├── .gitignore
├── README.md
├── __init__.py
├── data
└── pull_data.sh
├── data_utils.py
├── env_setup.sh
├── execute.py
├── neuralconvo.ini
├── seq2seq.ini
├── seq2seq_model.py
├── seq2seq_serve.ini
├── ui
├── .gitignore
├── README.md
├── __init.py__
├── app.py
├── requirements.txt
├── setup.sh
├── static
│ ├── css
│ │ ├── normalize.css
│ │ └── style.css
│ ├── js
│ │ ├── index.js
│ │ ├── jquery-latest.js
│ │ ├── jquery.mCustomScrollbar.concat.min.js
│ │ ├── jquery.mCustomScrollbar.min.css
│ │ └── jquery.min.js
│ ├── res
│ │ ├── botim.png
│ │ ├── hd1.jpg
│ │ └── hd2.jpg
│ └── scss
│ │ └── style.scss
└── templates
│ └── index.html
└── working_dir
├── vocab20000.dec
└── vocab20000.enc
/.gitignore:
--------------------------------------------------------------------------------
1 | data/
2 | *.pyc
3 |
--------------------------------------------------------------------------------
/README.md:
--------------------------------------------------------------------------------
1 | # easy\_seq2seq
2 |
3 | 
4 |
5 | * Go to [*practical_seq2seq*](https://github.com/suriyadeepan/practical_seq2seq)
6 |
7 | An implementation of Seq2Seq that actually works. I want to make it easy for people to train their own seq2seq model with any corpus. I am also adding the parameters of my trained model for people to just use it without training. If you have a model that works share your model params here, as external link or do a pull request. I have used Cornell Movie Dialog Corpus to train my model. A link to preprocessed data and scripts for preprocessing can be found in this repo.
8 |
9 | *Have Fun!*
10 |
11 | ## Update 1.1.2017
12 |
13 |
14 | I have created another repository - [practical_seq2seq](https://github.com/suriyadeepan/practical_seq2seq) to experiment with the seq2seq model. The new model trained on Twitter chat log and Cornell Movie Dialog corpus performs well. I wrote an article - [Practical seq2seq](http://suriyadeepan.github.io/2016-12-31-practical-seq2seq/), explaining the code.
15 |
16 | Happy New Year, **2017**
17 |
18 |
19 | ## Setup
20 |
21 |
22 | * Create temporary working directory prior to training
23 |
24 | ```bash
25 | mkdir working_dir
26 | ```
27 |
28 | * Download test/train data from Cornell Movie Dialog Corpus
29 |
30 | ```bash
31 | cd data/
32 | bash pull_data.sh
33 | ```
34 |
35 | ## Training
36 |
37 | ```bash
38 | # edit seq2seq.ini file to set
39 | # mode = train
40 | python execute.py
41 | # or use custom ini file
42 | # python execute.py my_custom_conf.ini
43 | ```
44 |
45 | ## Testing
46 |
47 | ```bash
48 | # edit seq2seq.ini file to set
49 | # mode = test
50 | python execute.py
51 | ```
52 |
53 | ## Serve
54 |
55 | ```bash
56 | # configuration : seq2seq_serve.ini
57 | python ui/app.py
58 | # wait until this message shows up
59 | # "Running on http://127.0.0.1:5000/ (Press CTRL+C to quit)"
60 | # open up the address in browser, chat with the bot
61 | ```
62 |
--------------------------------------------------------------------------------
/__init__.py:
--------------------------------------------------------------------------------
https://raw.githubusercontent.com/suriyadeepan/easy_seq2seq/c05b0d02f15e9bd1947ff79ddcd0b761a214e2a9/__init__.py
--------------------------------------------------------------------------------
/data/pull_data.sh:
--------------------------------------------------------------------------------
1 | wget -c 'https://www.dropbox.com/s/ncfa5t950gvtaeb/test.enc?dl=0' -O test.enc
2 | wget -c 'https://www.dropbox.com/s/48ro4759jaikque/test.dec?dl=0' -O test.dec
3 | wget -c 'https://www.dropbox.com/s/gu54ngk3xpwite4/train.enc?dl=0' -O train.enc
4 | wget -c 'https://www.dropbox.com/s/g3z2msjziqocndl/train.dec?dl=0' -O train.dec
5 |
--------------------------------------------------------------------------------
/data_utils.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 | """Utilities for downloading data from WMT, tokenizing, vocabularies."""
