int32) 207 | 208 | def _shape(batch_size, from_shape): 209 | if (not isinstance(from_shape, tensor_shape.TensorShape) or 210 | from_shape.ndims == 0): 211 | return tensor_shape.TensorShape(None) 212 | else: 213 | batch_size = tensor_util.constant_value( 214 | ops.convert_to_tensor( 215 | batch_size, name="batch_size")) 216 | return tensor_shape.TensorShape([batch_size]).concatenate(from_shape) 217 | 218 | def _create_ta(s, d): 219 | return tensor_array_ops.TensorArray( 220 | dtype=d, 221 | size=0, 222 | dynamic_size=True, 223 | element_shape=_shape(decoder.batch_size, s)) 224 | 225 | initial_outputs_ta = nest.map_structure(_create_ta, decoder.output_size, 226 | decoder.output_dtype) 227 | 228 | def condition(unused_time, unused_outputs_ta, unused_state, unused_inputs, 229 | finished, unused_sequence_lengths): 230 | return math_ops.logical_not(math_ops.reduce_all(finished)) 231 | 232 | def body(time, outputs_ta, state, inputs, finished, sequence_lengths): 233 | """Internal while_loop body. 234 | 235 | Args: 236 | time: scalar int32 tensor. 237 | outputs_ta: structure of TensorArray. 238 | state: (structure of) state tensors and TensorArrays. 239 | inputs: (structure of) input tensors. 240 | finished: bool tensor (keeping track of what's finished). 241 | sequence_lengths: int32 tensor (keeping track of time of finish). 242 | 243 | Returns: 244 | `(time + 1, outputs_ta, next_state, next_inputs, next_finished, 245 | next_sequence_lengths)`. 246 | ``` 247 | """ 248 | (next_outputs, decoder_state, next_inputs, 249 | decoder_finished) = decoder.step(time, inputs, state) 250 | if decoder.tracks_own_finished: 251 | next_finished = decoder_finished 252 | else: 253 | next_finished = math_ops.logical_or(decoder_finished, finished) 254 | if maximum_iterations is not None: 255 | next_finished = math_ops.logical_or( 256 | next_finished, time + 1 >= maximum_iterations) 257 | next_sequence_lengths = array_ops.where( 258 | math_ops.logical_and(math_ops.logical_not(finished), next_finished), 259 | array_ops.fill(array_ops.shape(sequence_lengths), time + 1), 260 | sequence_lengths) 261 | 262 | nest.assert_same_structure(state, decoder_state) 263 | nest.assert_same_structure(outputs_ta, next_outputs) 264 | nest.assert_same_structure(inputs, next_inputs) 265 | 266 | # Zero out output values past finish 267 | if impute_finished: 268 | emit = nest.map_structure( 269 | lambda out, zero: array_ops.where(finished, zero, out), 270 | next_outputs, 271 | zero_outputs) 272 | else: 273 | emit = next_outputs 274 | 275 | # Copy through states past finish 276 | def _maybe_copy_state(new, cur): 277 | # TensorArrays and scalar states get passed through. 278 | if isinstance(cur, tensor_array_ops.TensorArray): 279 | pass_through = True 280 | else: 281 | new.set_shape(cur.shape) 282 | pass_through = (new.shape.ndims == 0) 283 | return new if pass_through else array_ops.where(finished, cur, new) 284 | 285 | if impute_finished: 286 | next_state = nest.map_structure( 287 | _maybe_copy_state, decoder_state, state) 288 | else: 289 | next_state = decoder_state 290 | 291 | outputs_ta = nest.map_structure(lambda ta, out: ta.write(time, out), 292 | outputs_ta, emit) 293 | return (time + 1, outputs_ta, next_state, next_inputs, next_finished, 294 | next_sequence_lengths) 295 | 296 | res = control_flow_ops.while_loop( 297 | condition, 298 | body, 299 | loop_vars=[ 300 | initial_time, initial_outputs_ta, initial_state, initial_inputs, 301 | initial_finished, initial_sequence_lengths, 302 | ], 303 | parallel_iterations=parallel_iterations, 304 | swap_memory=swap_memory) 305 | 306 | final_outputs_ta = res[1] 307 | final_state = res[2] 308 | final_sequence_lengths = res[5] 309 | 310 | final_outputs = nest.map_structure(lambda ta: ta.stack(), final_outputs_ta) 311 | 312 | try: 313 | final_outputs, final_state = decoder.finalize( 314 | final_outputs, final_state, final_sequence_lengths) 315 | except NotImplementedError: 316 | pass 317 | 318 | if not output_time_major: 319 | final_outputs = nest.map_structure(_transpose_batch_time, final_outputs) 320 | 321 | return final_outputs, final_state, final_sequence_lengths 322 | -------------------------------------------------------------------------------- /Trip/data/TKY_split400_set.dat: -------------------------------------------------------------------------------- 1 | 187-21480 1 0 18 2 | 205-23354 9 1 7 3 | 187-21486 1 0 18 4 | 321-35723 16 1 9 1 5 | 321-35722 1 16 16 6 | 321-35725 16 1 9 7 | 257-28597 5 5 2 8 | 353-39412 14 3 3 10 9 | 151-17418 15 7 4 10 | 288-32428 0 17 1 1 11 | 151-17411 7 1 1 7 12 | 162-18806 25 20 20 25 13 | 192-22008 22 24 9 14 | 203-23260 5 4 6 15 | 203-23263 5 2 2 16 | 338-37598 3 3 0 17 | 192-22004 5 19 3 18 | 55-6316 0 10 0 19 | 251-28049 7 20 6 20 | 135-15393 3 12 12 3 21 | 117-13773 8 1 1 22 | 336-37434 1 9 14 23 | 29-3680 2 20 2 24 | 117-13779 6 20 20 25 | 8-788 4 1 19 26 | 118-13805 28 2 2 1 27 | 118-13806 1 4 17 0 10 26 28 28 | 65-7834 3 0 5 29 | 163-18882 0 28 18 30 | 134-15360 9 1 0 31 | 302-34017 0 6 12 32 | 288-32417 0 1 9 1 33 | 113-13048 12 0 0 34 | 396-44060 21 7 3 35 | 117-13783 1 1 6 36 | 396-44109 2 25 3 37 | 284-31715 0 1 1 38 | 51-5874 17 0 15 39 | 365-40845 2 0 0 40 | 336-37393 1 3 0 41 | 36-4372 3 3 15 42 | 159-18400 4 1 3 43 | 43-4908 22 6 1 44 | 63-7408 0 10 10 4 45 | 227-25674 0 3 15 46 | 227-25678 1 17 4 47 | 385-42788 4 1 9 48 | 235-26451 9 9 1 49 | 385-42786 23 4 1 50 | 385-42781 0 2 0 0 10 51 | 14-1494 1 1 21 52 | 196-22454 7 1 17 53 | 30-3739 15 0 0 54 | 195-22360 1 2 11 55 | 196-22452 3 17 4 7 56 | 12-1275 1 1 9 1 57 | 113-12997 0 2 12 58 | 12-1277 21 1 9 59 | 30-3731 15 3 14 14 15 60 | 12-1270 7 1 9 61 | 12-1273 1 8 9 62 | 113-12991 25 8 4 0 63 | 293-32875 4 15 15 64 | 243-27221 27 24 22 65 | 18-2243 1 9 10 66 | 342-37981 14 3 3 0 67 | 68-8190 0 25 4 1 68 | 29-3545 3 3 0 69 | 243-27223 1 24 22 70 | 35-4137 0 17 0 71 | 289-32514 20 6 23 72 | 289-32515 23 6 1 73 | 26-3077 3 3 2 74 | 102-11773 22 0 0 1 9 22 75 | 26-3075 3 2 3 76 | 26-3072 3 27 21 27 0 0 77 | 130-14983 28 0 15 78 | 196-22535 0 1 4 79 | 196-22532 17 1 7 80 | 161-18553 0 0 6 81 | 282-31521 2 1 0 82 | 250-27889 1 1 6 83 | 30-3748 15 3 0 84 | 30-3747 2 25 4 85 | 285-31914 28 0 3 86 | 324-36006 2 1 3 87 | 30-3741 15 0 0 88 | 83-9524 3 15 15 3 89 | 396-44093 3 25 6 12 90 | 196-22433 7 1 17 91 | 187-21600 3 9 3 92 | 35-4142 0 2 11 93 | 289-32519 7 1 29 94 | 277-30964 1 8 12 95 | 21-2519 1 9 1 96 | 13-1352 10 14 14 97 | 227-25695 2 3 15 98 | 227-25694 15 2 11 99 | 43-4884 24 6 1 100 | 21-2620 1 1 15 101 | 124-14432 7 7 9 102 | 284-31681 0 17 1 103 | 196-22508 1 1 25 104 | 147-17030 14 1 1 15 105 | 365-40796 23 0 10 106 | 113-13009 12 6 0 107 | 45-5035 1 9 21 108 | 339-37659 0 12 12 0 109 | 230-25971 10 17 1 110 | 113-13004 0 2 12 111 | 126-14577 6 1 9 112 | 151-17465 0 17 1 113 | 151-17463 7 1 7 114 | 151-17462 15 7 1 9 1 7 115 | 241-26960 0 1 1 116 | 36-4335 2 0 0 117 | 203-23251 5 6 2 2 118 | 336-37348 12 1 17 0 119 | 206-23359 0 1 9 4 120 | 365-40902 23 2 0 121 | 254-28315 2 6 5 122 | 146-16892 7 1 1 19 123 | 342-38144 6 6 2 124 | 13-1420 10 14 14 125 | 338-37622 4 17 1 1 6 22 126 | 124-14404 3 3 15 127 | 354-39503 2 0 3 128 | 338-37621 4 1 1 4 129 | 338-37627 25 16 16 130 | 235-26454 1 9 9 131 | 150-17361 22 1 9 17 132 | 359-40117 15 27 4 133 | 338-37619 4 1 1 4 134 | 118-13836 4 1 1 0 135 | 118-13830 0 14 0 1 1 1 4 136 | 244-27236 1 1 22 137 | 361-40260 2 1 9 138 | 68-8146 2 3 3 139 | 65-7867 5 19 3 140 | 302-34046 0 2 2 141 | 325-36187 27 9 7 142 | 201-23093 0 17 1 143 | 18-2048 1 9 10 144 | 18-1965 4 0 10 145 | 227-25783 15 0 2 2 3 7 146 | 119-13923 13 13 13 147 | 113-12975 0 6 12 148 | 96-11178 19 0 2 149 | 113-12979 12 0 0 150 | 76-8967 4 8 8 4 151 | 76-8963 1 1 4 152 | 146-16808 1 9 8 153 | 65-7783 3 3 5 154 | 390-43105 2 3 3 21 155 | 65-7788 4 4 5 156 | 235-26402 1 9 1 1 157 | 86-9746 12 5 5 158 | 86-9748 12 25 25 159 | 196-22403 7 1 17 160 | 285-31898 2 0 0 3 161 | 196-22407 7 1 17 162 | 369-41514 0 0 2 163 | 198-22859 20 2 15 164 | 196-22656 17 1 7 165 | 26-3042 3 2 2 166 | 196-22654 17 1 7 167 | 63-7541 21 1 9 168 | 1-14 1 9 9 1 7 169 | 1-17 7 1 1 170 | 1-13 27 4 0 171 | 289-32523 6 1 7 172 | 141-16263 11 10 10 173 | 96-11088 2 0 0 2 174 | 196-22663 7 1 17 175 | 227-25713 0 3 15 176 | 289-32538 2 6 29 177 | 49-5556 11 13 13 178 | 196-22544 17 1 7 179 | 365-40906 23 2 2 180 | 196-22540 17 1 7 181 | 342-38136 1 1 4 4 182 | 196-22548 9 1 7 183 | 12-1313 7 1 0 184 | 12-1317 1 4 24 23 5 185 | 12-1315 1 24 1 186 | 26-3038 3 3 2 187 | 187-21492 1 17 0 13 188 | 365-40750 23 2 2 0 189 | 187-21496 7 0 18 190 | 321-35734 9 1 16 191 | 102-11766 0 25 22 192 | 47-5121 12 23 23 193 | 365-40759 0 2 2 23 194 | 187-21556 1 0 18 195 | 187-21653 1 0 18 196 | 76-9006 1 1 2 197 | 76-9008 4 0 1 198 | 339-37759 0 1 1 17 199 | 205-23341 0 17 1 200 | 350-39085 11 11 0 201 | 8-772 19 25 4 202 | 288-32436 1 1 0 203 | 384-42749 3 0 0 10 204 | 187-21446 1 0 18 205 | 365-40831 2 23 23 0 206 | 55-6365 0 17 17 4 207 | 251-28038 12 1 9 208 | 205-23344 9 9 1 4 209 | 117-13764 6 1 1 210 | 79-9121 0 1 3 211 | 117-13760 8 1 1 6 212 | 248-27628 0 1 9 213 | 36-4391 4 1 7 214 | 105-12125 0 17 1 215 | 105-12123 6 2 0 0 216 | 105-12120 0 1 9 1 0 217 | 65-7820 3 0 2 2 5 218 | 353-39452 14 0 0 14 219 | 321-35740 12 12 16 220 | 98-11333 0 2 11 221 | 29-3603 2 3 3 222 | 288-32423 0 17 1 223 | 230-25943 10 17 1 9 9 27 224 | 385-42771 6 16 3 225 | 116-13686 25 0 0 226 | 187-21449 1 17 0 227 | 320-35625 1 9 1 228 | 164-19004 3 3 4 229 | 271-30356 1 9 1 230 | 283-31623 25 4 25 231 | 252-28148 20 23 5 232 | 204-23291 1 1 0 233 | 227-25665 0 3 15 234 | 302-34092 0 3 0 235 | 227-25660 3 3 15 236 | 147-16959 26 26 0 237 | 227-25663 1 1 27 238 | 1-5 1 1 7 239 | 5-457 1 1 2 240 | 398-44303 0 3 1 241 | 385-42778 10 4 8 242 | 235-26330 1 9 1 243 | 135-15520 3 8 8 12 244 | 336-37377 1 9 3 245 | 168-19371 10 10 0 246 | 65-7896 3 0 2 6 247 | 196-22442 7 1 17 248 | 347-38808 4 1 9 249 | 359-40111 15 27 1 7 250 | 124-14403 3 3 15 251 | 12-1266 7 1 8 2 1 252 | 65-7951 5 6 3 253 | 65-7952 3 1 5 254 | 289-32525 27 7 1 255 | 285-31860 11 2 3 256 | 12-1280 4 1 9 7 257 | 63-7500 1 4 10 10 0 25 258 | 99-11541 10 4 7 259 | 21-2593 1 9 1 260 | 102-11764 9 1 22 261 | 289-32528 23 6 2 1 7 262 | 33-3980 5 1 5 263 | 352-39346 27 21 0 264 | 205-23342 5 1 9 265 | 196-22506 1 1 7 266 | 371-41651 4 6 19 267 | 205-23349 25 1 3 3 1 268 | 196-22509 17 1 4 269 | 58-6792 8 23 5 270 | 49-5519 19 13 13 271 | 161-18546 2 0 16 272 | 161-18540 6 2 13 273 | 282-31536 2 0 0 2 274 | 332-36951 24 24 7 275 | 135-15459 3 12 8 3 276 | 233-26198 9 1 21 277 | 26-2994 3 7 21 7 278 | 96-11031 2 0 0 279 | 298-33660 13 0 17 0 280 | 30-3754 15 3 14 281 | 38-4513 6 4 1 282 | 274-30607 9 1 15 283 | 55-6339 2 2 6 284 | 38-4519 6 4 1 285 | 113-13124 12 6 0 286 | 327-36504 2 6 29 287 | 395-43957 7 2 1 288 | 196-22595 1 7 7 289 | 227-25680 1 9 1 290 | 196-22597 17 1 7 291 | 43-4894 1 3 1 292 | 196-22592 17 4 1 7 293 | 227-25685 15 2 11 294 | 100-11596 17 1 1 295 | 164-18991 17 3 3 296 | 187-21468 1 0 18 297 | 47-5173 0 12 23 298 | 187-21462 1 0 18 299 | 187-21461 18 0 1 17 300 | 353-39377 1 1 0 301 | 163-18855 19 19 5 302 | 353-39470 0 14 14 303 | 187-21501 1 0 18 304 | 339-37710 0 1 17 305 | 230-25985 10 17 9 306 | 45-5026 3 9 1 307 | 187-21509 1 0 18 308 | 151-17473 1 17 0 2 309 | 113-13010 0 4 17 0 310 | 187-21667 1 0 18 311 | 385-42782 0 16 6 312 | 196-22424 1 1 7 313 | 31-3794 25 4 7 314 | 31-3791 8 4 7 315 | 235-26326 1 9 9 9 316 | 13-1434 0 0 10 317 | 13-1436 0 14 14 318 | 146-16881 1 9 1 319 | 146-16884 2 0 17 1 320 | 338-37618 2 2 11 321 | 117-13717 6 1 1 322 | 78-9073 0 0 1 323 | 117-13714 5 1 1 324 | 13-1439 4 17 1 325 | 338-37611 17 1 1 4 326 | 126-14605 1 1 7 327 | 365-40883 23 0 0 328 | 196-22486 7 1 17 329 | 126-14602 0 0 15 330 | 196-22485 7 1 1 4 7 331 | 232-26125 12 1 9 7 332 | 318-35554 6 12 8 4 1 17 333 | 318-35555 1 9 1 4 8 6 334 | 124-14441 15 7 9 4 4 335 | 68-8159 27 4 0 336 | 68-8157 12 1 1 12 27 337 | 59-6857 14 10 28 338 | 68-8155 12 1 9 339 | 68-8153 0 17 1 1 7 340 | 68-8150 12 4 1 1 341 | 113-13066 12 0 0 342 | 201-23083 0 1 9 343 | 365-40826 23 10 10 344 | 30-3733 6 4 15 345 | 29-3672 3 2 1 3 346 | 113-12994 12 6 0 347 | 29-3677 3 25 25 3 348 | 43-4920 22 1 9 349 | 96-11160 2 14 2 0 350 | 128-14716 9 0 1 351 | 5-462 6 29 29 23 352 | 96-11168 2 0 0 2 353 | 293-32982 1 9 7 15 354 | 293-33018 7 1 15 355 | 12-1272 7 1 9 8 1 9 356 | 196-22467 9 1 7 357 | 30-3716 24 1 7 358 | 33-4012 5 26 1 359 | 196-22435 7 1 17 360 | 86-9771 4 10 26 361 | 338-37608 6 4 7 0 0 2 362 | 18-2222 1 9 10 363 | 14-1567 21 1 9 364 | 342-37998 14 3 3 365 | 63-7470 2 2 25 366 | 257-28689 2 11 19 5 5 367 | 257-28687 5 5 5 368 | 196-22558 1 7 15 369 | 194-22311 27 7 1 370 | 194-22313 1 12 12 371 | 265-29658 28 10 0 372 | 257-28696 5 16 5 5 373 | 334-37195 25 24 1 374 | 199-22883 15 3 1 9 0 375 | 336-37476 0 1 1 376 | 285-31831 11 11 11 377 | 289-32561 1 7 27 378 | 385-42774 6 3 0 10 379 | 385-42776 0 28 0 4 7 380 | 385-42777 2 3 0 381 | 285-31931 11 2 3 382 | 394-43872 0 26 10 10 383 | 353-39438 3 14 14 384 | 173-20014 0 28 28 0 385 | 227-25686 11 2 15 386 | 98-11405 14 0 0 387 | 187-21540 1 0 18 388 | 12-1295 29 29 8 1 1 7 7 389 | 187-21542 1 0 18 390 | 113-13111 0 2 2 391 | 18-2127 1 9 10 392 | 289-32511 7 23 12 393 | 80-9239 4 3 14 394 | 264-29517 27 9 1 395 | 289-32517 29 1 7 27 396 | 192-22025 3 2 3 397 | 227-25774 1 9 1 398 | 365-40963 23 2 0 399 | 365-40809 0 6 23 23 400 | 365-40967 23 2 2 0 401 | 105-12113 0 3 7 7 402 | 65-7694 8 5 5 403 | 105-12114 7 3 0 404 | 233-26226 19 5 23 405 | 284-31690 0 1 1 406 | 163-18866 18 0 0 18 407 | 187-21531 1 0 18 408 | 65-7814 5 25 2 5 409 | 353-39444 8 8 14 410 | 187-21537 1 0 18 411 | 248-27622 1 9 1 412 | 36-4312 20 20 16 2 413 | 164-19014 3 2 0 414 | 257-28703 5 6 4 5 415 | 230-25959 10 1 10 416 | 304-34213 9 1 4 417 | 161-18703 6 2 16 418 | 150-17281 22 8 22 419 | 78-9083 2 1 0 420 | 78-9080 2 2 1 421 | 65-7751 4 19 5 422 | 75-8898 3 0 11 423 | 21-2449 0 3 15 424 | 13-1373 10 0 2 11 425 | 371-41639 3 0 17 1 426 | 44-4923 2 1 1 427 | 196-22477 7 1 17 428 | 196-22476 7 1 17 429 | 371-41638 7 1 9 430 | 168-19362 9 1 16 431 | 359-40109 15 4 0 432 | 68-8160 1 1 12 433 | 65-7961 3 17 4 5 434 | 318-35567 21 27 4 8 435 | 318-35566 5 6 8 27 436 | 359-40106 17 1 4 7 437 | 65-7964 5 6 3 438 | 318-35563 1 9 1 439 | 