17 | from __future__ import absolute_import
18 | from __future__ import division
19 | from __future__ import print_function
20 |
21 | import os
22 | import re
23 |
24 | from six.moves import urllib
25 |
26 | from tensorflow.python.platform import gfile
27 |
28 | # Special vocabulary symbols - we always put them at the start.
29 | _PAD = b"_PAD"
30 | _GO = b"_GO"
31 | _EOS = b"_EOS"
32 | _UNK = b"_UNK"
33 | _START_VOCAB = [_PAD, _GO, _EOS, _UNK]
34 |
35 | PAD_ID = 0
36 | GO_ID = 1
37 | EOS_ID = 2
38 | UNK_ID = 3
39 |
40 | # Regular expressions used to tokenize.
41 | _WORD_SPLIT = re.compile(b"([.,!?\"':;)(])")
42 | _DIGIT_RE = re.compile(br"\d")
43 |
44 |
45 |
46 | def basic_tokenizer(sentence):
47 | """Very basic tokenizer: split the sentence into a list of tokens."""
48 | words = []
49 | for space_separated_fragment in sentence.strip().split():
50 | if isinstance(space_separated_fragment, str):
51 | word = str.encode(space_separated_fragment)
52 | else:
53 | word = space_separated_fragment
54 | words.extend(re.split(_WORD_SPLIT, word))
55 | return [w for w in words if w]
56 |
57 |
58 | def create_vocabulary(vocabulary_path, data_path, max_vocabulary_size,
59 | tokenizer=None, normalize_digits=True):
60 |
61 | if not gfile.Exists(vocabulary_path):
62 | print("Creating vocabulary %s from %s" % (vocabulary_path, data_path))
63 | vocab = {}
64 | with gfile.GFile(data_path, mode="rb") as f:
65 | counter = 0
66 | for line in f:
67 | counter += 1
68 | if counter % 100000 == 0:
69 | print(" processing line %d" % counter)
70 | tokens = tokenizer(line) if tokenizer else basic_tokenizer(line)
71 | for w in tokens:
72 | word = re.sub(_DIGIT_RE, b"0", w) if normalize_digits else w
73 | if word in vocab:
74 | vocab[word] += 1
75 | else:
76 | vocab[word] = 1
77 | vocab_list = _START_VOCAB + sorted(vocab, key=vocab.get, reverse=True)
78 | print('>> Full Vocabulary Size :',len(vocab_list))
79 | if len(vocab_list) > max_vocabulary_size:
80 | vocab_list = vocab_list[:max_vocabulary_size]
81 | with gfile.GFile(vocabulary_path, mode="wb") as vocab_file:
82 | for w in vocab_list:
83 | vocab_file.write(w + b"\n")
84 |
85 |
86 | def initialize_vocabulary(vocabulary_path):
87 |
88 | if gfile.Exists(vocabulary_path):
89 | rev_vocab = []
90 | with gfile.GFile(vocabulary_path, mode="rb") as f:
91 | rev_vocab.extend(f.readlines())
92 | rev_vocab = [line.strip() for line in rev_vocab]
93 | vocab = dict([(x, y) for (y, x) in enumerate(rev_vocab)])
94 | return vocab, rev_vocab
95 | else:
96 | raise ValueError("Vocabulary file %s not found.", vocabulary_path)
97 |
98 |
99 | def sentence_to_token_ids(sentence, vocabulary, tokenizer=None, normalize_digits=True):
100 |
101 | if tokenizer:
102 | words = tokenizer(sentence)
103 | else:
104 | words = basic_tokenizer(sentence)
105 | if not normalize_digits:
106 | return [vocabulary.get(w, UNK_ID) for w in words]
107 | # Normalize digits by 0 before looking words up in the vocabulary.
108 | return [vocabulary.get(re.sub(_DIGIT_RE, b"0", w), UNK_ID) for w in words]
109 |
110 |
111 | def data_to_token_ids(data_path, target_path, vocabulary_path,
112 | tokenizer=None, normalize_digits=True):
113 |
114 | if not gfile.Exists(target_path):
115 | print("Tokenizing data in %s" % data_path)
116 | vocab, _ = initialize_vocabulary(vocabulary_path)
117 | with gfile.GFile(data_path, mode="rb") as data_file:
118 | with gfile.GFile(target_path, mode="w") as tokens_file:
119 | counter = 0
120 | for line in data_file:
121 | counter += 1
122 | if counter % 100000 == 0:
123 | print(" tokenizing line %d" % counter)
124 | token_ids = sentence_to_token_ids(line, vocab, tokenizer,
125 | normalize_digits)
126 | tokens_file.write(" ".join([str(tok) for tok in token_ids]) + "\n")
127 |
128 |
129 |
130 | def prepare_custom_data(working_directory, train_enc, train_dec, test_enc, test_dec, enc_vocabulary_size, dec_vocabulary_size, tokenizer=None):
131 |
132 | # Create vocabularies of the appropriate sizes.
133 | enc_vocab_path = os.path.join(working_directory, "vocab%d.enc" % enc_vocabulary_size)
134 | dec_vocab_path = os.path.join(working_directory, "vocab%d.dec" % dec_vocabulary_size)
135 | create_vocabulary(enc_vocab_path, train_enc, enc_vocabulary_size, tokenizer)
136 | create_vocabulary(dec_vocab_path, train_dec, dec_vocabulary_size, tokenizer)
137 |
138 | # Create token ids for the training data.
139 | enc_train_ids_path = train_enc + (".ids%d" % enc_vocabulary_size)
140 | dec_train_ids_path = train_dec + (".ids%d" % dec_vocabulary_size)
141 | data_to_token_ids(train_enc, enc_train_ids_path, enc_vocab_path, tokenizer)
142 | data_to_token_ids(train_dec, dec_train_ids_path, dec_vocab_path, tokenizer)
143 |
144 | # Create token ids for the development data.
145 | enc_dev_ids_path = test_enc + (".ids%d" % enc_vocabulary_size)
146 | dec_dev_ids_path = test_dec + (".ids%d" % dec_vocabulary_size)
147 | data_to_token_ids(test_enc, enc_dev_ids_path, enc_vocab_path, tokenizer)
148 | data_to_token_ids(test_dec, dec_dev_ids_path, dec_vocab_path, tokenizer)
149 |
150 | return (enc_train_ids_path, dec_train_ids_path, enc_dev_ids_path, dec_dev_ids_path, enc_vocab_path, dec_vocab_path)
151 |
--------------------------------------------------------------------------------
/env_setup.sh:
--------------------------------------------------------------------------------
1 | echo 'export PYTHONPATH=$PYTHONPATH:$PWD' >> ~/.bashrc
2 |
--------------------------------------------------------------------------------
/execute.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 | from __future__ import absolute_import
17 | from __future__ import division
18 | from __future__ import print_function
19 |
20 | import math
21 | import os
22 | import random
23 | import sys
24 | import time
25 |
26 | import numpy as np
27 | from six.moves import xrange # pylint: disable=redefined-builtin
28 | import tensorflow as tf
29 |
30 | import data_utils
31 | import seq2seq_model
32 |
33 | try:
34 | from ConfigParser import SafeConfigParser
35 | except:
36 | from configparser import SafeConfigParser # In Python 3, ConfigParser has been renamed to configparser for PEP 8 compliance.
37 |
38 | gConfig = {}
39 |
40 | def get_config(config_file='seq2seq.ini'):
41 | parser = SafeConfigParser()
42 | parser.read(config_file)
43 | # get the ints, floats and strings
44 | _conf_ints = [ (key, int(value)) for key,value in parser.items('ints') ]
45 | _conf_floats = [ (key, float(value)) for key,value in parser.items('floats') ]
46 | _conf_strings = [ (key, str(value)) for key,value in parser.items('strings') ]
47 | return dict(_conf_ints + _conf_floats + _conf_strings)
48 |
49 | # We use a number of buckets and pad to the closest one for efficiency.