365-40856 23 23 2 440 | 13-1382 10 0 0 10 441 | 18-2062 14 3 3 0 442 | 117-13695 8 8 8 443 | 201-23070 2 2 0 444 | 196-22394 7 17 1 445 | 365-40817 2 2 0 446 | 96-11052 2 0 0 2 447 | 147-16945 0 0 0 448 | 63-7514 9 1 4 449 | 336-37360 15 0 17 1 9 450 | 102-11756 24 4 1 9 451 | 371-41647 21 1 4 452 | 205-23352 2 1 9 1 4 453 | 141-16237 11 17 1 454 | 49-5527 4 28 28 28 455 | 49-5526 1 9 1 0 456 | 21-2619 0 1 1 457 | 196-22512 17 1 7 458 | 21-2614 1 1 12 12 1 9 459 | 233-26178 1 9 1 460 | 26-3083 4 4 16 461 | 310-34863 0 25 24 462 | 350-39116 0 16 5 463 | 272-30461 9 7 27 3 464 | 187-21668 1 0 18 465 | 246-27430 12 1 2 1 466 | 369-41531 6 1 1 467 | 124-14502 15 3 1 1 468 | 96-11122 19 18 18 19 469 | 384-42762 3 2 3 470 | 384-42763 3 2 0 471 | 396-44091 3 0 27 472 | 339-37701 0 1 1 473 | 384-42764 0 0 2 474 | 384-42765 3 0 10 475 | 264-29556 1 4 6 476 | 250-27903 1 0 2 6 477 | 63-7560 2 0 10 0 478 | 360-40198 0 0 1 9 479 | 194-22320 21 27 27 480 | 296-33441 13 13 18 481 | 161-18522 6 2 13 482 | 290-32582 0 1 9 483 | 289-32532 27 7 1 484 | 285-31805 0 0 3 485 | 187-21478 1 0 18 486 | 187-21475 1 0 18 487 | 187-21476 0 1 3 488 | 1-2 1 9 1 7 489 | 187-21470 9 1 17 0 490 | 255-28489 1 9 1 491 | 1-7 7 1 9 9 1 7 492 | 388-42942 0 17 1 493 | 353-39403 1 1 14 494 | 187-21572 1 0 18 495 | 55-6394 20 8 6 496 | 151-17401 0 1 7 497 | 231-26032 3 0 27 498 | 192-22010 5 4 1 9 1 499 | 203-23278 5 1 1 500 | 18-2110 1 9 10 501 | 365-40727 2 23 23 0 502 | 398-44301 1 9 1 503 | 338-37600 0 2 11 504 | 338-37601 6 1 4 505 | 338-37603 4 1 1 4 506 | 338-37605 2 0 10 10 0 507 | 338-37607 6 1 1 4 508 | 13-1409 14 26 26 10 509 | 58-6744 5 6 8 510 | 248-27602 1 9 1 511 | 128-14744 17 0 0 17 512 | 65-7844 5 6 0 14 513 | 65-7840 3 1 5 514 | 196-22629 17 1 7 515 | 233-26258 6 4 1 516 | 359-40112 27 4 1 7 517 | 59-6840 7 1 0 518 | 113-13077 12 6 0 519 | 113-13074 12 2 0 520 | 113-13075 2 0 0 521 | 18-2184 1 9 10 522 | 151-17497 26 10 0 523 | 342-38011 14 1 7 14 524 | 117-13794 8 7 27 525 | 117-13793 1 1 6 526 | 369-41526 2 1 1 14 527 | 36-4340 23 23 15 528 | 18-2026 1 9 10 529 | 151-17498 0 1 7 530 | 96-11155 26 2 2 531 | 9-1057 7 1 8 12 12 6 532 | 159-18413 0 17 1 533 | 5-478 1 1 1 534 | 267-29919 2 0 25 535 | 243-27202 1 15 0 536 | 135-15421 3 12 12 537 | 13-1320 7 1 1 538 | 243-27204 1 24 22 539 | 13-1323 0 0 14 540 | 146-16829 16 1 23 541 | 243-27209 1 24 22 542 | 13-1328 10 0 0 10 543 | 196-22422 7 1 17 544 | 30-3729 0 14 3 15 545 | 196-22426 14 14 3 15 546 | 30-3726 0 3 15 547 | 113-12981 0 2 2 548 | 113-12982 12 0 0 549 | 293-32860 0 0 15 550 | 336-37380 12 1 9 551 | 63-7462 4 1 9 552 | 124-14399 12 8 4 1 553 | 293-32869 4 1 1 554 | 302-34096 0 3 3 0 555 | 293-32907 4 1 1 556 | 21-2577 1 1 4 557 | 57-6566 4 0 1 558 | 100-11639 1 9 1 559 | 342-38150 4 1 9 1 12 560 | 196-22473 1 7 21 561 | 196-22640 7 1 7 562 | 196-22528 17 1 7 563 | 63-7489 4 1 9 1 4 564 | 58-6757 5 8 1 565 | 377-42126 7 20 20 566 | 390-43158 1 1 24 567 | 336-37466 1 7 17 0 568 | 35-4191 10 0 2 569 | 338-37616 4 1 1 4 570 | 324-36010 2 0 14 571 | 76-8948 1 0 16 572 | 191-21905 9 8 8 573 | 369-41544 14 8 6 14 574 | 146-16910 19 4 1 575 | 296-33384 13 0 1 576 | 96-11204 2 0 2 577 | 191-21908 28 10 0 9 578 | 342-38096 17 4 0 579 | 124-14398 1 4 8 12 580 | 143-16516 17 1 9 581 | 151-17527 0 17 0 0 582 | 18-2096 1 9 10 583 | 327-36525 2 0 2 584 | 196-22633 17 1 7 585 | 196-22634 7 17 1 586 | 360-40141 3 1 9 587 | 332-37021 1 17 21 588 | 233-26212 0 1 1 589 | 196-22580 7 1 4 590 | 65-7916 4 6 5 591 | 116-13467 25 16 16 592 | 147-17041 25 25 0 593 | 100-11630 26 17 9 1 594 | 187-21528 1 0 18 595 | 187-21529 1 0 18 596 | 187-21522 1 0 18 597 | 230-25960 14 0 1 9 1 10 598 | 10-1112 2 1 9 7 599 | 151-17450 0 0 0 600 | 151-17457 0 17 1 601 | 284-31744 0 0 1 602 | 8-888 19 3 1 603 | 179-20532 5 23 23 5 604 | 18-1984 24 10 8 605 | 113-13039 12 0 0 606 | 65-7721 5 8 4 5 607 | 233-26236 0 17 1 1 608 | 12-1288 7 1 17 609 | 78-9076 0 0 3 610 | 338-37632 2 11 1 6 611 | 146-16860 1 9 1 4 19 612 | 78-9072 1 1 0 613 | 233-26230 26 10 0 17 4 6 614 | 126-14628 4 17 1 9 615 | 126-14629 1 7 12 616 | 13-1361 10 14 0 617 | 196-22466 9 1 7 618 | 118-13843 14 0 1 619 | 68-8176 15 2 3 27 620 | 365-40812 23 2 2 0 621 | 250-27827 24 6 4 622 | 107-12395 14 6 5 5 6 623 | 246-27437 8 4 4 624 | 68-8178 16 4 4 625 | 13-1444 10 0 0 626 | 35-4228 6 23 5 627 | 13-1446 6 2 11 2 628 | 8-839 19 4 1 4 629 | 365-40805 10 2 23 630 | 302-34036 4 0 3 631 | 102-11744 9 1 4 632 | 102-11745 4 10 26 633 | 76-8979 4 1 4 634 | 29-3576 2 20 2 635 | 371-41637 10 0 17 636 | 264-29643 1 9 0 637 | 49-5532 0 0 13 638 | 49-5536 9 1 0 0 639 | 146-16810 1 6 22 640 | 390-43117 1 24 22 641 | 196-22418 7 1 17 642 | 102-11759 12 1 9 1 0 643 | 196-22411 7 1 17 644 | 196-22413 7 1 17 645 | 124-14510 1 4 12 646 | 198-22866 4 7 15 647 | 193-22193 3 3 2 2 648 | 63-7458 25 10 0 25 649 | 396-44085 21 1 1 3 650 | 384-42754 3 0 18 651 | 384-42752 0 0 3 652 | 201-23085 0 2 2 1 2 0 653 | 9-987 1 1 4 654 | 289-32494 21 1 23 655 | 289-32493 23 23 1 656 | 227-25706 15 2 11 657 | 49-5547 11 2 13 13 658 | 396-44110 3 25 24 659 | 365-40938 23 16 0 660 | 196-22575 17 1 7 661 | 370-41605 11 2 12 12 2 11 662 | 22-2708 21 3 14 663 | 296-33450 0 10 13 664 | 205-23350 1 0 10 17 665 | 49-5548 13 2 11 666 | 285-31818 3 2 2 667 | 289-32549 2 2 1 668 | 12-1303 1 1 9 669 | 29-3609 2 20 2 670 | 205-23357 9 1 4 671 | -------------------------------------------------------------------------------- /Trip/data/Melb_set.dat: -------------------------------------------------------------------------------- 1 | 48453097@N00-2242 33 0 7 2 | 48453097@N00-2243 52 25 7 3 | 48453097@N00-2240 19 27 6 4 4 | 48453097@N00-2241 6 13 31 5 | 91151414@N00-4472 1 27 49 14 0 4 6 | 93913731@N00-4783 9 24 52 7 | 31148157@N00-1241 47 2 49 15 8 | 8502118@N08-4217 35 48 32 33 30 24 14 58 7 4 9 | 12900555@N02-212 29 27 24 10 | 17211040@N00-413 13 0 4 11 | 8502118@N08-4219 20 34 27 49 54 38 18 15 44 12 | 44124479801@N01-2042 13 14 0 4 13 | 17211040@N00-414 6 0 8 14 | 14197203@N05-328 11 76 24 15 | 64015960@N00-3128 29 2 3 16 | 14197203@N05-325 28 26 2 17 | 64015960@N00-3127 38 18 16 31 5 22 18 | 30206889@N00-1150 5 25 0 7 19 | 52618473@N02-2469 19 4 12 20 | 59551944@N00-2893 21 25 4 21 | 48213112@N00-2223 1 18 15 22 | 56384399@N00-2673 54 77 79 25 0 4 23 | 65638600@N05-3187 20 58 16 24 | 51035603197@N01-2408 24 5 4 25 | 79235216@N00-3888 46 3 5 22 26 | 65779544@N00-3201 1 36 2 27 | 23875378@N06-727 35 24 14 0 4 28 | 7414413@N05-3534 36 73 4 29 | 16867627@N06-394 20 51 38 18 16 44 53 56 30 | 16867627@N06-395 48 45 19 29 27 9 11 57 50 81 53 0 12 31 | 72327345@N00-3477 38 5 22 32 | 45676902@N00-2099 24 79 7 33 | 8502118@N08-4216 1 0 7 34 | 91256982@N00-4621 48 43 12 35 | 27759575@N05-959 29 51 56 36 | 91748271@N00-4650 42 49 8 37 | 52952793@N00-2505 19 24 14 0 7 38 | 56935359@N00-2678 45 17 13 39 | 37468950@N08-1685 48 28 32 17 33 9 30 3 38 44 21 0 61 12 40 | 37468950@N08-1686 1 6 15 31 52 7 8 41 | 60111909@N00-2903 27 15 24 14 0 56 7 8 42 | 60111909@N00-2904 24 58 5 43 0 56 7 43 | 58085609@N00-2824 5 66 7 44 | 78681550@N00-3869 9 30 11 49 12 45 | 78681550@N00-3868 5 0 4 46 | 62948807@N00-3007 33 2 4 47 | 28685313@N00-1075 28 9 71 75 4 12 48 | 87584208@N00-4389 1 9 4 49 | 33200530@N04-1320 1 33 26 3 24 0 7 50 | 24761036@N00-798 32 19 33 51 | 21160385@N02-560 9 14 12 52 | 10966462@N02-147 27 6 15 8 53 | 23806189@N00-721 28 19 36 14 55 22 54 | 23806189@N00-720 34 5 22 8 55 | 79925938@N00-3963 34 26 9 56 | 69931111@N00-3408 1 2 51 56 57 | 67774014@N00-3338 33 3 22 58 | 49456569@N06-2301 1 5 22 4 59 | 49456569@N06-2302 1 5 22 60 | 8429346@N03-4160 33 22 0 4 61 | 89743353@N00-4450 36 2 8 12 62 | 89165847@N00-4431 5 0 4 63 | 45676902@N00-2101 27 30 25 56 64 | 75873727@N00-3626 2 14 5 0 7 65 | 84987970@N00-4205 11 3 54 7 66 | 12284993@N00-198 1 10 18 5 67 | 14021337@N00-291 1 6 15 5 0 66 8 68 | 9433773@N08-4817 1 10 36 3 24 5 69 | 58673657@N07-2857 1 26 36 70 | 48583839@N07-2268 1 19 36 3 5 81 0 4 71 | 15213592@N03-385 19 30 0 4 72 | 87355829@N00-4316 32 33 26 2 73 | 87393953@N00-4368 28 30 5 0 74 | 57768536@N05-2700 20 1 32 17 38 18 0 75 | 99804259@N00-5096 19 58 12 76 | 60451415@N00-2912 18 21 25 0 77 | 36656940@N00-1593 42 2 3 8 78 | 13237839@N00-241 57 73 59 79 | 13237839@N00-245 28 36 9 2 14 39 0 80 | 27759575@N05-961 6 31 52 73 59 81 | 13233236@N04-232 0 4 12 82 | 27759575@N05-966 17 2 22 8 83 | 10559879@N00-64 20 1 2 18 84 | 10559879@N00-63 17 33 18 85 | 43929620@N05-1984 3 18 14 0 86 | 10936890@N00-145 19 18 16 5 53 22 4 8 87 | 43929620@N05-1982 38 63 61 88 | 26071009@N04-912 35 3 14 5 25 0 89 | 10681948@N00-129 45 36 18 90 | 19447876@N00-443 19 5 0 91 | 19447876@N00-445 14 73 59 7 92 | 33062815@N00-1315 32 30 3 4 93 | 84987970@N00-4210 2 16 44 94 | 84987970@N00-4213 65 47 54 95 | 84987970@N00-4212 34 1 17 26 2 49 58 8 96 | 84987970@N00-4214 1 47 38 53 12 97 | 24510946@N07-779 35 24 14 98 | 24510946@N07-778 1 3 0 99 | 25287507@N02-835 1 19 3 5 100 | 7878308@N04-3878 45 16 0 101 | 8329215@N04-4133 28 15 14 4 102 | 47980775@N02-2185 29 62 25 103 | 46778425@N00-2144 3 24 0 69 104 | 8329215@N04-4134 35 9 11 24 14 5 0 7 4 105 | 28067218@N00-1017 20 33 5 7 4 106 | 77152181@N06-3758 26 36 11 2 3 107 | 77152181@N06-3757 3 5 22 0 4 108 | 99395734@N08-5062 5 4 8 109 | 77152181@N06-3752 1 6 7 110 | 77152181@N06-3751 35 6 11 111 | 77152181@N06-3750 11 0 4 112 | 9662258@N02-4890 20 68 19 9 14 5 21 40 0 113 | 76395609@N00-3665 20 18 0 114 | 92828317@N00-4710 32 42 27 3 115 | 26085795@N02-916 20 34 28 17 33 29 3 51 18 50 8 116 | 87109756@N02-4290 20 32 33 18 5 22 117 | 30342939@N00-1203 48 28 1 32 17 19 33 9 30 2 3 50 24 14 5 43 0 7 4 12 118 | 99109486@N00-5049 9 30 39 119 | 71088526@N00-3470 9 24 14 39 63 7 120 | 97867688@N00-4935 64 11 51 121 | 8107002@N03-4069 45 6 36 0 4 122 | 87393953@N00-4369 37 7 4 123 | 30206889@N00-1149 30 64 11 24 16 5 55 40 25 53 0 7 124 | 14197203@N05-339 1 22 0 125 | 14197203@N05-337 35 20 18 126 | 64015960@N00-3134 17 2 25 127 | 44690184@N07-2075 19 24 8 128 | 44690184@N07-2071 19 6 24 7 4 129 | 49503156729@N01-2338 6 49 54 130 | 73014677@N05-3513 54 5 0 131 | 99804259@N00-5067 34 42 29 47 132 | 45904697@N05-2111 13 3 0 133 | 63022030@N00-3017 20 45 53 4 134 | 48453097@N00-2254 20 18 0 135 | 99804259@N00-5065 11 31 5 22 0 136 | 91872843@N00-4689 5 7 4 137 | 8281413@N06-4127 14 5 0 138 | 99804259@N00-5066 47 16 63 139 | 80373436@N00-4053 25 0 7 140 | 70995385@N00-3417 14 0 4 141 | 32394656@N00-1284 33 49 0 142 | 22670857@N03-665 3 40 25 143 | 35468141938@N01-1477 28 23 36 11 0 144 | 93012314@N00-4742 1 31 59 7 145 | 52952793@N00-2515 48 9 39 43 12 146 | 52952793@N00-2517 64 76 24 14 31 52 25 7 147 | 14481533@N00-368 1 19 2 15 24 5 0 4 8 148 | 52952793@N00-2519 28 1 26 36 2 0 149 | 25869855@N04-895 0 66 59 150 | 50556076@N00-2379 5 22 0 151 | 12238277@N04-191 70 36 14 0 61 152 | 89864780@N00-4454 35 11 0 153 | 28135215@N00-1042 6 13 15 8 154 | 82698298@N00-4124 76 31 0 7 155 | 32165728@N00-1277 65 62 82 53 156 | 58554451@N00-2850 77 25 0 157 | 42583704@N00-1935 68 28 1 17 36 9 30 2 24 14 5 22 0 7 4 158 | 42583704@N00-1936 27 2 15 8 159 | 42583704@N00-1938 35 45 47 9 11 3 15 76 24 14 16 5 44 53 22 0 63 7 160 | 58301516@N00-2846 1 10 3 18 22 161 | 48264126@N00-2225 13 7 4 162 | 51928417@N00-2430 11 2 66 7 163 | 24761036@N00-785 1 6 36 15 164 | 110608682@N04-150 30 52 0 7 165 | 71088526@N00-3467 3 15 21 25 0 80 4 166 | 41894181187@N01-1925 14 0 4 167 | 87584208@N00-4376 45 34 13 56 168 | 60444977@N00-2908 48 1 32 14 5 22 0 4 8 169 | 60444977@N00-2909 5 0 4 170 | 60444977@N00-2907 35 1 32 33 3 49 15 24 25 0 171 | 28964800@N03-1087 17 26 3 22 172 | 34948727@N00-1397 1 19 2 5 25 173 | 47988174@N04-2220 1 9 4 12 174 | 77235006@N00-3772 1 19 33 3 22 8 175 | 77235006@N00-3771 24 39 12 176 | 77235006@N00-3775 1 3 24 66 177 | 77235006@N00-3774 62 55 40 178 | 86292040@N00-4245 10 2 50 0 179 | 86292040@N00-4244 10 2 50 180 | 53701455@N00-2556 20 16 52 181 | 41894169655@N01-1918 15 14 8 182 | 12284993@N00-200 20 19 2 183 | 12284993@N00-202 20 34 5 0 4 184 | 76726772@N00-3736 9 25 0 4 8 185 | 76726772@N00-3735 62 55 40 186 | 14197203@N05-345 3 5 22 187 | 14197203@N05-342 1 3 15 0 7 4 188 | 87584208@N00-4375 1 9 8 189 | 90716728@N00-4468 3 39 22 190 | 10896675@N00-143 49 15 5 62 0 7 191 | 25287507@N02-836 49 54 18 192 | 25287507@N02-837 34 17 3 5 8 193 | 29174998@N05-1101 1 2 3 194 | 29465723@N00-1117 32 38 7 195 | 44124450371@N01-2032 34 6 2 49 18 15 44 0 7 196 | 84987970@N00-4211 20 1 9 3 22 7 4 197 | 28527865@N05-1066 3 39 5 0 61 198 | 97972855@N00-4982 14 0 7 199 | 63287673@N00-3062 35 11 0 200 | 52952793@N00-2527 9 14 0 7 4 201 | 52952793@N00-2526 48 20 38 18 22 202 | 48889122860@N01-2289 32 19 33 50 5 0 4 203 | 97972855@N00-4980 15 5 0 7 4 8 204 | 7640968@N02-3668 1 33 38 18 4 205 | 29728054@N00-1119 11 2 5 0 206 | 63287673@N00-3047 22 0 4 207 | 46421193@N07-2126 0 73 7 208 | 44669264@N06-2066 40 4 12 209 | 53836246@N00-2561 20 1 36 2 3 18 5 4 210 | 57841116@N03-2733 19 3 0 211 | 57841116@N03-2735 62 55 40 212 | 82397118@N00-4113 3 0 4 213 | 22670857@N03-614 11 52 0 214 | 7666975@N03-3671 29 15 16 53 0 8 215 | 22670857@N03-611 5 0 4 216 | 94482242@N00-4842 28 26 47 49 217 | 68529605@N00-3368 9 3 14 0 4 218 | 58085609@N00-2826 27 14 31 59 219 | 25976134@N04-902 33 36 3 5 220 | 57841116@N03-2789 17 19 12 221 | 84987970@N00-4209 34 1 19 26 6 36 47 2 49 8 222 | 82698298@N00-4123 1 3 5 223 | 84987970@N00-4202 13 11 2 12 224 | 84987970@N00-4203 18 4 8 225 | 84987970@N00-4206 28 6 36 11 8 226 | 84987970@N00-4204 1 36 11 2 227 | 39916636@N00-1786 3 5 22 228 | 54086610@N00-2575 14 5 25 0 7 229 | 87355829@N00-4315 10 32 33 230 | 87355829@N00-4314 17 33 0 231 | 54086610@N00-2572 9 72 75 4 232 | 63238390@N00-3034 9 0 4 233 | 75286866@N00-3600 45 10 11 234 | 89165847@N00-4440 2 51 69 12 235 | 8249050@N02-4121 49 18 43 236 | 41312242@N00-1906 35 9 24 7 237 | 28067218@N00-1023 1 6 15 238 | 93003266@N04-4726 20 68 28 1 2 3 38 22 239 | 18935424@N00-438 35 32 19 11 24 7 4 240 | 41725486@N08-1914 19 0 61 7 241 | 30265340@N00-1191 35 28 32 6 15 24 5 8 242 | 93987037@N00-4786 19 27 9 18 24 14 5 81 0 7 4 243 | 87801055@N06-4396 9 11 57 24 244 | 77152181@N06-3745 20 10 18 16 245 | 77152181@N06-3747 1 6 15 246 | 77152181@N06-3748 11 76 31 66 7 4 247 | 47334482@N00-2177 27 2 14 5 0 63 4 248 | 87801055@N06-4399 6 49 15 249 | 30265340@N00-1172 35 1 36 11 3 24 22 7 