50 | # See seq2seq_model.Seq2SeqModel for details of how they work.
51 | _buckets = [(5, 10), (10, 15), (20, 25), (40, 50)]
52 |
53 |
54 | def read_data(source_path, target_path, max_size=None):
55 | """Read data from source and target files and put into buckets.
56 |
57 | Args:
58 | source_path: path to the files with token-ids for the source language.
59 | target_path: path to the file with token-ids for the target language;
60 | it must be aligned with the source file: n-th line contains the desired
61 | output for n-th line from the source_path.
62 | max_size: maximum number of lines to read, all other will be ignored;
63 | if 0 or None, data files will be read completely (no limit).
64 |
65 | Returns:
66 | data_set: a list of length len(_buckets); data_set[n] contains a list of
67 | (source, target) pairs read from the provided data files that fit
68 | into the n-th bucket, i.e., such that len(source) < _buckets[n][0] and
69 | len(target) < _buckets[n][1]; source and target are lists of token-ids.
70 | """
71 | data_set = [[] for _ in _buckets]
72 | with tf.gfile.GFile(source_path, mode="r") as source_file:
73 | with tf.gfile.GFile(target_path, mode="r") as target_file:
74 | source, target = source_file.readline(), target_file.readline()
75 | counter = 0
76 | while source and target and (not max_size or counter < max_size):
77 | counter += 1
78 | if counter % 100000 == 0:
79 | print(" reading data line %d" % counter)
80 | sys.stdout.flush()
81 | source_ids = [int(x) for x in source.split()]
82 | target_ids = [int(x) for x in target.split()]
83 | target_ids.append(data_utils.EOS_ID)
84 | for bucket_id, (source_size, target_size) in enumerate(_buckets):
85 | if len(source_ids) < source_size and len(target_ids) < target_size:
86 | data_set[bucket_id].append([source_ids, target_ids])
87 | break
88 | source, target = source_file.readline(), target_file.readline()
89 | return data_set
90 |
91 |
92 | def create_model(session, forward_only):
93 |
94 | """Create model and initialize or load parameters"""
95 | model = seq2seq_model.Seq2SeqModel( gConfig['enc_vocab_size'], gConfig['dec_vocab_size'], _buckets, gConfig['layer_size'], gConfig['num_layers'], gConfig['max_gradient_norm'], gConfig['batch_size'], gConfig['learning_rate'], gConfig['learning_rate_decay_factor'], forward_only=forward_only)
96 |
97 | if 'pretrained_model' in gConfig:
98 | model.saver.restore(session,gConfig['pretrained_model'])
99 | return model
100 |
101 | ckpt = tf.train.get_checkpoint_state(gConfig['working_directory'])
102 | if ckpt and ckpt.model_checkpoint_path:
103 | print("Reading model parameters from %s" % ckpt.model_checkpoint_path)
104 | model.saver.restore(session, ckpt.model_checkpoint_path)
105 | else:
106 | print("Created model with fresh parameters.")
107 | session.run(tf.global_variables_initializer())
108 | return model
109 |
110 |
111 | def train():
112 | # prepare dataset
113 | print("Preparing data in %s" % gConfig['working_directory'])
114 | enc_train, dec_train, enc_dev, dec_dev, _, _ = data_utils.prepare_custom_data(gConfig['working_directory'],gConfig['train_enc'],gConfig['train_dec'],gConfig['test_enc'],gConfig['test_dec'],gConfig['enc_vocab_size'],gConfig['dec_vocab_size'])
115 |
116 | # setup config to use BFC allocator
117 | config = tf.ConfigProto()
118 | config.gpu_options.allocator_type = 'BFC'
119 |
120 | with tf.Session(config=config) as sess:
121 | # Create model.
122 | print("Creating %d layers of %d units." % (gConfig['num_layers'], gConfig['layer_size']))
123 | model = create_model(sess, False)
124 |
125 | # Read data into buckets and compute their sizes.
126 | print ("Reading development and training data (limit: %d)."
127 | % gConfig['max_train_data_size'])
128 | dev_set = read_data(enc_dev, dec_dev)
129 | train_set = read_data(enc_train, dec_train, gConfig['max_train_data_size'])
130 | train_bucket_sizes = [len(train_set[b]) for b in xrange(len(_buckets))]
131 | train_total_size = float(sum(train_bucket_sizes))
132 |
133 | # A bucket scale is a list of increasing numbers from 0 to 1 that we'll use
134 | # to select a bucket. Length of [scale[i], scale[i+1]] is proportional to
135 | # the size if i-th training bucket, as used later.
136 | train_buckets_scale = [sum(train_bucket_sizes[:i + 1]) / train_total_size
137 | for i in xrange(len(train_bucket_sizes))]
138 |
139 | # This is the training loop.
140 | step_time, loss = 0.0, 0.0
141 | current_step = 0
142 | previous_losses = []
143 | while True:
144 | # Choose a bucket according to data distribution. We pick a random number
145 | # in [0, 1] and use the corresponding interval in train_buckets_scale.
146 | random_number_01 = np.random.random_sample()
147 | bucket_id = min([i for i in xrange(len(train_buckets_scale))
148 | if train_buckets_scale[i] > random_number_01])
149 |
150 | # Get a batch and make a step.
151 | start_time = time.time()
152 | encoder_inputs, decoder_inputs, target_weights = model.get_batch(
153 | train_set, bucket_id)
154 | _, step_loss, _ = model.step(sess, encoder_inputs, decoder_inputs,
155 | target_weights, bucket_id, False)
156 | step_time += (time.time() - start_time) / gConfig['steps_per_checkpoint']
157 | loss += step_loss / gConfig['steps_per_checkpoint']
158 | current_step += 1
159 |
160 | # Once in a while, we save checkpoint, print statistics, and run evals.
161 | if current_step % gConfig['steps_per_checkpoint'] == 0:
162 | # Print statistics for the previous epoch.
163 | perplexity = math.exp(loss) if loss < 300 else float('inf')
164 | print ("global step %d learning rate %.4f step-time %.2f perplexity "
165 | "%.2f" % (model.global_step.eval(), model.learning_rate.eval(),
166 | step_time, perplexity))
167 | # Decrease learning rate if no improvement was seen over last 3 times.
168 | if len(previous_losses) > 2 and loss > max(previous_losses[-3:]):
169 | sess.run(model.learning_rate_decay_op)
170 | previous_losses.append(loss)
171 | # Save checkpoint and zero timer and loss.
172 | checkpoint_path = os.path.join(gConfig['working_directory'], "seq2seq.ckpt")
173 | model.saver.save(sess, checkpoint_path, global_step=model.global_step)
174 | step_time, loss = 0.0, 0.0
175 | # Run evals on development set and print their perplexity.
176 | for bucket_id in xrange(len(_buckets)):
177 | if len(dev_set[bucket_id]) == 0:
178 | print(" eval: empty bucket %d" % (bucket_id))
179 | continue
180 | encoder_inputs, decoder_inputs, target_weights = model.get_batch(
181 | dev_set, bucket_id)
182 | _, eval_loss, _ = model.step(sess, encoder_inputs, decoder_inputs,
183 | target_weights, bucket_id, True)
184 | eval_ppx = math.exp(eval_loss) if eval_loss < 300 else float('inf')
185 | print(" eval: bucket %d perplexity %.2f" % (bucket_id, eval_ppx))
186 | sys.stdout.flush()
187 |
188 |
189 | def decode():
190 | with tf.Session() as sess:
191 | # Create model and load parameters.
192 | model = create_model(sess, True)
193 | model.batch_size = 1 # We decode one sentence at a time.
194 |
195 | # Load vocabularies.