250 | 43929620@N05-2005 2 15 0 251 | 30265340@N00-1170 6 2 15 8 252 | 82698298@N00-4126 28 3 5 22 4 253 | 30265340@N00-1176 1 6 18 254 | 71088526@N00-3468 20 32 19 2 31 5 22 255 | 62744044@N00-3004 1 2 0 4 256 | 63357531@N00-3092 5 22 0 257 | 63357531@N00-3091 1 26 6 5 258 | 62744044@N00-3002 13 3 8 259 | 63357531@N00-3094 1 26 2 260 | 97867688@N00-4922 32 33 5 0 261 | 76225887@N00-3659 14 0 7 4 262 | 97867688@N00-4929 19 69 12 263 | 21225669@N00-562 2 3 61 264 | 43408458@N08-1956 2 38 22 265 | 13956722@N06-290 20 66 4 266 | 78518246@N00-3846 35 34 22 267 | 81697278@N02-4085 19 5 8 268 | 78518246@N00-3844 42 9 11 269 | 9438227@N06-4828 33 3 15 5 0 270 | 55579714@N00-2646 6 2 0 8 271 | 31058815@N00-1231 5 40 25 272 | 38331851@N00-1724 1 15 39 273 | 57841116@N03-2771 32 22 7 274 | 36424619@N00-1562 15 5 79 59 275 | 45904697@N05-2109 36 2 12 276 | 72744226@N00-3506 1 36 2 277 | 72744226@N00-3505 9 30 0 7 278 | 99804259@N00-5070 2 83 8 279 | 28461660@N00-1062 32 9 24 7 4 280 | 28461660@N00-1063 27 36 47 16 4 281 | 25287507@N02-841 28 1 32 36 282 | 44204242@N04-2059 1 5 0 283 | 25369515@N00-856 11 3 7 284 | 75546495@N03-3609 1 49 5 52 0 66 7 4 285 | 75546495@N03-3607 28 1 32 286 | 24413182@N00-765 13 24 14 7 287 | 24413182@N00-766 1 3 38 5 22 0 7 4 288 | 40742367@N04-1895 27 30 59 289 | 40742367@N04-1894 1 17 14 31 43 53 4 8 290 | 44124401641@N01-2030 36 13 5 291 | 63357531@N00-3072 32 54 5 22 0 4 292 | 86589646@N00-4261 1 19 42 29 6 36 3 15 5 25 0 293 | 89743353@N00-4449 17 30 12 294 | 83029300@N00-4130 24 14 25 4 295 | 82256568@N03-4106 19 3 14 5 22 0 4 296 | 82256568@N03-4107 24 14 7 297 | 54352459@N03-2598 1 0 7 298 | 82256568@N03-4108 24 5 7 299 | 31148157@N00-1249 11 24 0 300 | 77967821@N00-3811 3 15 5 4 301 | 74693095@N00-3568 1 37 2 302 | 54352459@N03-2608 29 3 51 303 | 44124479801@N01-2040 3 7 12 304 | 31148157@N00-1240 28 17 19 305 | 57990672@N04-2818 24 5 25 0 7 306 | 31148157@N00-1245 28 17 26 0 307 | 16767930@N05-392 67 23 11 57 24 31 73 66 7 4 308 | 44259454@N00-2063 11 24 66 309 | 97064055@N00-4902 11 24 7 310 | 58196392@N00-2839 15 4 8 311 | 42583704@N00-1940 20 57 3 18 31 0 66 312 | 68892129@N00-3387 11 78 31 5 0 313 | 24998164@N06-819 35 19 9 24 14 7 4 314 | 14136614@N03-313 1 29 6 315 | 14136614@N03-311 1 19 2 5 316 | 24998164@N06-817 24 14 5 0 7 4 317 | 41894175689@N01-1921 2 3 15 5 318 | 41894175689@N01-1923 1 11 54 5 319 | 30628871@N00-1217 32 50 5 0 320 | 89165847@N00-4436 70 1 2 3 38 50 24 14 21 25 0 61 80 7 4 321 | 69041444@N00-3395 35 24 0 322 | 37086457@N00-1677 20 5 0 323 | 84385698@N00-4171 45 17 6 0 7 324 | 94520611@N05-4852 3 5 40 52 7 4 325 | 34948727@N00-1386 34 11 24 0 7 326 | 53840569@N07-2566 29 6 8 327 | 53840569@N07-2567 35 11 7 328 | 29954226@N00-1134 19 13 15 0 329 | 22131083@N08-587 70 21 63 330 | 66172503@N00-3248 35 34 27 11 2 3 18 15 5 22 0 7 4 331 | 77235006@N00-3767 20 3 18 5 44 332 | 58589111@N04-2852 15 52 25 333 | 8473570@N02-4182 37 2 5 334 | 8473570@N02-4183 32 2 78 31 335 | 99804259@N00-5101 2 58 43 336 | 8473570@N02-4185 11 2 39 44 0 337 | 48453097@N00-2255 20 1 15 338 | 76726772@N00-3729 35 24 40 52 7 4 339 | 96256161@N00-4887 45 34 16 53 340 | 49787588@N00-2345 1 3 15 5 22 0 4 8 341 | 72562013@N06-3478 14 5 7 342 | 49503207397@N01-2340 1 3 18 0 343 | 49503207397@N01-2341 48 16 43 344 | 25963075@N04-901 35 11 18 345 | 24122821@N05-749 15 44 0 346 | 87356726@N00-4328 38 0 63 347 | 31124964@N00-1233 30 3 49 53 348 | 52618473@N02-2495 15 14 7 349 | 32446767@N03-1286 24 14 4 350 | 37103729@N02-1679 36 2 3 5 22 351 | 86589646@N00-4257 1 3 0 352 | 8485766@N07-4198 24 31 66 7 353 | 79235216@N00-3891 38 18 44 25 354 | 79235216@N00-3890 28 1 6 36 30 2 54 51 15 50 355 | 66883888@N00-3274 27 6 31 356 | 24413182@N00-763 1 9 30 3 14 5 7 357 | 36330826404@N01-1521 28 32 19 2 5 44 22 0 4 358 | 36330826404@N01-1522 20 1 30 2 40 25 7 359 | 11674539@N06-166 1 24 7 360 | 93384183@N00-4749 35 28 32 29 2 361 | 22670857@N03-607 34 2 15 362 | 91821745@N00-4682 5 0 4 363 | 40082311@N00-1823 17 3 18 364 | 40082311@N00-1820 28 3 18 365 | 40082311@N00-1826 3 38 18 366 | 40082311@N00-1827 20 17 3 38 18 367 | 40575018@N07-1854 32 5 22 368 | 96155869@N00-4885 17 19 33 51 12 369 | 80021819@N00-4049 29 36 2 370 | 51986774@N00-2455 3 50 22 371 | 82955378@N05-4129 29 2 3 15 5 44 56 372 | 98999190@N00-5047 65 32 19 22 7 4 373 | 68529605@N00-3370 28 33 12 374 | 21929094@N02-582 15 16 0 375 | 8755651@N05-4372 46 6 4 8 376 | 92496717@N00-4698 35 43 4 377 | 57841116@N03-2808 38 18 44 378 | 82155102@N07-4105 32 33 4 379 | 63357531@N00-3084 1 3 0 380 | 94327650@N03-4790 42 15 8 381 | 69539023@N00-3398 48 32 26 36 5 0 382 | 42656687@N00-1941 1 2 50 383 | 49975776@N07-2352 20 1 6 57 3 14 5 25 0 63 4 384 | 97867688@N00-4915 26 27 9 2 15 79 0 69 12 385 | 37912374670@N01-1703 18 5 22 386 | 97972855@N00-4983 19 3 18 16 44 12 387 | 10195518@N02-14 62 55 21 66 61 7 388 | 66647015@N06-3271 3 76 63 7 389 | 26670541@N07-931 28 42 26 6 2 8 390 | 79925938@N00-3977 34 26 9 391 | 9438227@N06-4832 17 29 30 16 392 | 57841116@N03-2747 50 44 56 393 | 24453935@N00-771 24 0 66 7 394 | 44925192@N00-2079 28 1 32 26 9 11 3 14 5 22 0 395 | 44925192@N00-2078 13 64 7 396 | 19289077@N00-440 34 19 33 49 15 22 7 4 8 397 | 31043650@N04-1230 33 2 21 398 | 97304820@N00-4906 2 5 22 0 399 | 97304820@N00-4907 34 1 42 29 6 36 3 15 5 22 0 8 12 400 | 33195950@N00-1318 35 42 29 27 3 15 24 22 0 73 7 8 401 | 63287673@N00-3042 35 61 7 402 | 22853276@N05-681 3 38 61 403 | 22853276@N05-680 25 0 59 7 404 | 94482242@N00-4849 20 6 8 405 | 22853276@N05-686 52 0 7 406 | 22853276@N05-685 17 9 30 14 25 0 4 12 407 | 49462309@N04-2323 31 25 7 408 | 83332749@N00-4141 35 17 9 11 44 0 7 409 | 92496717@N00-4702 77 21 7 4 410 | 56935359@N00-2680 42 26 4 411 | 36656940@N00-1583 26 2 3 412 | 48737213@N00-2282 64 11 24 413 | 13233236@N04-228 9 2 0 414 | 13233236@N04-224 32 19 18 415 | 13233236@N04-225 1 32 33 26 36 416 | 68183366@N05-3354 34 47 8 417 | 39297387@N00-1769 9 2 14 0 418 | 43929620@N05-1990 1 36 2 8 419 | 10559879@N00-78 42 6 15 420 | 60422641@N04-2905 38 15 61 421 | 68892129@N00-3391 11 5 7 422 | 24998164@N06-820 28 1 32 19 33 26 36 0 4 423 | 30628871@N00-1208 19 5 0 4 424 | 95464641@N00-4868 1 17 19 425 | 31404528@N02-1265 2 39 0 4 426 | 56140378@N07-2672 20 38 15 5 427 | 56140378@N07-2671 6 49 8 428 | 22852813@N04-676 9 30 50 16 53 429 | 12505664@N00-206 35 0 7 430 | 36521966482@N01-1563 30 0 4 431 | 23809369@N08-724 20 27 2 18 16 44 432 | 23809369@N08-725 33 26 36 31 433 | 23809369@N08-723 32 19 33 11 3 5 0 7 434 | 36521966482@N01-1565 13 14 7 435 | 57608438@N08-2690 35 3 7 4 436 | 57608438@N08-2691 9 30 24 5 25 437 | 99804259@N00-5064 35 38 14 31 438 | 49992082@N08-2354 1 6 3 15 439 | 77152181@N06-3762 17 5 0 7 4 440 | 77152181@N06-3761 32 17 19 33 11 3 24 5 0 7 441 | 84987970@N00-4199 65 13 38 18 61 442 | 61579728@N05-2956 68 64 11 3 12 443 | -------------------------------------------------------------------------------- /Trip/data/Edin_set.dat: -------------------------------------------------------------------------------- 1 | 7512717@N06-3700 0 10 20 2 | 31397036@N00-1656 4 8 1 3 | 96423179@N00-4823 6 5 1 4 | 16048448@N00-449 4 3 0 6 5 1 5 | 67975030@N00-3214 4 11 3 0 15 13 6 14 7 5 16 19 1 6 | 60611127@N00-2978 4 2 6 14 5 1 7 | 60611127@N00-2979 17 11 3 0 6 1 8 | 34427470616@N01-1839 2 0 10 1 9 | 88185060@N00-4358 11 13 6 16 19 10 | 26436271@N00-1294 3 6 7 11 | 28543475@N08-1495 2 8 9 12 | 37574483@N07-2072 2 7 5 13 | 88185060@N00-4352 0 6 16 14 | 88185060@N00-4353 0 6 5 15 | 88185060@N00-4355 2 6 8 7 16 | 88185060@N00-4356 2 7 5 17 | 88185060@N00-4357 11 0 7 19 18 | 11154928@N04-150 0 6 8 5 1 19 | 39745335@N06-2137 0 7 16 1 20 | 72429059@N00-3372 2 0 6 8 21 | 47616704@N00-2495 3 0 6 14 7 1 22 | 47616704@N00-2496 4 0 1 23 | 56617401@N00-2845 3 0 6 14 5 1 24 | 60006733@N05-2953 2 3 9 25 | 56617401@N00-2840 4 3 5 26 | 54022106@N00-2749 11 19 9 27 | 12771303@N00-240 11 13 10 19 28 | 29057345@N04-1522 2 0 21 29 | 65958138@N00-3138 4 0 6 1 30 | 90113037@N00-4475 2 6 8 31 | 60206297@N00-2969 23 0 8 32 | 60206297@N00-2968 4 20 1 33 | 32812410@N06-1774 4 2 0 6 7 5 1 34 | 19473027@N00-758 3 6 16 35 | 51367743@N00-2651 4 17 3 13 14 36 | 27948364@N00-1427 2 0 6 14 1 37 | 84533614@N00-4210 15 12 16 38 | 59446027@N00-2942 4 0 13 8 1 39 | 27948364@N00-1428 4 2 6 40 | 88185060@N00-4369 4 2 0 13 41 | 81795065@N00-4120 15 13 16 42 | 78461351@N00-3968 3 0 7 5 1 43 | 42629370@N00-2272 17 11 15 13 16 44 | 42629370@N00-2273 2 0 1 45 | 42629370@N00-2270 3 0 5 1 46 | 42629370@N00-2271 4 2 1 47 | 92742787@N04-4702 2 0 15 1 48 | 79335735@N06-4020 3 6 14 49 | 8043663@N04-4053 4 13 6 7 50 | 25984556@N07-1271 6 8 1 51 | 25984556@N07-1270 4 3 0 10 6 14 8 1 52 | 44124427374@N01-2379 11 0 6 7 21 1 53 | 10354154@N05-46 14 7 1 54 | 20587560@N00-924 2 10 5 55 | 52439376@N00-2683 11 3 8 5 1 56 | 7150832@N06-3324 17 15 13 57 | 33278177@N00-1799 4 11 7 58 | 48762356@N00-2528 3 0 1 59 | 30963564@N00-1607 17 15 14 60 | 98833136@N00-4989 0 15 13 61 | 20799701@N00-931 4 17 2 26 15 13 8 7 5 18 16 1 62 | 35034357529@N01-1884 0 6 14 63 | 20483509@N00-826 11 3 0 6 64 | 36539950@N00-1989 15 10 16 65 | 63079834@N00-3076 4 7 1 66 | 30671962@N05-1575 0 8 18 67 | 11037560@N04-142 2 0 9 68 | 28355390@N03-1452 0 13 6 14 16 1 69 | 18966576@N00-542 3 0 14 7 1 70 | 10091295@N02-4 15 8 9 71 | 42027916@N00-2230 6 5 21 72 | 42027916@N00-2231 4 7 1 73 | 37493306@N00-2065 2 3 0 14 8 9 1 74 | 42027916@N00-2233 6 9 1 75 | 71812696@N00-3344 3 12 5 76 | 28638538@N00-1508 4 3 0 1 77 | 16497759@N07-481 11 0 21 19 78 | 10335112@N05-41 4 17 3 6 1 79 | 10335112@N05-42 15 13 16 80 | 23363966@N02-1097 17 11 3 1 81 | 23363966@N02-1094 14 7 1 82 | 23363966@N02-1095 17 11 6 83 | 40139809@N00-2151 17 0 16 1 84 | 90514086@N00-4491 4 0 6 14 21 1 85 | 28543475@N08-1507 10 12 1 86 | 23351536@N07-1075 2 0 12 8 1 87 | 23351536@N07-1074 3 12 9 88 | 23351536@N07-1077 2 0 8 9 1 89 | 77321230@N00-3882 4 0 10 1 90 | 77036619@N00-3811 2 3 0 6 8 5 1 91 | 21712240@N00-969 4 11 12 5 19 9 1 92 | 7512717@N06-3601 0 9 1 93 | 12982579@N00-266 3 0 14 5 16 1 94 | 7512717@N06-3609 3 0 8 95 | 62136333@N02-3056 4 17 11 3 15 96 | 23351536@N07-1080 2 3 15 5 16 9 97 | 40230716@N03-2175 4 3 0 98 | 23351536@N07-1083 3 15 13 12 14 16 99 | 67008765@N00-3181 17 11 19 100 | 54458860@N00-2791 17 11 0 101 | 67008765@N00-3183 4 3 0 6 14 5 1 102 | 23351536@N07-1087 4 2 8 103 | 50802125@N00-2618 4 0 6 104 | 7359973@N08-3417 2 0 6 7 9 1 105 | 32481757@N04-1761 4 17 11 0 8 16 1 106 | 80997939@N00-4083 4 0 8 107 | 14360576@N06-365 4 2 3 0 15 13 10 6 8 7 16 1 108 | 14360576@N06-366 23 18 22 109 | 77704116@N00-3896 3 6 7 5 1 110 | 14360576@N06-368 0 20 19 111 | 25396215@N00-1219 3 0 6 14 7 1 112 | 88185060@N00-4362 0 6 1 113 | 53836246@N00-2718 17 11 3 15 114 | 18276635@N00-528 4 0 8 1 115 | 18276635@N00-529 13 6 7 116 | 93511097@N00-4722 0 8 1 117 | 19473027@N00-652 14 5 1 118 | 29978367@N00-1542 4 17 2 8 1 119 | 22607526@N00-1027 2 0 8 1 120 | 22607526@N00-1029 4 2 1 121 | 95974557@N08-4812 3 14 5 122 | 24401095@N00-1160 4 2 3 12 123 | 36749913@N00-1996 10 6 7 124 | 16190595@N05-464 0 8 1 125 | 60006733@N05-2950 3 15 13 14 16 126 | 64119933@N00-3113 11 15 13 6 7 127 | 50551683@N00-2608 4 2 0 6 128 | 11261325@N03-164 4 17 11 0 5 1 129 | 32263295@N00-1717 3 12 6 9 130 | 32263295@N00-1714 6 5 9 131 | 23480782@N05-1103 4 2 0 132 | 42348675@N00-2254 15 13 9 133 | 19909714@N00-807 4 2 1 134 | 20545573@N00-913 6 14 7 135 | 9047223@N07-4490 26 7 1 136 | 24697990@N00-1181 11 0 19 1 137 | 28536568@N00-1490 17 11 5 138 | 19909714@N00-808 3 6 5 139 | 27681741@N06-1351 0 15 13 16 140 | 20483509@N00-867 0 6 14 141 | 32111003@N00-1703 3 6 14 1 142 | 43619118@N04-2335 4 3 6 143 | 7790703@N02-3920 4 17 11 0 144 | 73124528@N00-3394 4 17 11 145 | 43619118@N04-2339 4 3 0 14 1 146 | 80052783@N00-4046 4 17 11 0 5 16 1 147 | 7790703@N02-3928 0 6 1 148 | 23108377@N07-1051 11 3 13 6 8 24 149 | 46980778@N07-2463 4 3 1 150 | 36424619@N00-1970 2 11 23 16 151 | 99185699@N00-5001 3 14 1 152 | 32693718@N07-1765 2 7 5 153 | 31450835@N05-1669 2 3 0 14 8 9 1 154 | 21254955@N04-945 17 0 1 155 | 21254955@N04-944 4 0 1 156 | 27462949@N00-1338 11 18 21 157 | 27462949@N00-1339 4 2 0 15 5 158 | 61897811@N00-3038 17 11 3 0 12 1 159 | 7880512@N06-3985 23 8 19 160 | 61369090@N07-2998 0 8 5 161 | 38943965@N05-2109 4 3 0 21 1 162 | 30129139@N04-1560 2 13 16 9 163 | 24183489@N00-1146 2 15 13 164 | 8684258@N02-4295 4 2 0 13 16 1 165 | 7512717@N06-3654 4 0 10 8 1 166 | 7512717@N06-3657 3 6 14 1 167 | 7512717@N06-3656 3 6 9 168 | 8308954@N06-4157 4 0 10 169 | 11037560@N04-101 11 15 5 170 | 43405950@N07-2319 2 3 10 14 7 1 171 | 43405950@N07-2318 2 0 1 172 | 9813542@N08-4955 4 3 0 6 14 5 173 | 25389244@N00-1216 4 17 11 0 5 1 174 | 62136333@N02-3049 4 3 1 175 | 44124372821@N01-2370 0 6 16 176 | 44124372821@N01-2371 4 2 0 177 | 62136333@N02-3046 4 3 15 7 1 178 | 39927207@N00-2138 4 0 8 5 179 | 25148740@N03-1203 4 17 11 0 15 13 1 180 | 63679518@N00-3099 4 7 5 181 | 28404646@N00-1483 11 5 16 19 182 | 25396215@N00-1220 3 12 8 5 9 183 | 84533614@N00-4213 23 3 0 6 5 184 | 16048448@N00-450 23 15 13 7 185 | 84533614@N00-4212 2 9 1 186 | 49998283@N00-2578 4 0 5 187 | 15462727@N07-438 4 2 3 0 6 7 5 188 | 43126182@N00-2307 2 0 5 189 | 19153501@N00-543 17 11 13 16 190 | 46237053@N00-2436 15 5 1 191 | 34961066@N00-1876 11 0 1 192 | 86752930@N00-4292 3 0 6 193 | 27195496@N00-1323 11 3 0 194 | 34792843@N04-1875 0 8 7 1 195 | 51035737977@N01-2639 11 0 1 196 | 58554451@N00-2893 17 11 0 1 197 | 58034287@N00-2879 3 0 6 14 5 1 198 | 27718315@N02-1367 0 5 1 199 | 24616128@N00-1173 4 0 6 5 1 200 | 36749913@N00-2003 4 2 10 1 201 | 8349002@N04-4168 4 17 11 3 6 7 5 1 202 | 54022106@N00-2752 2 3 1 203 | 36749913@N00-2007 15 13 16 204 | 84035351@N00-4180 4 0 7 1 205 | 25034321@N05-1201 3 10 6 7 206 | 97768382@N00-4951 4 17 11 3 0 16 1 207 | 21916524@N07-981 17 11 8 208 | 8520591@N03-4227 0 5 21 9 1 209 | 8057454@N06-4066 17 11 13 5 16 210 | 37996646802@N01-2091 4 23 0 211 | 37996646802@N01-2092 3 6 14 212 | 26226560@N02-1282 2 0 6 8 213 | 26226560@N02-1281 0 8 1 214 | 50833784@N00-2625 4 17 2 11 3 7 5 16 19 1 215 | 51367743@N00-2647 17 11 0 15 13 6 8 216 | 57192952@N00-2850 4 3 14 1 217 | 8997662@N06-4468 15 13 6 14 5 16 1 218 | 8997662@N06-4469 4 5 1 219 | 38783516@N00-2108 2 3 0 220 | 51367743@N00-2648 4 14 1 221 | 47495910@N05-2491 3 0 14 8 1 222 | 47495910@N05-2493 4 2 0 15 13 10 8 7 9 223 | 47495910@N05-2492 4 0 1 224 | 25302607@N00-1208 4 3 0 15 13 225 | 71668585@N00-3329 4 0 1 226 | 19473027@N00-581 0 10 7 5 227 | 9352211@N06-4727 4 3 0 10 7 1 228 | 14146962@N07-349 4 3 1 229 | 55826454@N00-2820 4 3 0 15 10 6 5 16 9 1 230 | 12495774@N02-219 2 11 23 0 1 231 | 99962592@N00-5027 4 3 6 232 | 71401076@N00-3308 4 10 20 14 7 233 | 9352211@N06-4728 0 6 1 234 | 20483509@N00-907 14 7 9 235 | 77047514@N00-3862 2 0 6 14 236 | 66227636@N04-3140 3 0 1 237 | 77047514@N00-3825 2 12 20 6 7 5 16 238 | 90001203@N00-4474 0 10 8 1 239 | 90001203@N00-4473 4 0 1 240 | 31803571@N02-1683 17 16 1 241 | 51035639266@N01-2633 4 10 6 5 242 | 27817416@N02-1392 0 7 5 243 | 27817416@N02-1394 0 6 1 244 | 20483509@N00-838 11 6 14 5 1 245 | 20483509@N00-837 17 11 19 246 | 42072165@N08-2237 4 0 1 247 | 37029314@N00-2054 4 17 11 0 10 6 14 7 1 248 | 37029314@N00-2055 4 15 13 8 16 249 | 85217508@N00-4229 4 0 6 5 1 250 | 48277923@N00-2508 4 2 0 1 251 | 8110030@N05-4087 17 23 0 15 13 6 14 5 16 1 252 | 33613749@N00-1811 4 0 6 1 253 | 7512717@N06-3681 0 6 21 254 | 30099537@N02-1555 4 3 1 255 | 7512717@N06-3684 0 21 9 256 | 88185060@N00-4346 0 10 6 257 | 9351020@N06-4719 0 5 1 258 | 9351020@N06-4718 4 0 6 259 | 71088526@N00-3303 2 6 14 1 260 | 61663261@N00-3032 15 13 14 5 261 | 71088526@N00-3301 17 23 3 0 6 14 7 5 262 | 7512717@N06-3667 0 6 14 9 263 | 34427470616@N01-1862 2 0 9 1 264 | 7512717@N06-3505 0 6 7 1 265 | 32054787@N00-1700 17 0 8 266 | 7512717@N06-3616 3 6 21 267 | 7512717@N06-3611 0 10 6 1 268 | 34427470616@N01-1850 0 10 12 18 269 | 7512717@N06-3597 0 6 14 270 | 61635869@N00-3025 4 3 0 14 5 1 271 | 28103568@N00-1432 3 0 13 14 7 5 16 272 | 26835091@N04-1308 12 6 8 9 273 | 68493542@N00-3225 3 0 6 5 22 274 | 15786211@N00-446 4 2 14 275 | 77456049@N00-3895 17 3 0 6 14 5 19 1 276 | 20483509@N00-882 2 0 13 277 | 89187910@N00-4421 2 0 15 13 278 | 20483509@N00-881 11 0 19 279 | 50234824@N07-2587 4 0 7 1 280 | 83308875@N00-4165 2 0 15 13 8 5 16 281 | 83308875@N00-4164 3 6 14 5 1 282 | 65393953@N03-3136 2 0 14 8 283 | 14360576@N06-371 0 6 5 1 284 | 14360576@N06-370 0 15 10 20 5 21 16 1 285 | 86533211@N00-4287 11 8 7 5 19 286 | 14915235@N00-411 2 0 10 287 | 96241281@N00-4818 4 2 3 13 1 288 | 23062920@N00-1046 2 3 0 12 1 289 | 23062920@N00-1045 3 6 7 21 290 | 63238112@N02-3085 0 8 1 291 | 9588476@N02-4799 3 6 7 292 | 29812295@N07-1537 4 2 8 1 293 | 7512717@N06-3701 3 7 5 294 | 12930951@N07-265 3 14 5 295 | 29812295@N07-1538 2 0 13 16 296 | 19473027@N00-705 3 6 7 297 | 92237387@N00-4593 17 23 19 298 | 19473027@N00-702 6 7 1 299 | 31612046@N04-1680 4 11 0 8 300 | 45972156@N04-2428 11 23 3 13 7 19 301 | 45972156@N04-2427 4 2 0 6 1 302 | 22607526@N00-1030 3 6 7 303 | 9156485@N02-4568 3 0 6 14 1 304 | 80021819@N00-4043 4 17 11 0 14 8 305 | 22415346@N06-1017 0 15 13 1 306 | 80021819@N00-4044 15 13 16 307 | 120694055@N06-203 4 2 1 308 | 19473027@N00-682 0 5 16 309 | 19473027@N00-680 3 0 5 310 | 60107315@N00-2963 0 6 7 311 | 52706710@N03-2688 4 2 3 6 312 | 9159505@N02-4574 4 0 1 313 | 48122140@N00-2503 4 2 3 0 15 8 5 1 314 | 7666975@N03-3796 4 2 1 315 | 63458311@N00-3096 4 0 1 316 | 48321512@N06-2514 3 6 7 5 21 317 | 22087304@N07-986 15 13 16 318 | 9588476@N02-4805 7 21 24 319 | 94735786@N00-4744 4 3 6 5 1 320 | 94735786@N00-4745 4 17 2 11 3 10 8 5 9 1 321 | 94735786@N00-4746 4 3 6 5 322 | 21283177@N00-948 17 15 14 16 1 323 | 94735786@N00-4748 4 17 2 3 0 15 6 8 5 21 24 1 324 | 40725466@N03-2185 4 3 6 5 1 325 | 71088526@N00-3302 4 3 1 326 | 43222683@N00-2312 11 0 7 1 327 | 28536568@N00-1487 4 0 5 1 328 | 7633518@N08-3789 4 0 1 329 | 37185577@N00-2060 4 0 7 1 330 | 86765090@N00-4293 11 13 16 19 331 | 27681741@N06-1349 2 7 5 332 | 10655856@N04-82 3 12 14 5 1 333 | 10655856@N04-81 0 10 21 1 334 | 71057941@N00-3288 3 6 8 335 | 7790703@N02-3930 2 10 5 336 | 97534175@N00-4922 4 2 0 1 337 | 97534175@N00-4923 2 0 8 1 338 | 44232237@N03-2388 3 0 6 14 8 339 | 86401769@N00-4277 4 17 6 1 340 | 90585585@N00-4496 0 5 9 341 | 62308216@N00-3059 4 0 1 342 | 90585585@N00-4493 4 3 19 343 | 11613952@N05-185 8 7 1 344 | 50542255@N00-2605 3 10 6 345 | 73173768@N00-3399 4 20 6 14 1 346 | 7512717@N06-3625 2 20 8 9 347 | 7512717@N06-3623 0 8 1 348 | 7512717@N06-3719 4 0 10 349 | 21492331@N03-956 3 7 5 9 350 | 17643132@N00-513 2 0 13 12 5 9 1 351 | 10332960@N03-39 2 0 8 352 | 17643132@N00-514 4 11 0 5 1 353 | 10340081@N02-44 4 2 3 354 | 19663529@N00-800 4 0 5 16 355 | 19663529@N00-802 11 19 1 356 | 58433801@N07-2891 11 6 5 357 | 60444977@N00-2972 3 0 1 358 | 85603833@N00-4239 15 16 9 359 | 73478860@N00-3413 0 5 9 1 360 | 75951171@N02-3776 7 9 1 361 | 58695432@N00-2907 15 13 5 362 | 7512717@N06-3530 0 14 7 363 | 7512717@N06-3651 3 10 6 7 1 364 | 7512717@N06-3650 2 11 6 14 365 | 29624180@N04-1535 11 3 19 366 | 29624180@N04-1533 4 3 0 1 367 | 52683275@N00-2684 4 2 0 368 | 87658636@N00-4316 0 12 1 369 | 27828336@N00-1423 14 7 1 370 | 9878315@N04-4969 0 18 19 371 | 88185060@N00-4335 4 0 20 372 | 87658636@N00-4318 2 0 1 373 | 85217508@N00-4232 13 5 16 374 | 71668585@N00-3340 3 5 19 375 | 9351020@N06-4721 4 5 1 376 | 8326485@N06-4160 4 2 11 0 16 1 377 | 40055757@N00-2144 11 3 5 378 | 71909564@N00-3364 15 13 1 379 | 26680618@N02-1302 0 6 14 9 380 | 48355243@N00-2515 0 6 14 381 | 23518714@N00-1107 4 3 0 6 5 19 1 382 | 12061846@N00-202 4 3 0 15 6 5 1 383 | 43002463@N00-2286 2 3 5 384 | 71668585@N00-3332 6 14 7 385 | 71668585@N00-3335 5 18 21 386 | 7929654@N05-4012 17 11 3 0 7 387 | 7929654@N05-4011 2 11 23 15 13 16 388 | 8997662@N06-4470 17 11 3 6 7 5 19 389 | 37804979@N00-2078 3 5 1 390 | 11037560@N04-98 4 11 0 391 | 11037560@N04-99 11 15 10 6 14 5 392 | 19473027@N00-672 3 0 14 1 393 | 64876844@N00-3122 0 13 8 1 394 | 9159505@N02-4575 17 11 3 15 13 5 16 19 395 | 59271446@N00-2924 6 21 1 396 | 12495774@N02-222 4 2 0 6 7 397 | 48355243@N00-2516 4 2 0 398 | 11881702@N00-194 6 14 8 399 | 11881702@N00-193 4 2 0 8 1 400 | 30409117@N07-1567 2 5 9 401 | 24803078@N00-1186 3 0 6 14 5 1 402 | 24645033@N00-1180 3 13 5 16 403 | 14869313@N00-398 17 11 6 404 | 80649505@N08-4071 4 23 19 405 | 44124427374@N01-2380 17 11 0 6 406 | 58598613@N00-2896 17 11 0 407 | 66434265@N00-3145 4 11 15 13 8 1 408 | 66434265@N00-3146 23 10 8 19 409 | 66434265@N00-3147 2 0 8 410 | 26265986@N00-1287 4 14 7 18 24 411 | 26265986@N00-1286 4 2 7 412 | 31363435@N07-1652 2 0 15 13 8 5 16 1 413 | 31363435@N07-1651 2 0 8 7 5 1 414 | 71445279@N00-3323 2 0 12 6 14 7 9 1 415 | 7310714@N06-3388 11 0 8 1 416 | 7310714@N06-3389 3 0 9 417 | 64332066@N00-3115 4 17 2 0 13 8 5 1 418 | 70384696@N00-3257 4 2 1 419 | 62136333@N02-3047 4 11 3 15 1 420 | 42907325@N00-2280 2 0 16 421 | 83511002@N00-4169 23 13 6 16 422 | 38115734@N00-2095 23 6 7 423 | 7512717@N06-3699 3 6 5 424 | 38115734@N00-2093 4 11 19 425 | 7512717@N06-3696 2 0 8 1 426 | 32329639@N00-1747 11 3 0 6 7 1 427 | 91357983@N05-4557 3 13 10 428 | 41692261@N00-2209 4 0 1 429 | 13518023@N03-274 4 19 1 430 | 13518023@N03-275 4 2 3 14 5 9 1 431 | 44756385@N00-2401 2 11 3 0 1 432 | 44756385@N00-2402 2 0 8 9 433 | 105307156@N06-74 4 17 23 434 | 96241281@N00-4817 0 13 8 435 | 18402635@N00-540 4 2 0 436 | 51272891@N00-2645 2 0 5 16 1 437 | 7512717@N06-3677 0 20 6 7 438 | 8997662@N06-4471 2 11 3 14 7 5 19 1 439 | 97595808@N00-4941 2 12 8 9 440 | 66913692@N00-3155 3 0 16 1 441 | 66913692@N00-3154 4 3 12 442 | 57258627@N00-2853 3 0 14 8 5 443 | 7512717@N06-3572 2 12 9 444 | 77036619@N00-3807 4 17 2 11 3 0 14 8 5 1 445 | 8095506@N07-4080 4 3 0 10 6 7 1 446 | 13016459@N08-267 8 5 16 447 | 7512717@N06-3586 0 6 14 8 448 | 60611127@N00-2983 4 2 0 15 13 449 | 60611127@N00-2981 4 3 0 13 8 1 450 | 60611127@N00-2980 4 3 0 1 451 | 32054787@N00-1692 11 0 8 452 | 42583704@N00-2268 4 3 6 5 453 | 63238112@N02-3080 10 6 7 5 454 | 68121184@N00-3216 0 14 1 455 | 42583704@N00-2267 4 3 14 7 1 456 | 42583704@N00-2266 17 2 11 23 8 5 19 457 | 62744044@N00-3070 3 0 5 1 458 | 97595808@N00-4937 11 15 13 19 459 | 42509361@N05-2262 4 6 5 1 460 | 35703376@N00-1953 4 17 1 461 | 53921762@N00-2723 0 6 7 462 | 77351046@N00-3889 3 0 13 14 463 | 53921762@N00-2727 3 5 1 464 | 47905916@N00-2498 4 17 0 15 8 16 1 465 | 8389569@N08-4176 0 15 5 466 | 42479008@N02-2261 17 11 15 13 16 467 | 41894148532@N01-2217 2 6 14 1 468 | 22583729@N06-1022 0 5 9 469 | 19473027@N00-739 17 2 0 19 470 | 21712240@N00-970 2 15 13 5 16 19 471 | 19473027@N00-701 3 6 5 9 472 | 8389569@N08-4173 3 0 15 8 1 473 | 14057878@N00-311 2 0 1 474 | 14057878@N00-317 0 8 9 1 475 | 14057878@N00-316 3 12 6 7 476 | 19473027@N00-734 11 0 1 477 | 54549576@N00-2795 2 11 19 478 | 75136617@N00-3734 0 6 7 479 | 60006733@N05-2947 4 0 7 480 | 65326158@N00-3135 4 17 0 1 481 | 31191687@N06-1639 3 6 5 482 | 60006733@N05-2949 15 13 5 16 483 | 8590701@N06-4242 4 11 3 0 5 1 484 | 19473027@N00-691 10 20 21 9 485 | 19473027@N00-690 12 8 5 9 486 | 84989911@N00-4220 4 2 11 14 8 9 1 487 | 19473027@N00-695 6 7 1 488 | 19473027@N00-694 3 14 5 489 | 23837403@N03-1122 4 17 15 13 16 490 | 23837403@N03-1121 4 17 13 16 491 | 23837403@N03-1120 6 7 5 492 | 30985799@N00-1627 11 3 19 493 | 30985799@N00-1626 4 2 14 8 1 494 | 7512717@N06-3693 4 0 10 20 8 1 495 | 94869158@N00-4750 3 6 1 496 | 24645033@N00-1178 3 5 21 497 | 24645033@N00-1179 15 13 5 498 | 45972156@N04-2429 0 5 1 499 | 46952961@N00-2461 3 15 13 5 16 500 | 81719291@N00-4117 15 13 16 501 | 8389569@N08-4175 4 17 11 3 15 13 14 7 1 502 | 8389569@N08-4174 4 17 11 3 0 15 13 6 14 5 16 9 1 503 | 79295762@N00-4010 0 5 1 504 | 7512717@N06-3665 3 6 14 5 505 | 7790703@N02-3948 3 6 1 506 | 68838281@N00-3233 3 14 1 507 | 7958312@N08-4031 4 3 0 6 1 508 | 72429059@N00-3371 11 0 1 509 | 96721948@N00-4858 4 11 0 15 13 8 1 510 | 80246494@N00-4051 4 17 11 16 1 511 | 10486919@N00-67 0 6 9 512 | 36101699310@N01-1966 11 3 19 1 513 | 36101699310@N01-1965 3 0 15 6 5 514 | 21116747@N02-937 2 11 9 1 515 | 97206958@N00-4894 4 5 16 516 | 9788232@N03-4952 4 3 5 517 | 22890158@N08-1042 4 17 11 0 518 | 75286866@N00-3758 4 2 3 13 8 519 | 8970099@N04-4456 2 0 8 1 520 | 44036847@N05-2348 3 0 14 1 521 | 8970099@N04-4455 2 0 8 522 | 8286908@N02-4155 4 3 5 523 | 54549576@N00-2794 4 17 11 15 7 16 524 | 27594569@N04-1345 0 8 9 1 525 | 75181076@N00-3748 11 3 13 14 5 16 1 526 | 28404646@N00-1473 2 13 5 9 527 | 10159442@N00-13 4 11 3 0 15 13 6 7 5 16 19 1 528 | 99646216@N00-5017 11 3 6 529 | 99646216@N00-5016 17 11 19 530 | 7512717@N06-3711 3 15 13 6 21 16 531 | 7512717@N06-3709 3 6 14 532 | 11037560@N04-127 6 5 1 533 | 14711846@N00-387 4 17 11 3 0 8 5 1 534 | 7512717@N06-3707 0 14 8 1 535 | 52015062@N00-2673 4 3 0 6 5 1 536 | 73422480@N00-3411 2 12 7 537 | 11037560@N04-122 0 6 8 538 | 84276425@N00-4196 2 11 0 15 7 1 539 | 84276425@N00-4197 2 0 6 8 7 21 9 1 540 | 30420228@N05-1568 4 2 0 541 | 33925347@N05-1821 0 10 6 5 542 | 34792843@N04-1874 4 6 7 543 | 50509003@N00-2602 4 3 0 6 7 1 544 | 50509003@N00-2600 4 3 0 7 545 | 81311832@N00-4095 4 5 1 546 | 77036619@N00-3810 3 0 8 5 1 547 | 7354829@N03-3414 4 0 1 548 | 7354829@N03-3416 4 0 14 8 7 5 21 16 1 549 | 14711846@N00-386 15 13 6 550 | 29172291@N00-1526 4 17 2 11 3 0 15 13 1 551 | 60206297@N00-2967 0 9 1 552 | 78361112@N00-3966 17 0 1 553 | 73788492@N00-3423 4 17 11 23 0 1 554 | 44925192@N00-2404 3 0 6 14 555 | 44925192@N00-2405 17 11 0 556 | 7512717@N06-3604 2 12 9 557 | 73788492@N00-3425 17 11 3 0 6 8 5 1 558 | 73788492@N00-3424 3 15 13 6 7 5 21 559 | 49503132817@N01-2572 0 9 1 560 | 7512717@N06-3644 0 6 8 561 | 13996991@N07-284 4 2 0 1 562 | 13996991@N07-282 15 6 5 16 563 | 50794252@N04-2617 4 6 1 564 | 26052518@N05-1273 0 6 1 565 | 62579590@N00-3064 4 2 19 566 | 60107315@N00-2962 2 11 0 567 | 85318305@N00-4235 4 2 0 12 14 8 5 1 568 | 85318305@N00-4236 2 0 10 12 8 5 9 569 | 92237387@N00-4592 3 0 6 7 5 570 | 29797746@N08-1536 4 17 11 3 0 15 13 6 14 5 16 1 571 | 54022106@N00-2757 2 8 5 572 | 67433276@N05-3199 2 14 1 573 | 10287726@N02-32 17 2 3 0 8 1 574 | 10287726@N02-33 4 2 8 7 1 575 | 23873159@N08-1128 2 8 5 1 576 | 8057454@N06-4065 2 0 1 577 | 10287726@N02-34 13 8 16 578 | 25561968@N00-1226 17 13 16 579 | 53836246@N00-2722 17 11 15 13 580 | 53836246@N00-2720 4 2 0 1 581 | 53836246@N00-2721 0 15 13 14 5 582 | 17505188@N06-512 3 0 16 583 | 27105072@N00-1320 2 9 1 584 | 84035351@N00-4181 0 15 8 585 | 84989911@N00-4219 4 17 11 0 8 7 5 16 1 586 | 19473027@N00-644 0 6 14 587 | 67008765@N00-3180 0 8 1 588 | 81397230@N00-4108 2 0 6 1 589 | 8970099@N04-4457 11 23 0 15 13 16 590 | 8328114@N02-4161 4 3 14 1 591 | 9588476@N02-4804 4 11 0 10 1 592 | 8328114@N02-4162 3 14 1 593 | 30713600@N00-1577 4 2 6 14 594 | 24401095@N00-1158 4 0 6 7 1 595 | 77047514@N00-3840 2 10 12 596 | 77047514@N00-3842 2 11 3 6 16 19 597 | 8503026@N04-4221 4 0 6 1 598 | 27017895@N02-1317 2 8 1 599 | 24931020@N02-1193 4 11 15 13 14 8 5 16 1 600 | 42348675@N00-2249 0 6 8 24 601 | 17424601@N00-507 3 6 7 602 | 42348675@N00-2246 2 0 6 7 1 603 | 77047514@N00-3839 2 10 8 1 604 | 45457437@N00-2421 3 0 13 605 | 7512717@N06-3638 0 6 14 606 | 28370443@N08-1471 17 11 0 16 607 | 29365083@N06-1529 2 0 15 608 | 69349066@N00-3245 6 7 1 609 | 36967808@N00-2052 4 3 14 7 9 1 610 | 43405950@N07-2320 2 5 16 611 | 92134758@N00-4591 3 6 5 612 | 7888746@N08-3991 3 6 8 7 613 | 70938076@N00-3282 2 3 12 614 | 70938076@N00-3285 15 13 16 615 | 70938076@N00-3284 17 11 16 616 | 77047514@N00-3814 0 20 5 617 | 25866217@N00-1264 4 7 1 618 | 25866217@N00-1262 0 7 19 619 | 26276803@N05-1290 3 0 6 7 5 16 620 | 12917962@N00-260 2 12 6 14 621 | 23539578@N06-1111 4 0 6 5 622 | 23539578@N06-1110 4 3 0 5 623 | 12658277@N00-229 4 3 0 13 624 | 7512717@N06-3565 4 0 8 625 | 54069048@N07-2769 3 5 1 626 | 73014677@N05-3381 4 2 11 6 627 | 23351536@N07-1055 2 6 8 9 628 | 50195515@N00-2583 4 3 0 13 6 14 1 629 | 8701764@N02-4297 4 0 6 1 630 | 74691311@N00-3466 4 15 14 16 631 | 43498286@N00-2334 3 6 5 632 | 62136333@N02-3057 2 11 3 15 7 1 633 | 52015062@N00-2674 17 11 0 5 634 | 43498286@N00-2333 4 0 9 1 635 | -------------------------------------------------------------------------------- /seq2seq_c/python/ops/helper.py: -------------------------------------------------------------------------------- 1 | # Copyright 2016 The TensorFlow Authors. All Rights Reserved. 2 | # 3 | # Licensed under the Apache License, Version 2.0 (the "License"); 4 | # you may not use this file except in compliance with the License. 