196 | enc_vocab_path = os.path.join(gConfig['working_directory'],"vocab%d.enc" % gConfig['enc_vocab_size'])
197 | dec_vocab_path = os.path.join(gConfig['working_directory'],"vocab%d.dec" % gConfig['dec_vocab_size'])
198 |
199 | enc_vocab, _ = data_utils.initialize_vocabulary(enc_vocab_path)
200 | _, rev_dec_vocab = data_utils.initialize_vocabulary(dec_vocab_path)
201 |
202 | # Decode from standard input.
203 | sys.stdout.write("> ")
204 | sys.stdout.flush()
205 | sentence = sys.stdin.readline()
206 | while sentence:
207 | # Get token-ids for the input sentence.
208 | token_ids = data_utils.sentence_to_token_ids(tf.compat.as_bytes(sentence), enc_vocab)
209 | # Which bucket does it belong to?
210 | bucket_id = min([b for b in xrange(len(_buckets))
211 | if _buckets[b][0] > len(token_ids)])
212 | # Get a 1-element batch to feed the sentence to the model.
213 | encoder_inputs, decoder_inputs, target_weights = model.get_batch(
214 | {bucket_id: [(token_ids, [])]}, bucket_id)
215 | # Get output logits for the sentence.
216 | _, _, output_logits = model.step(sess, encoder_inputs, decoder_inputs,
217 | target_weights, bucket_id, True)
218 | # This is a greedy decoder - outputs are just argmaxes of output_logits.
219 | outputs = [int(np.argmax(logit, axis=1)) for logit in output_logits]
220 | # If there is an EOS symbol in outputs, cut them at that point.
221 | if data_utils.EOS_ID in outputs:
222 | outputs = outputs[:outputs.index(data_utils.EOS_ID)]
223 | # Print out French sentence corresponding to outputs.
224 | print(" ".join([tf.compat.as_str(rev_dec_vocab[output]) for output in outputs]))
225 | print("> ", end="")
226 | sys.stdout.flush()
227 | sentence = sys.stdin.readline()
228 |
229 |
230 | def self_test():
231 | """Test the translation model."""
232 | with tf.Session() as sess:
233 | print("Self-test for neural translation model.")
234 | # Create model with vocabularies of 10, 2 small buckets, 2 layers of 32.
235 | model = seq2seq_model.Seq2SeqModel(10, 10, [(3, 3), (6, 6)], 32, 2,
236 | 5.0, 32, 0.3, 0.99, num_samples=8)
237 | sess.run(tf.initialize_all_variables())
238 |
239 | # Fake data set for both the (3, 3) and (6, 6) bucket.
240 | data_set = ([([1, 1], [2, 2]), ([3, 3], [4]), ([5], [6])],
241 | [([1, 1, 1, 1, 1], [2, 2, 2, 2, 2]), ([3, 3, 3], [5, 6])])
242 | for _ in xrange(5): # Train the fake model for 5 steps.
243 | bucket_id = random.choice([0, 1])
244 | encoder_inputs, decoder_inputs, target_weights = model.get_batch(
245 | data_set, bucket_id)
246 | model.step(sess, encoder_inputs, decoder_inputs, target_weights,
247 | bucket_id, False)
248 |
249 |
250 | def init_session(sess, conf='seq2seq.ini'):
251 | global gConfig
252 | gConfig = get_config(conf)
253 |
254 | # Create model and load parameters.
255 | model = create_model(sess, True)
256 | model.batch_size = 1 # We decode one sentence at a time.
257 |
258 | # Load vocabularies.
259 | enc_vocab_path = os.path.join(gConfig['working_directory'],"vocab%d.enc" % gConfig['enc_vocab_size'])
260 | dec_vocab_path = os.path.join(gConfig['working_directory'],"vocab%d.dec" % gConfig['dec_vocab_size'])
261 |
262 | enc_vocab, _ = data_utils.initialize_vocabulary(enc_vocab_path)
263 | _, rev_dec_vocab = data_utils.initialize_vocabulary(dec_vocab_path)
264 |
265 | return sess, model, enc_vocab, rev_dec_vocab
266 |
267 | def decode_line(sess, model, enc_vocab, rev_dec_vocab, sentence):
268 | # Get token-ids for the input sentence.
269 | token_ids = data_utils.sentence_to_token_ids(tf.compat.as_bytes(sentence), enc_vocab)
270 |
271 | # Which bucket does it belong to?