5 | # You may obtain a copy of the License at 6 | # 7 | # http://www.apache.org/licenses/LICENSE-2.0 8 | # 9 | # Unless required by applicable law or agreed to in writing, software 10 | # distributed under the License is distributed on an "AS IS" BASIS, 11 | # WITHOUT WARRANTIES OR CONDITIONS OF ANY KIND, either express or implied. 12 | # See the License for the specific language governing permissions and 13 | # limitations under the License. 14 | # ============================================================================== 15 | """A library of helpers for use with SamplingDecoders. 16 | """ 17 | 18 | from __future__ import absolute_import 19 | from __future__ import division 20 | from __future__ import print_function 21 | 22 | import abc 23 | 24 | import six 25 | import tensorflow as tf 26 | from seq2seq_c.python.ops import decoder 27 | from tensorflow.python.framework import dtypes 28 | from tensorflow.python.framework import ops 29 | from tensorflow.python.framework import tensor_shape 30 | from tensorflow.python.ops import array_ops 31 | from tensorflow.python.ops import control_flow_ops 32 | from tensorflow.python.ops import embedding_ops 33 | from tensorflow.python.ops import gen_array_ops 34 | from tensorflow.python.ops import math_ops 35 | from tensorflow.python.ops import tensor_array_ops 36 | from tensorflow.python.ops.distributions import bernoulli 37 | from tensorflow.python.ops.distributions import categorical 38 | from tensorflow.python.util import nest 39 | 40 | __all__ = [ 41 | "Helper", 42 | "TrainingHelper", 43 | "GreedyEmbeddingHelper", 44 | "GreedyEmbeddingHelper2", 45 | "SampleEmbeddingHelper", 46 | "CustomHelper", 47 | "ScheduledEmbeddingTrainingHelper", 48 | "ScheduledOutputTrainingHelper", 49 | "InferenceHelper", 50 | ] 51 | 52 | _transpose_batch_time = decoder._transpose_batch_time # pylint: disable=protected-access 53 | 54 | 55 | def _unstack_ta(inp): 56 | return tensor_array_ops.TensorArray( 57 | dtype=inp.dtype, size=array_ops.shape(inp)[0], 58 | element_shape=inp.get_shape()[1:]).unstack(inp) 59 | 60 | 61 | @six.add_metaclass(abc.ABCMeta) 62 | class Helper(object): 63 | """Interface for implementing sampling in seq2seq decoders. 64 | 65 | Helper instances are used by `BasicDecoder`. 66 | """ 67 | 68 | @abc.abstractproperty 69 | def batch_size(self): 70 | """Batch size of tensor returned by `sample`. 71 | 72 | Returns a scalar int32 tensor. 73 | """ 74 | raise NotImplementedError("batch_size has not been implemented") 75 | 76 | @abc.abstractproperty 77 | def sample_ids_shape(self): 78 | """Shape of tensor returned by `sample`, excluding the batch dimension. 79 | 80 | Returns a `TensorShape`. 81 | """ 82 | raise NotImplementedError("sample_ids_shape has not been implemented") 83 | 84 | @abc.abstractproperty 85 | def sample_ids_dtype(self): 86 | """DType of tensor returned by `sample`. 87 | 88 | Returns a DType. 89 | """ 90 | raise NotImplementedError("sample_ids_dtype has not been implemented") 91 | 92 | @abc.abstractmethod 93 | def initialize(self, name=None): 94 | """Returns `(initial_finished, initial_inputs)`.""" 95 | pass 96 | 97 | @abc.abstractmethod 98 | def sample(self, time, outputs, state, name=None): 99 | """Returns `sample_ids`.""" 100 | pass 101 | 102 | @abc.abstractmethod 103 | def next_inputs(self, time, outputs, state, sample_ids, name=None): 104 | """Returns `(finished, next_inputs, next_state)`.""" 105 | pass 106 | 107 | 108 | class CustomHelper(Helper): 109 | """Base abstract class that allows the user to customize sampling.""" 110 | 111 | def __init__(self, initialize_fn, sample_fn, next_inputs_fn, 112 | sample_ids_shape=None, sample_ids_dtype=None): 113 | """Initializer. 114 | 115 | Args: 116 | initialize_fn: callable that returns `(finished, next_inputs)` 117 | for the first iteration. 118 | sample_fn: callable that takes `(time, outputs, state)` 119 | and emits tensor `sample_ids`. 120 | next_inputs_fn: callable that takes `(time, outputs, state, sample_ids)` 121 | and emits `(finished, next_inputs, next_state)`. 122 | sample_ids_shape: Either a list of integers, or a 1-D Tensor of type 123 | `int32`, the shape of each value in the `sample_ids` batch. Defaults to 124 | a scalar. 125 | sample_ids_dtype: The dtype of the `sample_ids` tensor. Defaults to int32. 126 | """ 127 | self._initialize_fn = initialize_fn 128 | self._sample_fn = sample_fn 129 | self._next_inputs_fn = next_inputs_fn 130 | self._batch_size = None 131 | self._sample_ids_shape = tensor_shape.TensorShape(sample_ids_shape or []) 132 | self._sample_ids_dtype = sample_ids_dtype or dtypes.int32 133 | 134 | @property 135 | def batch_size(self): 136 | if self._batch_size is None: 137 | raise ValueError("batch_size accessed before initialize was called") 138 | return self._batch_size 139 | 140 | @property 141 | def sample_ids_shape(self): 142 | return self._sample_ids_shape 143 | 144 | @property 145 | def sample_ids_dtype(self): 146 | return self._sample_ids_dtype 147 | 148 | def initialize(self, name=None): 149 | with ops.name_scope(name, "%sInitialize" % type(self).__name__): 150 | (finished, next_inputs) = self._initialize_fn() 151 | if self._batch_size is None: 152 | self._batch_size = array_ops.size(finished) 153 | return (finished, next_inputs) 154 | 155 | def sample(self, time, outputs, state, name=None): 156 | with ops.name_scope( 157 | name, "%sSample" % type(self).__name__, (time, outputs, state)): 158 | return self._sample_fn(time=time, outputs=outputs, state=state) 159 | 160 | def next_inputs(self, time, outputs, state, sample_ids, name=None): 161 | with ops.name_scope( 162 | name, "%sNextInputs" % type(self).__name__, (time, outputs, state)): 163 | return self._next_inputs_fn( 164 | time=time, outputs=outputs, state=state, sample_ids=sample_ids) 165 | 166 | 167 | class TrainingHelper(Helper): 168 | """A helper for use during training. Only reads inputs. 169 | 170 | Returned sample_ids are the argmax of the RNN output logits. 171 | """ 172 | 173 | def __init__(self, inputs, sequence_length, time_major=False, name=None): 174 | """Initializer. 175 | 176 | Args: 177 | inputs: A (structure of) input tensors. 178 | sequence_length: An int32 vector tensor. 179 | time_major: Python bool. Whether the tensors in `inputs` are time major. 180 | If `False` (default), they are assumed to be batch major. 181 | name: Name scope for any created operations. 182 | 183 | Raises: 184 | ValueError: if `sequence_length` is not a 1D tensor. 185 | """ 186 | with ops.name_scope(name, "TrainingHelper", [inputs, sequence_length]): 187 | inputs = ops.convert_to_tensor(inputs, name="inputs") 188 | self._inputs = inputs 189 | if not time_major: 190 | inputs = nest.map_structure(_transpose_batch_time, inputs) 191 | 192 | self._input_tas = nest.map_structure(_unstack_ta, inputs) 193 | self._sequence_length = ops.convert_to_tensor( 194 | sequence_length, name="sequence_length") 195 | if self._sequence_length.get_shape().ndims != 1: 196 | raise ValueError( 197 | "Expected sequence_length to be a vector, but received shape: %s" % 198 | self._sequence_length.get_shape()) 199 | 200 | self._zero_inputs = nest.map_structure( 201 | lambda inp: array_ops.zeros_like(inp[0, :]), inputs) 202 | 203 | self._batch_size = array_ops.size(sequence_length) 204 | 205 | @property 206 | def inputs(self): 207 | return self._inputs 208 | 209 | @property 210 | def sequence_length(self): 211 | return self._sequence_length 212 | 213 | @property 214 | def batch_size(self): 215 | return self._batch_size 216 | 217 | @property 218 | def sample_ids_shape(self): 219 | return tensor_shape.TensorShape([]) 220 | 221 | @property 222 | def sample_ids_dtype(self): 223 | return dtypes.int32 224 | 225 | def initialize(self, name=None): 226 | with ops.name_scope(name, "TrainingHelperInitialize"): 227 | finished = math_ops.equal(0, self._sequence_length) 228 | all_finished = math_ops.reduce_all(finished) 229 | next_inputs = control_flow_ops.cond( 230 | all_finished, lambda: self._zero_inputs, 231 | lambda: nest.map_structure(lambda inp: inp.read(0), self._input_tas)) 232 | return (finished, next_inputs) 233 | 234 | def sample(self, time, outputs, name=None, **unused_kwargs): 235 | with ops.name_scope(name, "TrainingHelperSample", [time, outputs]): 236 | sample_ids = math_ops.cast( 237 | math_ops.argmax(outputs, axis=-1), dtypes.int32) 238 | return sample_ids 239 | 240 | def next_inputs(self, time, outputs, state, name=None, **unused_kwargs): 241 | """next_inputs_fn for TrainingHelper.""" 242 | with ops.name_scope(name, "TrainingHelperNextInputs", 243 | [time, outputs, state]): 244 | next_time = time + 1 245 | finished = (next_time >= self._sequence_length) 246 | all_finished = math_ops.reduce_all(finished) 247 | def read_from_ta(inp): 248 | return inp.read(next_time) 249 | next_inputs = control_flow_ops.cond( 250 | all_finished, lambda: self._zero_inputs, 251 | lambda: nest.map_structure(read_from_ta, self._input_tas)) 252 | return (finished, next_inputs, state) 253 | 254 | 255 | class ScheduledEmbeddingTrainingHelper(TrainingHelper): 256 | """A training helper that adds scheduled sampling. 257 | 258 | Returns -1s for sample_ids where no sampling took place; valid sample id 259 | values elsewhere. 260 | """ 261 | 262 | def __init__(self, inputs, sequence_length, embedding, sampling_probability, 263 | time_major=False, seed=None, scheduling_seed=None, name=None): 264 | """Initializer. 265 | 266 | Args: 267 | inputs: A (structure of) input tensors. 268 | sequence_length: An int32 vector tensor. 269 | embedding: A callable that takes a vector tensor of `ids` (argmax ids), 270 | or the `params` argument for `embedding_lookup`. 271 | sampling_probability: A 0D `float32` tensor: the probability of sampling 272 | categorically from the output ids instead of reading directly from the 273 | inputs. 274 | time_major: Python bool. Whether the tensors in `inputs` are time major. 275 | If `False` (default), they are assumed to be batch major. 276 | seed: The sampling seed. 277 | scheduling_seed: The schedule decision rule sampling seed. 278 | name: Name scope for any created operations. 279 | 280 | Raises: 281 | ValueError: if `sampling_probability` is not a scalar or vector. 282 | """ 283 | with ops.name_scope(name, "ScheduledEmbeddingSamplingWrapper", 284 | [embedding, sampling_probability]): 285 | if callable(embedding): 286 | self._embedding_fn = embedding 287 | else: 288 | self._embedding_fn = ( 289 | lambda ids: embedding_ops.embedding_lookup(embedding, ids)) 290 | self._sampling_probability = ops.convert_to_tensor( 291 | sampling_probability, name="sampling_probability") 292 | if self._sampling_probability.get_shape().ndims not in (0, 1): 293 | raise ValueError( 294 | "sampling_probability must be either a scalar or a vector. " 295 | "saw shape: %s" % (self._sampling_probability.get_shape())) 296 | self._seed = seed 297 | self._scheduling_seed = scheduling_seed 298 | super(ScheduledEmbeddingTrainingHelper, self).__init__( 299 | inputs=inputs, 300 | sequence_length=sequence_length, 301 | time_major=time_major, 302 | name=name) 303 | 304 | def initialize(self, name=None): 305 | return super(ScheduledEmbeddingTrainingHelper, self).initialize(name=name) 306 | 307 | def sample(self, time, outputs, state, name=None): 308 | with ops.name_scope(name, "ScheduledEmbeddingTrainingHelperSample", 309 | [time, outputs, state]): 310 | # Return -1s where we did not sample, and sample_ids elsewhere 311 | select_sampler = bernoulli.Bernoulli( 312 | probs=self._sampling_probability, dtype=dtypes.bool) 313 | select_sample = select_sampler.sample( 314 | sample_shape=self.batch_size, seed=self._scheduling_seed) 315 | sample_id_sampler = categorical.Categorical(logits=outputs) 316 | return array_ops.where( 317 | select_sample, 318 | sample_id_sampler.sample(seed=self._seed), 319 | gen_array_ops.fill([self.batch_size], -1)) 320 | 321 | def next_inputs(self, time, outputs, state, sample_ids, name=None): 322 | with ops.name_scope(name, "ScheduledEmbeddingTrainingHelperNextInputs", 323 | [time, outputs, state, sample_ids]): 324 | (finished, base_next_inputs, state) = ( 325 | super(ScheduledEmbeddingTrainingHelper, self).next_inputs( 326 | time=time, 327 | outputs=outputs, 328 | state=state, 329 | sample_ids=sample_ids, 330 | name=name)) 331 | 332 | def maybe_sample(): 333 | """Perform scheduled sampling.""" 334 | where_sampling = math_ops.cast( 335 | array_ops.where(sample_ids > -1), dtypes.int32) 336 | where_not_sampling = math_ops.cast( 337 | array_ops.where(sample_ids <= -1), dtypes.int32) 338 | sample_ids_sampling = array_ops.gather_nd(sample_ids, where_sampling) 339 | inputs_not_sampling = array_ops.gather_nd( 340 | base_next_inputs, where_not_sampling) 341 | sampled_next_inputs = self._embedding_fn(sample_ids_sampling) 342 | base_shape = array_ops.shape(base_next_inputs) 343 | return (array_ops.scatter_nd(indices=where_sampling, 344 | updates=sampled_next_inputs, 345 | shape=base_shape) 346 | + array_ops.scatter_nd(indices=where_not_sampling, 347 | updates=inputs_not_sampling, 348 | shape=base_shape)) 349 | 350 | all_finished = math_ops.reduce_all(finished) 351 | next_inputs = control_flow_ops.cond( 352 | all_finished, lambda: base_next_inputs, maybe_sample) 353 | return (finished, next_inputs, state) 354 | 355 | 356 | class ScheduledOutputTrainingHelper(TrainingHelper): 357 | """A training helper that adds scheduled sampling directly to outputs. 358 | 359 | Returns False for sample_ids where no sampling took place; True elsewhere. 360 | """ 361 | 362 | def __init__(self, inputs, sequence_length, sampling_probability, 363 | time_major=False, seed=None, next_inputs_fn=None, 364 | auxiliary_inputs=None, name=None): 365 | """Initializer. 366 | 367 | Args: 368 | inputs: A (structure) of input tensors. 369 | sequence_length: An int32 vector tensor. 370 | sampling_probability: A 0D `float32` tensor: the probability of sampling 371 | from the outputs instead of reading directly from the inputs. 372 | time_major: Python bool. Whether the tensors in `inputs` are time major. 373 | If `False` (default), they are assumed to be batch major. 374 | seed: The sampling seed. 375 | next_inputs_fn: (Optional) callable to apply to the RNN outputs to create 376 | the next input when sampling. If `None` (default), the RNN outputs will 377 | be used as the next inputs. 378 | auxiliary_inputs: An optional (structure of) auxiliary input tensors with 379 | a shape that matches `inputs` in all but (potentially) the final 380 | dimension. These tensors will be concatenated to the sampled output or 381 | the `inputs` when not sampling for use as the next input. 382 | name: Name scope for any created operations. 383 | 384 | Raises: 385 | ValueError: if `sampling_probability` is not a scalar or vector. 386 | """ 387 | with ops.name_scope(name, "ScheduledOutputTrainingHelper", 388 | [inputs, auxiliary_inputs, sampling_probability]): 389 | self._sampling_probability = ops.convert_to_tensor( 390 | sampling_probability, name="sampling_probability") 391 | if self._sampling_probability.get_shape().ndims not in (0, 1): 392 | raise ValueError( 393 | "sampling_probability must be either a scalar or a vector. " 394 | "saw shape: %s" % (self._sampling_probability.get_shape())) 395 | 396 | if auxiliary_inputs is None: 397 | maybe_concatenated_inputs = inputs 398 | else: 399 | inputs = ops.convert_to_tensor(inputs, name="inputs") 400 | auxiliary_inputs = ops.convert_to_tensor( 401 | auxiliary_inputs, name="auxiliary_inputs") 402 | maybe_concatenated_inputs = nest.map_structure( 403 | lambda x, y: array_ops.concat((x, y), -1), 404 | inputs, auxiliary_inputs) 405 | if not time_major: 406 | auxiliary_inputs = nest.map_structure( 407 | _transpose_batch_time, auxiliary_inputs) 408 | 409 | self._auxiliary_input_tas = ( 410 | nest.map_structure(_unstack_ta, auxiliary_inputs) 411 | if auxiliary_inputs is not None else None) 412 | 413 | self._seed = seed 414 | 415 | self._next_inputs_fn = next_inputs_fn 416 | 417 | super(ScheduledOutputTrainingHelper, self).__init__( 418 | inputs=maybe_concatenated_inputs, 419 | sequence_length=sequence_length, 420 | time_major=time_major, 421 | name=name) 422 | 423 | def initialize(self, name=None): 424 | return super(ScheduledOutputTrainingHelper, self).initialize(name=name) 425 | 426 | def sample(self, time, outputs, state, name=None): 427 | with ops.name_scope(name, "ScheduledOutputTrainingHelperSample", 428 | [time, outputs, state]): 429 | sampler = bernoulli.Bernoulli(probs=self._sampling_probability) 430 | return sampler.sample(sample_shape=self.batch_size, seed=self._seed) 431 | 432 | def next_inputs(self, time, outputs, state, sample_ids, name=None): 433 | with ops.name_scope(name, "ScheduledOutputTrainingHelperNextInputs", 434 | [time, outputs, state, sample_ids]): 435 | (finished, base_next_inputs, state) = ( 436 | super(ScheduledOutputTrainingHelper, self).next_inputs( 437 | time=time, 438 | outputs=outputs, 439 | state=state, 440 | sample_ids=sample_ids, 441 | name=name)) 442 | sample_ids = math_ops.cast(sample_ids, dtypes.bool) 443 | 444 | def maybe_sample(): 445 | """Perform scheduled sampling.""" 446 | 447 | def maybe_concatenate_auxiliary_inputs(outputs_, indices=None): 448 | """Concatenate outputs with auxiliary inputs, if they exist.""" 449 | if self._auxiliary_input_tas is None: 450 | return outputs_ 451 | 452 | next_time = time + 1 453 | auxiliary_inputs = nest.map_structure( 454 | lambda ta: ta.read(next_time), self._auxiliary_input_tas) 455 | if indices is not None: 456 | auxiliary_inputs = array_ops.gather_nd(auxiliary_inputs, indices) 457 | return nest.map_structure( 458 | lambda x, y: array_ops.concat((x, y), -1), 459 | outputs_, auxiliary_inputs) 460 | 461 | if self._next_inputs_fn is None: 462 | return array_ops.where( 463 | sample_ids, maybe_concatenate_auxiliary_inputs(outputs), 464 | base_next_inputs) 465 | 466 | where_sampling = math_ops.cast( 467 | array_ops.where(sample_ids), dtypes.int32) 468 | where_not_sampling = math_ops.cast( 469 | array_ops.where(math_ops.logical_not(sample_ids)), dtypes.int32) 470 | outputs_sampling = array_ops.gather_nd(outputs, where_sampling) 471 | inputs_not_sampling = array_ops.gather_nd(base_next_inputs, 472 | where_not_sampling) 473 | sampled_next_inputs = maybe_concatenate_auxiliary_inputs( 474 | self._next_inputs_fn(outputs_sampling), where_sampling) 475 | 476 | base_shape = array_ops.shape(base_next_inputs) 477 | return (array_ops.scatter_nd(indices=where_sampling, 478 | updates=sampled_next_inputs, 479 | shape=base_shape) 480 | + array_ops.scatter_nd(indices=where_not_sampling, 481 | updates=inputs_not_sampling, 482 | shape=base_shape)) 483 | 484 | all_finished = math_ops.reduce_all(finished) 485 | no_samples = math_ops.logical_not(math_ops.reduce_any(sample_ids)) 486 | next_inputs = control_flow_ops.cond( 487 | math_ops.logical_or(all_finished, no_samples), 488 | lambda: base_next_inputs, maybe_sample) 489 | return (finished, next_inputs, state) 490 | 491 | 492 | class GreedyEmbeddingHelper(Helper): 493 | """A helper for use during inference. 494 | 495 | Uses the argmax of the output (treated as logits) and passes the 496 | result through an embedding layer to get the next input. 497 | """ 498 | 499 | def __init__(self, embedding, start_tokens, end_token): 500 | """Initializer. 501 | 502 | Args: 503 | embedding: A callable that takes a vector tensor of `ids` (argmax ids), 504 | or the `params` argument for `embedding_lookup`. The returned tensor 505 | will be passed to the decoder input. 506 | start_tokens: `int32` vector shaped `[batch_size]`, the start tokens. 507 | end_token: `int32` scalar, the token that marks end of decoding. 508 | 509 | Raises: 510 | ValueError: if `start_tokens` is not a 1D tensor or `end_token` is not a 511 | scalar. 512 | """ 513 | if callable(embedding): 514 | self._embedding_fn = embedding 515 | else: 516 | self._embedding_fn = ( 517 | lambda ids: embedding_ops.embedding_lookup(embedding, ids)) 518 | 519 | self._start_tokens = ops.convert_to_tensor( 520 | start_tokens, dtype=dtypes.int32, name="start_tokens") 521 | self._end_token = ops.convert_to_tensor( 522 | end_token, dtype=dtypes.int32, name="end_token") 523 | if self._start_tokens.get_shape().ndims != 1: 524 | raise ValueError("start_tokens must be a vector") 525 | self._batch_size = array_ops.size(start_tokens) 526 | if self._end_token.get_shape().ndims != 0: 527 | raise ValueError("end_token must be a scalar") 528 | self._start_inputs = self._embedding_fn(self._start_tokens) 529 | 530 | @property 531 | def batch_size(self): 532 | return self._batch_size 533 | 534 | @property 535 | def sample_ids_shape(self): 536 | return tensor_shape.TensorShape([]) 537 | 538 | @property 539 | def sample_ids_dtype(self): 540 | return dtypes.int32 541 | 542 | def initialize(self, name=None): 543 | finished = array_ops.tile([False], [self._batch_size]) 544 | return (finished, self._start_inputs) 545 | 546 | def sample(self, time, outputs, state, name=None): 547 | """sample for GreedyEmbeddingHelper.""" 548 | del time, state # unused by sample_fn 549 | # Outputs are logits, use argmax to get the most probable id 550 | if not isinstance(outputs, ops.Tensor): 551 | raise TypeError("Expected outputs to be a single Tensor, got: %s" % 552 | type(outputs)) 553 | sample_ids = math_ops.argmax(outputs, axis=-1, output_type=dtypes.int32) 554 | return sample_ids 555 | 556 | def next_inputs(self, time, outputs, state, sample_ids, name=None): 557 | """next_inputs_fn for GreedyEmbeddingHelper.""" 558 | del time, outputs # unused by next_inputs_fn 559 | finished = math_ops.equal(sample_ids, self._end_token) 560 | all_finished = math_ops.reduce_all(finished) 561 | next_inputs = control_flow_ops.cond( 562 | all_finished, 563 | # If we're finished, the next_inputs value doesn't matter 564 | lambda: self._start_inputs, 565 | lambda: self._embedding_fn(sample_ids)) 566 | return (finished, next_inputs, state) 567 | 568 | class GreedyEmbeddingHelper2(Helper): 569 | """A helper for use during training. Only reads inputs. 570 | 571 | Returned sample_ids are the argmax of the RNN output logits. 572 | """ 573 | 574 | def __init__(self, embedding, sequence_length, start_tokens,name=None): 575 | """Initializer. 576 | 577 | Args: 578 | inputs: A (structure of) input tensors. 579 | sequence_length: An int32 vector tensor. 580 | time_major: Python bool. Whether the tensors in `inputs` are time major. 581 | If `False` (default), they are assumed to be batch major. 582 | name: Name scope for any created operations. 583 | 584 | Raises: 585 | ValueError: if `sequence_length` is not a 1D tensor. 586 | """ 587 | with ops.name_scope(name, "TrainingHelper", [sequence_length]): 588 | self._sequence_length = ops.convert_to_tensor( 589 | sequence_length, name="sequence_length") 590 | if self._sequence_length.get_shape().ndims != 1: 591 | raise ValueError( 592 | "Expected sequence_length to be a vector, but received shape: %s" % 593 | self._sequence_length.get_shape()) 594 | 595 | 596 | self._batch_size = array_ops.size(sequence_length) 597 | if callable(embedding): 598 | self._embedding_fn = embedding 599 | else: 600 | self._embedding_fn = ( 601 | lambda ids: embedding_ops.embedding_lookup(embedding, ids)) 602 | 603 | self._start_tokens = ops.convert_to_tensor( 604 | start_tokens, dtype=dtypes.int32, name="start_tokens") 605 | self._start_inputs = self._embedding_fn(self._start_tokens) 606 | 607 | 608 | @property 609 | def sequence_length(self): 610 | return self._sequence_length 611 | @property 612 | def batch_size(self): 613 | return self._batch_size 614 | 615 | @property 616 | def sample_ids_shape(self): 617 | return tensor_shape.TensorShape([]) 618 | 619 | @property 620 | def sample_ids_dtype(self): 621 | return dtypes.int32 622 | 623 | def initialize(self, name=None): 624 | with ops.name_scope(name, "TrainingHelperInitialize"): 625 | finished = math_ops.equal(0, self._sequence_length) 626 | 627 | return (finished, self._start_inputs) 628 | 629 | def sample(self, time, outputs, name=None, **unused_kwargs): 630 | with ops.name_scope(name, "TrainingHelperSample", [time, outputs]): 631 | sample_ids = math_ops.cast( 632 | math_ops.argmax(outputs, axis=-1), dtypes.int32) 633 | return sample_ids 634 | 635 | def next_inputs(self, time, outputs, state, sample_ids,name=None, **unused_kwargs): 636 | """next_inputs_fn for TrainingHelper.""" 637 | with ops.name_scope(name, "TrainingHelperNextInputs", 638 | [time, outputs, state]): 639 | next_time = time + 1 640 | finished = (next_time >= self._sequence_length) 641 | all_finished = math_ops.reduce_all(finished) 642 | next_inputs = control_flow_ops.cond( 643 | all_finished, 644 | # If we're finished, the next_inputs value doesn't matter 645 | lambda: self._start_inputs, 646 | lambda: self._embedding_fn(sample_ids)) 647 | 648 | 649 | return (finished, next_inputs, state) 650 | 651 | class SampleEmbeddingHelper(GreedyEmbeddingHelper): 652 | """A helper for use during inference. 653 | 654 | Uses sampling (from a distribution) instead of argmax and passes the 655 | result through an embedding layer to get the next input. 656 | """ 657 | 658 | def __init__(self, embedding, start_tokens, end_token, 659 | softmax_temperature=None, seed=None): 660 | """Initializer. 661 | 662 | Args: 663 | embedding: A callable that takes a vector tensor of `ids` (argmax ids), 664 | or the `params` argument for `embedding_lookup`. The returned tensor 665 | will be passed to the decoder input. 666 | start_tokens: `int32` vector shaped `[batch_size]`, the start tokens. 667 | end_token: `int32` scalar, the token that marks end of decoding. 668 | softmax_temperature: (Optional) `float32` scalar, value to divide the 669 | logits by before computing the softmax. Larger values (above 1.0) result 670 | in more random samples, while smaller values push the sampling 671 | distribution towards the argmax. Must be strictly greater than 0. 672 | Defaults to 1.0. 673 | seed: (Optional) The sampling seed. 674 | 675 | Raises: 676 | ValueError: if `start_tokens` is not a 1D tensor or `end_token` is not a 677 | scalar. 678 | """ 679 | super(SampleEmbeddingHelper, self).__init__( 680 | embedding, start_tokens, end_token) 681 | self._softmax_temperature = softmax_temperature 682 | self._seed = seed 683 | 684 | def sample(self, time, outputs, state, name=None): 685 | """sample for SampleEmbeddingHelper.""" 686 | del time, state # unused by sample_fn 687 | # Outputs are logits, we sample instead of argmax (greedy). 688 | if not isinstance(outputs, ops.Tensor): 689 | raise TypeError("Expected outputs to be a single Tensor, got: %s" % 690 | type(outputs)) 691 | if self._softmax_temperature is None: 692 | logits = outputs 693 | else: 694 | logits = outputs / self._softmax_temperature 695 | 696 | sample_id_sampler = categorical.Categorical(logits=logits) 697 | sample_ids = sample_id_sampler.sample(seed=self._seed) 698 | 699 | return sample_ids 700 | 701 | 702 | class InferenceHelper(Helper): 703 | """A helper to use during inference with a custom sampling function.""" 704 | 705 | def __init__(self, sample_fn, sample_shape, sample_dtype, 706 | start_inputs, end_fn, next_inputs_fn=None): 707 | """Initializer. 708 | 709 | Args: 710 | sample_fn: A callable that takes `outputs` and emits tensor `sample_ids`. 711 | sample_shape: Either a list of integers, or a 1-D Tensor of type `int32`, 712 | the shape of the each sample in the batch returned by `sample_fn`. 713 | sample_dtype: the dtype of the sample returned by `sample_fn`. 714 | start_inputs: The initial batch of inputs. 715 | end_fn: A callable that takes `sample_ids` and emits a `bool` vector 716 | shaped `[batch_size]` indicating whether each sample is an end token. 717 | next_inputs_fn: (Optional) A callable that takes `sample_ids` and returns 718 | the next batch of inputs. If not provided, `sample_ids` is used as the 719 | next batch of inputs. 