272 | bucket_id = min([b for b in xrange(len(_buckets)) if _buckets[b][0] > len(token_ids)])
273 |
274 | # Get a 1-element batch to feed the sentence to the model.
275 | encoder_inputs, decoder_inputs, target_weights = model.get_batch({bucket_id: [(token_ids, [])]}, bucket_id)
276 |
277 | # Get output logits for the sentence.
278 | _, _, output_logits = model.step(sess, encoder_inputs, decoder_inputs, target_weights, bucket_id, True)
279 |
280 | # This is a greedy decoder - outputs are just argmaxes of output_logits.
281 | outputs = [int(np.argmax(logit, axis=1)) for logit in output_logits]
282 |
283 | # If there is an EOS symbol in outputs, cut them at that point.
284 | if data_utils.EOS_ID in outputs:
285 | outputs = outputs[:outputs.index(data_utils.EOS_ID)]
286 |
287 | return " ".join([tf.compat.as_str(rev_dec_vocab[output]) for output in outputs])
288 |
289 | if __name__ == '__main__':
290 | if len(sys.argv) - 1:
291 | gConfig = get_config(sys.argv[1])
292 | else:
293 | # get configuration from seq2seq.ini
294 | gConfig = get_config()
295 |
296 | print('\n>> Mode : %s\n' %(gConfig['mode']))
297 |
298 | if gConfig['mode'] == 'train':
299 | # start training
300 | train()
301 | elif gConfig['mode'] == 'test':
302 | # interactive decode
303 | decode()
304 | else:
305 | # wrong way to execute "serve"
306 | # Use : >> python ui/app.py
307 | # uses seq2seq_serve.ini as conf file
308 | print('Serve Usage : >> python ui/app.py')
309 | print('# uses seq2seq_serve.ini as conf file')
310 |
--------------------------------------------------------------------------------
/neuralconvo.ini:
--------------------------------------------------------------------------------
1 | [strings]
2 | # Mode : train, test, serve
3 | mode = train
4 | train_enc = data/train.enc
5 | train_dec = data/train.dec
6 | test_enc = data/test.enc
7 | test_dec = data/test.enc
8 | # folder where checkpoints, vocabulary, temporary data will be stored
9 | working_directory = working_dir/
10 | [ints]
11 | # vocabulary size
12 | # 20,000 is a reasonable size
13 | enc_vocab_size = 45000
14 | dec_vocab_size = 45000
15 | # number of LSTM layers : 1/2/3
16 | num_layers = 2
17 | # typical options : 128, 256, 512, 1024
18 | layer_size = 900
19 | # dataset size limit; typically none : no limit
20 | max_train_data_size = 50000
21 | batch_size = 10
22 | # steps per checkpoint
23 | # Note : At a checkpoint, models parameters are saved, model is evaluated
24 | # and results are printed
25 | steps_per_checkpoint = 200
26 | [floats]
27 | learning_rate = 0.001
28 | learning_rate_decay_factor = 0.9
29 | max_gradient_norm = 5.0
30 | ##############################################################################
31 | # Note : Edit the bucket sizes at line47 of execute.py (_buckets)
32 | #
33 | # Learn more about the configurations from this link
34 | # https://www.tensorflow.org/versions/r0.9/tutorials/seq2seq/index.html
35 | ##############################################################################
36 |
--------------------------------------------------------------------------------
/seq2seq.ini:
--------------------------------------------------------------------------------
1 | [strings]
2 | # Mode : train, test, serve
3 | mode = train
4 | train_enc = data/train.enc
5 | train_dec = data/train.dec
6 | test_enc = data/test.enc
7 | test_dec = data/test.dec
8 | # folder where checkpoints, vocabulary, temporary data will be stored
9 | working_directory = working_dir/
10 | [ints]
11 | # vocabulary size
12 | # 20,000 is a reasonable size
13 | enc_vocab_size = 20000
14 | dec_vocab_size = 20000
15 | # number of LSTM layers : 1/2/3
16 | num_layers = 3
17 | # typical options : 128, 256, 512, 1024
18 | layer_size = 256
19 | # dataset size limit; typically none : no limit
20 | max_train_data_size = 0
21 | batch_size = 64
22 | # steps per checkpoint
23 | # Note : At a checkpoint, models parameters are saved, model is evaluated
24 | # and results are printed
25 | steps_per_checkpoint = 300
26 | [floats]
27 | learning_rate = 0.5
28 | learning_rate_decay_factor = 0.99
29 | max_gradient_norm = 5.0
30 | ##############################################################################
31 | # Note : Edit the bucket sizes at line47 of execute.py (_buckets)
32 | #
33 | # Learn more about the configurations from this link
34 | # https://www.tensorflow.org/versions/r0.9/tutorials/seq2seq/index.html
35 | ##############################################################################
36 |
--------------------------------------------------------------------------------
/seq2seq_model.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 | """Sequence-to-sequence model with an attention mechanism."""