720 | """ 721 | self._sample_fn = sample_fn 722 | self._end_fn = end_fn 723 | self._sample_shape = tensor_shape.TensorShape(sample_shape) 724 | self._sample_dtype = sample_dtype 725 | self._next_inputs_fn = next_inputs_fn 726 | self._batch_size = array_ops.shape(start_inputs)[0] 727 | self._start_inputs = ops.convert_to_tensor( 728 | start_inputs, name="start_inputs") 729 | 730 | @property 731 | def batch_size(self): 732 | return self._batch_size 733 | 734 | @property 735 | def sample_ids_shape(self): 736 | return self._sample_shape 737 | 738 | @property 739 | def sample_ids_dtype(self): 740 | return self._sample_dtype 741 | 742 | def initialize(self, name=None): 743 | finished = array_ops.tile([False], [self._batch_size]) 744 | return (finished, self._start_inputs) 745 | 746 | def sample(self, time, outputs, state, name=None): 747 | del time, state # unused by sample 748 | return self._sample_fn(outputs) 749 | 750 | def next_inputs(self, time, outputs, state, sample_ids, name=None): 751 | del time, outputs # unused by next_inputs 752 | if self._next_inputs_fn is None: 753 | next_inputs = sample_ids 754 | else: 755 | next_inputs = self._next_inputs_fn(sample_ids) 756 | finished = self._end_fn(sample_ids) 757 | return (finished, next_inputs, state) 758 | -------------------------------------------------------------------------------- /Trip/gae_context.py: -------------------------------------------------------------------------------- 1 | # -*- coding: UTF-8 -*- 2 | from __future__ import division 3 | import math 4 | from tensorflow.python.layers.core import Dense 5 | import seq2seq_c as seqc 6 | from metric import * 7 | from ops import * 8 | import time 9 | import datetime 10 | from tensorflow.python.ops.rnn_cell_impl import LSTMStateTuple 11 | # =============================== vars ====================================== # 12 | is_initial=True 13 | EPSILON = 1e-6 14 | batch_size=8 15 | n_hidden=512 16 | AE_learning_rate=0.6 17 | initializer=tf.truncated_normal_initializer(stddev=0.02) 18 | critic_lr=1e-02 19 | gen_lr=1e-02 20 | z_dim=128 21 | c_dim=512 22 | train_iters =20 # 遍历样本次数 for training 23 | embedding_size=256 24 | dynamic_traning=True 25 | dat_suffix = ['Osak', 'Glas', 'Edin', 'Toro', 'Melb','TKY_split200'] 26 | dat_ix=5 27 | poi_name="poi-"+dat_suffix[dat_ix]+".csv" #Edin 28 | tra_name="traj-"+dat_suffix[dat_ix]+".csv" 29 | embedding_name=dat_suffix[dat_ix] 30 | model='./logs/model_'+embedding_name+'.pkt' 31 | # =============================== data load ====================================== # 32 | #load original data 33 | op_tdata = open('origin_data/'+poi_name, 'r') 34 | 35 | ot_tdata = open('origin_data/'+tra_name, 'r') 36 | print 'To Train',dat_suffix[dat_ix] 37 | POIs=[] 38 | Trajectory=[] 39 | for line in op_tdata.readlines(): 40 | lineArr = line.split(',') 41 | temp_line=list() 42 | for item in lineArr: 43 | temp_line.append(item.strip('\n')) 44 | POIs.append(temp_line) 45 | POIs=POIs[1:] #remove first line 46 | 47 | def calc_dist_vec(longitudes1, latitudes1, longitudes2, latitudes2): 48 | """Calculate the distance (unit: km) between two places on earth, vectorised""" 49 | # convert degrees to radians 50 | lng1 = np.radians(longitudes1) 51 | lat1 = np.radians(latitudes1) 52 | lng2 = np.radians(longitudes2) 53 | lat2 = np.radians(latitudes2) 54 | radius = 6371.0088 # mean earth radius, en.wikipedia.org/wiki/Earth_radius#Mean_radius 55 | 56 | # The haversine formula, en.wikipedia.org/wiki/Great-circle_distance 57 | dlng = np.fabs(lng1 - lng2) 58 | dlat = np.fabs(lat1 - lat2) 59 | dist = 2 * radius * np.arcsin( np.sqrt( 60 | (np.sin(0.5*dlat))**2 + np.cos(lat1) * np.cos(lat2) * (np.sin(0.5*dlng))**2 )) 61 | return dist 62 | # 63 | 64 | 65 | 66 | #print POIs 67 | get_POIs={} 68 | char_pois=[] #pois chars 69 | 70 | for items in POIs: 71 | char_pois.append(items[0]) 72 | get_POIs.setdefault(items[0],[]).append([items[2],items[3]]) # pois to category 73 | Users=[] 74 | poi_count={} 75 | for line in ot_tdata.readlines(): 76 | lineArr=line.split(',') 77 | temp_line=list() 78 | if lineArr[0]=='userID': 79 | continue 80 | poi_count.setdefault(lineArr[2], []).append(lineArr[2]) 81 | for i in range(len(lineArr)): 82 | 83 | if i==0: 84 | user = lineArr[i] 85 | Users.append(user) # add user id 86 | temp_line.append(user) 87 | continue 88 | temp_line.append(lineArr[i].strip('\n')) 89 | Trajectory.append(temp_line) 90 | Users=sorted(list(set(Users))) 91 | print 'user number',len(Users) 92 | 93 | TRAIN_TRA=[] 94 | TRAIN_USER=[] 95 | TRAIN_TIME=[] 96 | TRAIN_DIST=[] 97 | DATA={} #temp_data 98 | for index in range(len(Trajectory)): 99 | if(int(Trajectory[index][-2])>=3): #the length of the trajectory must over than 3 100 | DATA.setdefault(Trajectory[index][0]+'-'+Trajectory[index][1],[]).append([Trajectory[index][2],Trajectory[index][3],Trajectory[index][4]]) #userID+trajID 101 | 102 | 103 | 104 | # #calc distance 105 | # distance_count=[] 106 | # for i in range(len(POIs)): 107 | # lon1=float(POIs[i][2]) 108 | # lat1=float(POIs[i][3]) 109 | # for j in range(i+1,len(POIs)): 110 | # lon2 = float(POIs[j][2]) 111 | # lat2 =float( POIs[j][3]) 112 | # dist=calc_dist_vec(lon1,lat1,lon2,lat2) 113 | # distance_count.append(dist) 114 | # print 'max',max(distance_count) 115 | # print 'min',min(distance_count) 116 | 117 | #calc_distance 118 | distance_count=[] 119 | for key in DATA.keys(): 120 | traj=DATA[key] 121 | #print traj 122 | for i in range(len(traj)): 123 | #print get_POIs[traj[i][0]][0][0] 124 | lon1=float(get_POIs[traj[i][0]][0][0]) 125 | lat1=float(get_POIs[traj[i][0]][0][1]) 126 | for j in range(i+1,len(traj)): 127 | lon2 = float(get_POIs[traj[j][0]][0][0]) 128 | lat2 = float(get_POIs[traj[j][0]][0][1]) 129 | distance_count.append(calc_dist_vec(lon1,lat1,lon2,lat2)) 130 | upper_dis=max(distance_count) 131 | lower_dis=min(distance_count) 132 | print len(distance_count) 133 | # rand_l=np.random.rand(-1,1) 134 | # rand_u=np.random.rand(-1,1) 135 | # print rand_l 136 | # val=((min(distance_count)-min(distance_count))*rand_l+(max(distance_count)-min(distance_count))*rand_u)/(min(distance_count)-min(distance_count)+max(distance_count)-min(distance_count)) 137 | # # def calc_distance(distance): 138 | # # val = ((distance - min(distance_count)) * rand_l + (max(distance_count) - distance) * rand_u)/(distance - min(distance_count) + max(distance_count) - distance) 139 | # # return val 140 | 141 | for keys in DATA.keys(): 142 | user_traj=DATA[keys] 143 | temp_poi=[] 144 | temp_time=[] 145 | temp_dist=[] 146 | for i in range(len(user_traj)): 147 | temp_poi.append(user_traj[i][0]) #add poi id 148 | lon1=float(get_POIs[user_traj[i][0]][0][0]) 149 | lat1=float(get_POIs[user_traj[i][0]][0][1]) 150 | lons=float(get_POIs[user_traj[0][0]][0][0]) 151 | lats=float(get_POIs[user_traj[0][0]][0][1]) 152 | lone=float(get_POIs[user_traj[-1][0]][0][0]) 153 | late=float(get_POIs[user_traj[-1][0]][0][1]) 154 | sd=calc_dist_vec(lon1,lat1,lons,lats) 155 | ed = calc_dist_vec(lon1, lat1, lone, late) 156 | value1=0.5*(sd)/max(distance_count) 157 | value2=0.5*(ed)/max(distance_count) 158 | #print value 159 | temp_dist.append([value1,value2]) #lon,lat 160 | 161 | dt = time.strftime("%H:%M:%S", time.localtime(int(user_traj[i][1:][0]))) 162 | #print dt.split(":")[0] 163 | temp_time.append(int(dt.split(":")[0])) #add poi time 164 | TRAIN_USER.append(keys) 165 | TRAIN_TRA.append(temp_poi) 166 | TRAIN_TIME.append(temp_time) 167 | TRAIN_DIST.append(temp_dist) 168 | dictionary={} 169 | for key in poi_count.keys(): 170 | count=len(poi_count[key]) 171 | dictionary[key]=count 172 | dictionary['GO']=1 173 | dictionary['PAD']=1 174 | dictionary['END']=1 175 | new_dict=sorted(dictionary.items(),key = lambda x:x[1],reverse = True) 176 | 177 | print 'poi number is',len(new_dict)-3 178 | voc_poi=list() 179 | 180 | for item in new_dict: 181 | voc_poi.append(item[0]) #has been sorted by frequency 182 | 183 | def extract_words_vocab(): 184 | int_to_vocab = {idx: word for idx, word in enumerate(voc_poi)} 185 | vocab_to_int = {word: idx for idx, word in int_to_vocab.items()} 186 | return int_to_vocab, vocab_to_int 187 | int_to_vocab, vocab_to_int=extract_words_vocab() 188 | 189 | #generate pre-traning dataset 190 | new_trainT = list() 191 | for i in range(len(TRAIN_TRA)): #TRAIN 192 | temp = list() 193 | temp.append(vocab_to_int['GO']) 194 | for j in range(len(TRAIN_TRA[i])): 195 | temp.append(vocab_to_int[TRAIN_TRA[i][j]]) 196 | temp.append(vocab_to_int['END']) 197 | temp.append(vocab_to_int['PAD']) 198 | new_trainT.append(temp) 199 | 200 | #generate traning dataset 201 | new_trainTs = list() 202 | for i in range(len(TRAIN_TRA)): #TRAIN 203 | temp = list() 204 | for j in range(len(TRAIN_TRA[i])): 205 | temp.append(vocab_to_int[TRAIN_TRA[i][j]]) 206 | new_trainTs.append(temp) 207 | 208 | dataset=open('data/'+embedding_name+'_set.dat','w') 209 | for i in range(len(new_trainTs)): 210 | dataset.write(str(TRAIN_USER[i])+'\t') 211 | for j in range(len(new_trainTs[i])): 212 | dataset.write(str(new_trainTs[i][j])+'\t') 213 | dataset.write('\n') 214 | dataset.close() 215 | 216 | #embeddings 217 | if dynamic_traning is True: 218 | embeddings=tf.Variable(tf.random_uniform([len(voc_poi),embedding_size],-1.0,1.0)) 219 | time_embeddings = tf.Variable(tf.random_uniform([24,32], -1.0, 1.0)) 220 | distance_embeddings1=tf.Variable(tf.random_uniform([32], -1.0, 1.0)) 221 | distance_embeddings2=tf.Variable(tf.random_uniform([32], -1.0, 1.0)) 222 | weights=tf.Variable(tf.truncated_normal([embedding_size,embedding_size],stddev=1.0/math.sqrt(embedding_size))) 223 | bias=tf.Variable(tf.zeros([embedding_size]),dtype=tf.float32) 224 | embeddings=tf.nn.xw_plus_b(embeddings,weights,bias) 225 | else: 226 | embeddings=[] 227 | fread_emb=open('data/'+embedding_name+'vec.dat','r') 228 | for line in fread_emb.readlines(): 229 | lineArr=line.split() 230 | temp=list() 231 | for i in range(1,len(lineArr)): 232 | temp.append(float(lineArr[i])) 233 | embeddings.append(temp) 234 | embeddings=tf.constant(embeddings) 235 | 236 | # =============================== data load end====================================== # 237 | 238 | # =============================== tf.vars ====================================== # 239 | keep_prob = tf.placeholder("float") 240 | lens=tf.placeholder(dtype=tf.int32) 241 | input_X=tf.placeholder(dtype=tf.int32, shape=[batch_size, None]) 242 | input_X_de=tf.placeholder(dtype=tf.int32, shape=[batch_size, None]) 243 | input_t=tf.placeholder(dtype=tf.int32, shape=[batch_size, None]) 244 | input_d1=tf.placeholder(dtype=tf.float32, shape=[batch_size, None]) 245 | input_d2=tf.placeholder(dtype=tf.float32, shape=[batch_size, None]) 246 | target_sequence_length = tf.placeholder(tf.int32, [None], name='target_sequence_length') 247 | max_target_sequence_length = tf.reduce_max(target_sequence_length, name='max_target_len') 248 | 249 | code = tf.placeholder(tf.float32, shape=[None, c_dim]) 250 | z = tf.placeholder(dtype=tf.int32, shape=[batch_size, None]) 251 | z_t=tf.placeholder(dtype=tf.int32, shape=[batch_size, None]) 252 | z_d1=tf.placeholder(dtype=tf.float32, shape=[batch_size, None]) 253 | z_d2=tf.placeholder(dtype=tf.float32, shape=[batch_size, None]) 254 | 255 | # =============================== Encoder ====================================== # 256 | def encoder(X,context,keep_prob=0.5): 257 | """ 258 | encode discrete feature to continuous latent vector 259 | :param tensor: [batch_size,length,embedding_size]. 260 | :return:encoded latent vector 261 | """ 262 | with tf.variable_scope("encoder"): 263 | tensor=tf.nn.embedding_lookup(embeddings,X) #find embeddings of trajectory:[batch_size,length,embedding_size]. 264 | time_t=tf.nn.embedding_lookup(time_embeddings,context[0]) 265 | print 'time_t',time_t 266 | space=tf.tensordot(context[1],distance_embeddings1,0)+tf.tensordot(context[2],distance_embeddings2,0) 267 | print 'space',space 268 | tensor=tf.concat([tensor,time_t],2) 269 | tensor=tf.concat([tensor,space],2) 270 | trans_tensor=tf.transpose(tensor,[1,0,2]) #[length,batch_size,embedding_size]. 271 | lstm_cell=tf.nn.rnn_cell.LSTMCell(n_hidden) 272 | dr_lstm_cell=tf.nn.rnn_cell.DropoutWrapper(lstm_cell,output_keep_prob=keep_prob) 273 | (output,states)=tf.nn.dynamic_rnn(dr_lstm_cell,trans_tensor,time_major=True,dtype=tf.float32) 274 | latent_code=output[-1] 275 | latent_code=tf.nn.l2_normalize(latent_code) 276 | print 'latentcode',latent_code 277 | #latent_code= fully_connected(latent_code, c_dim, initializer=initializer, is_last=True, scope="encoder_output") 278 | return latent_code,states 279 | 280 | # =============================== Decoder ====================================== # 281 | def decoder(tensor,X,en_state,reuse=False): 282 | """ 283 | decode continuous vector to probability of pixel 284 | :param tensor: 2-D tensor. 285 | :param output_dim: Integer. dimension of output 286 | :param is_mnist: Boolean. mnist or not 287 | :param reuse: Boolean. reuse or not 288 | :param is_train:Boolean. train or not 289 | :return: 2-D tensor. decoded vector (image) 290 | """ 291 | with tf.variable_scope('decoder',reuse=reuse) as scope: 292 | 293 | decode_lstm=tf.nn.rnn_cell.LSTMCell(n_hidden) 294 | decode_dr_lstm = tf.nn.rnn_cell.DropoutWrapper(decode_lstm, output_keep_prob=keep_prob) 295 | output_layer=Dense(len(vocab_to_int)) 296 | decoder_initial_state=en_state#LSTMStateTuple(c_state, h_state) 297 | 298 | copy = tf.tile(tf.constant([vocab_to_int['GO']]), [batch_size]) 299 | training_helper = seqc.GreedyEmbeddingHelper2(embeddings, 300 | sequence_length=target_sequence_length, start_tokens=copy) 301 | training_decoder = seqc.BasicDecoder(decode_dr_lstm, training_helper, decoder_initial_state,tensor,output_layer) # cell,helper, initial_state, out_layer(convert rnn_size to vocab_size) 302 | output, _, _ = seqc.dynamic_decode(training_decoder, 303 | impute_finished=True, 304 | maximum_iterations=max_target_sequence_length) 305 | predicting_logits = tf.identity(output.sample_id, name='predictions') 306 | training_logits = tf.identity(output.rnn_output, 'logits') 307 | masks = tf.sequence_mask(target_sequence_length, max_target_sequence_length, dtype=tf.float32, name='masks') 308 | target = X 309 | return output, predicting_logits, training_logits, masks, target 310 | 311 | 312 | # =============================== Generator ====================================== # 313 | def generator(z,context,reuse=False): 314 | """ 315 | generator of WGAN 316 | :param z: 2-D tensor. noise with standard normal distribution 317 | :param reuse: Boolean. reuse or not 318 | :return: 2-D tensor. latent vector 319 | """ 320 | """ 321 | encode discrete feature to continuous latent vector 322 | :param tensor: [batch_size,length,embedding_size]. 323 | :return:encoded latent vector 324 | """ 325 | with tf.variable_scope("generator"): 326 | tensor=tf.nn.embedding_lookup(embeddings,z) #find embeddings of trajectory:[batch_size,length,embedding_size]. 327 | time_t=tf.nn.embedding_lookup(time_embeddings,context[0]) 328 | space=tf.tensordot(context[1],distance_embeddings1,0)+tf.tensordot(context[2],distance_embeddings2,0) 329 | tensor=tf.concat([tensor,time_t],2) 330 | tensor=tf.concat([tensor,space],2) 331 | trans_tensor=tf.transpose(tensor,[1,0,2]) #[length,batch_size,embedding_size]. 332 | lstm_cell=tf.nn.rnn_cell.LSTMCell(n_hidden) 333 | dr_lstm_cell=tf.nn.rnn_cell.DropoutWrapper(lstm_cell,output_keep_prob=keep_prob) 334 | (output,states)=tf.nn.dynamic_rnn(dr_lstm_cell,trans_tensor,time_major=True,dtype=tf.float32) 335 | latent_code=output[-1] 336 | latent_code=tf.nn.l2_normalize(latent_code) 337 | return latent_code,states 338 | # =============================== Discriminator(Critic) ====================================== # 339 | def critic(latent,reuse=False): 340 | """ 341 | discriminator of WGAN 342 | :param latent: 2-D tensor. latent vector 343 | :param reuse: Boolean. reuse or not 344 | :return: 2-D tensor. logit of data or noise 345 | """ 346 | with tf.variable_scope("critic",reuse=reuse): 347 | fc_100 = fully_connected(latent, 100, initializer=initializer, scope="fc_100") 348 | fc_60 = fully_connected(fc_100, 60, initializer=initializer, scope="fc_60") 349 | fc_20 = fully_connected(fc_60, 20, initializer=initializer, scope="fc_20") 350 | output=fully_connected(fc_20,1,initializer=initializer,is_last=True,scope="critic_output") 351 | #WGAN does not using activate 352 | return output 353 | 354 | # =============================== Function ====================================== # 355 | def autoencoder(X,de_X,context,keep_prob): 356 | """ 357 | deep autoencoder. reconstruction the input data 358 | :param data: 2-D tensor. data for reconstruction 359 | :return: 2-D tensor. reconstructed data and latent vector 360 | """ 361 | with tf.variable_scope("autoencoder"): 362 | latent,en_state=encoder(X,context,keep_prob) 363 | output_, predicting_logits_, training_logits_, masks_, target_=decoder(latent,de_X,en_state) 364 | return training_logits_,masks_,target_,latent,predicting_logits_,en_state 365 | def pad_sentence_batch(sentence_batch, pad_int): 366 | ''' 367 | 对batch中的序列进行补全，保证batch中的每行都有相同的sequence_length 368 | 369 | 参数： 370 | - sentence batch 371 | - pad_int: 对应索引号 372 | ''' 373 | max_sentence = max([len(sentence) for sentence in sentence_batch]) # 取最大长度 374 | return [sentence + [pad_int] * (max_sentence - len(sentence)) for sentence in sentence_batch] 375 | 376 | def pad_time_batch(time_batch): 377 | ''' 378 | 对batch中的序列进行补全，保证batch中的每行都有相同的sequence_length 379 | 380 | 参数： 381 | - sentence batch 382 | - pad_int: 对应索引号 383 | ''' 384 | max_sentence = max([len(sentence) for sentence in time_batch]) # 取最大长度 385 | return [sentence + [0] * (max_sentence - len(sentence)) for sentence in time_batch] 386 | def pad_dist_batch(dist_batch): 387 | ''' 388 | 对batch中的序列进行补全，保证batch中的每行都有相同的sequence_length 389 | 390 | 参数： 391 | - sentence batch 392 | - pad_int: 对应索引号 393 | ''' 394 | max_sentence = max([len(sentence) for sentence in dist_batch]) # 取最大长度 395 | return [sentence + [sentence[-1]] * (max_sentence - len(sentence)) for sentence in dist_batch] 396 | def eos_sentence_batch(sentence_batch, eos_in): 397 | return [sentence + [eos_in] for sentence in sentence_batch] 398 | 399 | # =============================== Graph ====================================== # 400 | """ 401 | build network 402 | :return: 403 | """ 404 | context=[input_t,input_d1,input_d2] 405 | z_context=[z_t,z_d1,z_d2] 406 | training_logits_,masks_,target_,real_code,predicting_logits_,encoder_state=autoencoder(input_X,input_X_de,context,keep_prob) 407 | g_code,g_state=generator(z,z_context) #flat c 408 | print g_code 409 | critic_real=critic(real_code) 410 | print critic_real 411 | critic_fake=critic(g_code,reuse=True) 412 | print critic_fake 413 | #WGAN loss 414 | disc_real_loss=tf.reduce_mean(critic_real) 415 | disc_fake_loss=-tf.reduce_mean(critic_fake) #Negative sample phase 416 | critic_loss=tf.reduce_mean(critic_real)-tf.reduce_mean(critic_fake) 417 | gen_loss=tf.reduce_mean(critic_fake) #Train the generator 418 | 419 | #for discrete input, use cross entropy loss 420 | AE_loss=seqc.sequence_loss(training_logits_, target_, masks_) 421 | 422 | #get trainable variables 423 | AE_variables=tf.get_collection(tf.GraphKeys.TRAINABLE_VARIABLES,scope="autoencoder") 424 | gen_variables=tf.get_collection(tf.GraphKeys.TRAINABLE_VARIABLES,scope="generator") 425 | critic_variables=tf.get_collection(tf.GraphKeys.TRAINABLE_VARIABLES,scope='critic') 426 | encoder_variables=tf.get_collection(tf.GraphKeys.TRAINABLE_VARIABLES,scope='autoencoder/encoder') 427 | 428 | #set optimizer for each module 429 | disc_op=tf.train.AdamOptimizer(learning_rate=critic_lr) 430 | gen_op=tf.train.AdamOptimizer(learning_rate=gen_lr) 431 | AE_op=tf.train.GradientDescentOptimizer(learning_rate=AE_learning_rate) 432 | 433 | 434 | #compute gradients 435 | pos_critic_grad=disc_op.compute_gradients(disc_real_loss,critic_variables+encoder_variables) 436 | 437 | neg_critic_grad=disc_op.compute_gradients(disc_fake_loss,critic_variables) 438 | 439 | #clipping gradients for negative samples 440 | neg_critic_grad = [(tf.clip_by_value(grad, -1., 1.), var) for grad, var in neg_critic_grad] 441 | gen_grad = gen_op.compute_gradients(gen_loss, gen_variables) 442 | AE_grad = AE_op.compute_gradients(AE_loss,AE_variables) 443 | #AE_grad = [(tf.clip_by_value(grad, -1., 1.), var) for grad, var in AE_grad] 444 | 445 | #apply gradients 446 | update_critic_pos=disc_op.apply_gradients(pos_critic_grad) 447 | update_critic_neg=disc_op.apply_gradients(neg_critic_grad) 448 | update_G=gen_op.apply_gradients(gen_grad) 449 | update_AE=AE_op.apply_gradients(AE_grad) 450 | 451 | #reconstruction 452 | with tf.variable_scope("autoencoder"): 453 | toutput_, tpredicting_logits_, ttraining_logits_, tmasks_, ttarget_=decoder(real_code,input_X_de,encoder_state,reuse=True) 454 | foutput_, fpredicting_logits_, ftraining_logits_, fmasks_, ftarget_ = decoder(g_code, input_X_de,g_state,reuse=True) 455 | def train(data): 456 | init=tf.global_variables_initializer() 457 | with tf.Session() as sess: 458 | sess.run(init) 459 | train_variables, test_variables=data 460 | encoder_train, decoder_train, train_batch_lenth, n_trainTime, n_trainDist1, n_trainDist2, z_train, z_train_time, z_train_dist1, z_train_dist2=train_variables 461 | 462 | max_F1=[] 463 | max_pair_F1=[] 464 | test_max_F1=[] 465 | test_max_pair_F1=[] 466 | ftest_max_F1=[] 467 | ftest_max_pair_F1=[] 468 | res = {} 469 | fres={} 470 | for epoch in range(train_iters): 471 | step = 0 472 | ACC=0 473 | F1=[] 474 | pairs_F1=[] 475 | LOSS=[] 476 | gen_LOSS=[] 477 | Critic_Loss=[] 478 | Fake_Loss=[] 479 | GAN_Loss=[] 480 | while step < len(encoder_train) // batch_size: 481 | start_i = step * batch_size 482 | dist_1_=n_trainDist1[start_i:start_i + batch_size] 483 | dist_2_=n_trainDist2[start_i:start_i + batch_size] 484 | input_time_=n_trainTime[start_i:start_i + batch_size] 485 | encode_batch =encoder_train[start_i:start_i + batch_size] 486 | decode_batch = decoder_train[start_i:start_i + batch_size] 487 | pad_source_lengths =train_batch_lenth[start_i:start_i + batch_size] 488 | z_in =z_train[start_i:start_i + batch_size] 489 | z_time=z_train_time[start_i:start_i + batch_size] 490 | z_dist1=z_train_dist1[start_i:start_i + batch_size] 491 | z_dist2=z_train_dist2[start_i:start_i + batch_size] 492 | 493 | #update AE 494 | _,_AEloss=sess.run([update_AE,AE_loss],feed_dict={target_sequence_length: pad_source_lengths, 495 | input_X: encode_batch,input_X_de: decode_batch,input_t:input_time_,input_d1:dist_1_,input_d2:dist_2_,keep_prob: 0.5}) 496 | # update critic & encoder at positive sample phase 497 | for k in range(10): 498 | _, _critic_loss, real_loss = sess.run([update_critic_pos,critic_loss,disc_real_loss],feed_dict={target_sequence_length: pad_source_lengths, 499 | input_X: encode_batch,input_X_de: decode_batch,z: z_in,z_t:z_time,z_d1:z_dist1,z_d2:z_dist2,input_t:input_time_,input_d1:dist_1_,input_d2:dist_2_,keep_prob: 0.5}) 500 | fake_loss, _ = sess.run([disc_fake_loss, update_critic_neg], feed_dict={z: z_in, z_t:z_time,z_d1:z_dist1,z_d2:z_dist2,keep_prob: 0.5}) 501 | # update generate 502 | _, gan_loss = sess.run([update_G, gen_loss], feed_dict={z: z_in, z_t:z_time,z_d1:z_dist1,z_d2:z_dist2,keep_prob: 0.5}) 503 | #training result 504 | values= sess.run(predicting_logits_, feed_dict={target_sequence_length: pad_source_lengths, 505 | input_X: encode_batch,input_X_de: decode_batch,input_t:input_time_,input_d1:dist_1_,input_d2:dist_2_, 506 | z_t: z_time, z_d1: z_dist1, z_d2: z_dist2,keep_prob: 0.5}) 507 | LOSS.append(_AEloss) 508 | gen_LOSS.append(real_loss) 509 | Critic_Loss.append(_critic_loss) 510 | Fake_Loss.append(fake_loss) 511 | GAN_Loss.append(gan_loss) 512 | 513 | for v in range(len(values)): 514 | length=pad_source_lengths[v]-1 515 | if (decode_batch[v][:length]==values[v][:length]).all(): 516 | ACC+=1 517 | f=calc_F1(decode_batch[v][:length],values[v][:length]) 518 | p_f=calc_pairsF1(decode_batch[v][:length],values[v][:length]) 519 | F1.append(f) 520 | pairs_F1.append(p_f) 521 | step += 1 522 | #saver.save(sess, model) 523 | #print 'train F1',np.mean(F1) 524 | #print 'pairs-F1',np.mean(pairs_F1) 525 | #print 'epoch',epoch,'train accuracy',ACC/len(trainU) 526 | #print 'testing------------------>loss',epoch,np.sum(LOSS) 527 | test_f1,test_pairs_f1,ftest_f1,ftest_pairs_f1=test(sess=sess,test_variables=test_variables) # 528 | max_F1.append(np.mean(F1)) 529 | max_pair_F1.append(np.mean(pairs_F1)) 530 | #print test_f1,test_pairs_f1,ftest_f1,ftest_pairs_f1 531 | max_epoch = -1 532 | res.setdefault(test_f1, []).append(test_pairs_f1) 533 | fres.setdefault(ftest_f1, []).append(ftest_pairs_f1) 534 | keys = res.keys() 535 | fkeys =fres.keys() 536 | keys = sorted(keys) 537 | fkeys = sorted(fkeys) 538 | return (max_F1[max_epoch]), (max_pair_F1)[max_epoch],keys[-1],max(res[keys[-1]]),fkeys[-1],max(fres[fkeys[-1]]) 539 | 540 | def test(sess,test_variables):#,Test_Time=None 541 | step = 0 542 | Pred_ACC = 0 543 | F1 = [] 544 | pairs_F1 = [] 545 | fake_F1=[] 546 | fake_pairs_F1=[] 547 | encoder_test, decoder_test, test_batch_lenth, n_testTime, n_testDist1, n_testDist2, z_test, z_test_time, z_test_dist1, z_test_dist2 = test_variables 548 | while step < len(encoder_test) // batch_size: 549 | start_i = step * batch_size 550 | dist_1_ = n_testDist1[start_i:start_i + batch_size] 551 | dist_2_ = n_testDist2[start_i:start_i + batch_size] 552 | input_time_ = n_testTime[start_i:start_i + batch_size] 553 | encode_batch = encoder_test[start_i:start_i + batch_size] 554 | decode_batch = decoder_test[start_i:start_i + batch_size] 555 | pad_source_lengths = test_batch_lenth[start_i:start_i + batch_size] 556 | z_in = z_test[start_i:start_i + batch_size] 557 | z_time = z_test_time[start_i:start_i + batch_size] 558 | z_dist1 = z_test_dist1[start_i:start_i + batch_size] 559 | z_dist2 = z_test_dist2[start_i:start_i + batch_size] 560 | otpredicting_logits_,fake_ = sess.run([tpredicting_logits_,fpredicting_logits_],feed_dict={target_sequence_length: pad_source_lengths, 561 | input_X: z_in,input_t:z_time,input_d1:z_dist1,input_d2:z_dist2,z:z_in,z_t:z_time,z_d1:z_dist1,z_d2:z_dist2, 562 | keep_prob: 1.0}) # 梯度更新后的结果 563 | 564 | for v in range(len(otpredicting_logits_[:1])): 565 | length = pad_source_lengths[v] - 1 566 | if (decode_batch[v][:length]== otpredicting_logits_[v][:length]).all(): 567 | Pred_ACC += 1 568 | actual = decode_batch[v][:length] 569 | recommend = np.concatenate([[actual[0]], otpredicting_logits_[v][1:length - 1]], axis=0) 570 | recommend = np.concatenate([recommend, [actual[-1]]], axis=0) 571 | 572 | frecommend = np.concatenate([[actual[0]], fake_[v][1:length - 1]], axis=0) 573 | frecommend = np.concatenate([frecommend, [actual[-1]]], axis=0) 574 | #print actual,recommend,frecommend 575 | f_f = calc_F1(actual, frecommend) 576 | f_p_f = calc_pairsF1(actual, frecommend) 577 | f = calc_F1(actual, recommend) 578 | p_f = calc_pairsF1(actual, recommend) 579 | F1.append(f) 580 | pairs_F1.append(p_f) 581 | fake_F1.append(f_f) 582 | fake_pairs_F1.append(f_p_f) 583 | step += 1 584 | #print 'trajectory length',length 585 | return np.mean(F1),np.mean(pairs_F1),np.mean(fake_F1),np.mean((fake_pairs_F1)) 586 | def get_data(index,K): 587 | # sort original data 588 | index_T = {} 589 | trainT = [] 590 | trainU = [] 591 | trainTime=[] 592 | trainDist=[] 593 | for i in range(len(new_trainTs)): 594 | index_T[i] = len(new_trainTs[i]) 595 | temp_size = sorted(index_T.items(), key=lambda item: item[1]) 596 | for i in range(len(temp_size)): 597 | id = temp_size[i][0] 598 | trainT.append(new_trainTs[id]) 599 | trainU.append(TRAIN_USER[id]) 600 | trainTime.append(TRAIN_TIME[id]) 601 | trainDist.append(TRAIN_DIST[id]) 602 | value=int(math.ceil(len(trainT)/K)) 603 | if index==K-1: 604 | testT=trainT[-value:] 605 | testU=trainU[-value:] 606 | trainT=trainT[:-value] 607 | trainU=trainU[:-value] 608 | 609 | testTime=trainTime[-value:] 610 | testDist=trainDist[-value:] 611 | trainTime=trainTime[:-value] 612 | trainDist=trainDist[:-value] 613 | 614 | elif index==0: 615 | testT=trainT[:(index+1)*value] 616 | testU=trainU[:(index+1)*value] 617 | trainT =trainT[(index+1)*value:] 618 | trainU =trainU[(index+1)*value:] 619 | 620 | testTime=trainTime[:(index+1)*value] 621 | testDist=trainDist[:(index+1)*value] 622 | trainTime=trainTime[(index+1)*value:] 623 | trainDist=trainDist[(index+1)*value:] 624 | 625 | else: 626 | testT=trainT[index*value:(index+1)*value] 627 | testU=trainU[index*value:(index+1)*value] 628 | trainT = trainT[0:index*value]+trainT[(index+1)*value:] 629 | trainU = trainU[0:index*value]+trainU[(index+1)*value:] 630 | 631 | testTime=trainTime[index*value:(index+1)*value] 632 | testDist=trainDist[index*value:(index+1)*value] 633 | trainTime=trainTime[0:index*value]+trainTime[(index+1)*value:] 634 | trainDist=trainDist[0:index*value]+trainDist[(index+1)*value:] 635 | train_size = len(trainT) % batch_size 636 | #if 637 | trainT = trainT + [trainT[-1]]*(batch_size-train_size) # copy data and fill the last batch size 638 | trainU = trainU + [trainU[-1]]*(batch_size-train_size) 639 | trainTime=trainTime+[trainTime[-1]]*(batch_size-train_size) 640 | trainDist = trainDist + [trainDist[-1]] * (batch_size - train_size) 641 | #print 'Text', testT,index,K 642 | test_size = len(testT) % batch_size 643 | if test_size!=0: 644 | testT = testT + [testT[-1]]*(batch_size-test_size) # copy data and fill the last batch size 645 | testU = testU + [testU[-1]]*(batch_size-test_size) #BUG for test_size