17 |
18 | from __future__ import absolute_import
19 | from __future__ import division
20 | from __future__ import print_function
21 |
22 | import random
23 |
24 | import numpy as np
25 | from six.moves import xrange # pylint: disable=redefined-builtin
26 | import tensorflow as tf
27 |
28 | from tensorflow.models.rnn.translate import data_utils
29 |
30 |
31 | class Seq2SeqModel(object):
32 | """Sequence-to-sequence model with attention and for multiple buckets.
33 |
34 | This class implements a multi-layer recurrent neural network as encoder,
35 | and an attention-based decoder. This is the same as the model described in
36 | this paper: http://arxiv.org/abs/1412.7449 - please look there for details,
37 | or into the seq2seq library for complete model implementation.
38 | This class also allows to use GRU cells in addition to LSTM cells, and
39 | sampled softmax to handle large output vocabulary size. A single-layer
40 | version of this model, but with bi-directional encoder, was presented in
41 | http://arxiv.org/abs/1409.0473
42 | and sampled softmax is described in Section 3 of the following paper.
43 | http://arxiv.org/abs/1412.2007
44 | """
45 |
46 | def __init__(self, source_vocab_size, target_vocab_size, buckets, size,
47 | num_layers, max_gradient_norm, batch_size, learning_rate,
48 | learning_rate_decay_factor, use_lstm=False,
49 | num_samples=512, forward_only=False):
50 | """Create the model.
51 |
52 | Args:
53 | source_vocab_size: size of the source vocabulary.
54 | target_vocab_size: size of the target vocabulary.
55 | buckets: a list of pairs (I, O), where I specifies maximum input length
56 | that will be processed in that bucket, and O specifies maximum output
57 | length. Training instances that have inputs longer than I or outputs
58 | longer than O will be pushed to the next bucket and padded accordingly.
59 | We assume that the list is sorted, e.g., [(2, 4), (8, 16)].
60 | size: number of units in each layer of the model.
61 | num_layers: number of layers in the model.
62 | max_gradient_norm: gradients will be clipped to maximally this norm.
63 | batch_size: the size of the batches used during training;
64 | the model construction is independent of batch_size, so it can be
65 | changed after initialization if this is convenient, e.g., for decoding.
66 | learning_rate: learning rate to start with.
67 | learning_rate_decay_factor: decay learning rate by this much when needed.
68 | use_lstm: if true, we use LSTM cells instead of GRU cells.
69 | num_samples: number of samples for sampled softmax.
70 | forward_only: if set, we do not construct the backward pass in the model.
71 | """
72 | self.source_vocab_size = source_vocab_size
73 | self.target_vocab_size = target_vocab_size
74 | self.buckets = buckets
75 | self.batch_size = batch_size
76 | self.learning_rate = tf.Variable(float(learning_rate), trainable=False)
77 | self.learning_rate_decay_op = self.learning_rate.assign(
78 | self.learning_rate * learning_rate_decay_factor)
79 | self.global_step = tf.Variable(0, trainable=False)
80 |
81 | # If we use sampled softmax, we need an output projection.
82 | output_projection = None
83 | softmax_loss_function = None
84 | # Sampled softmax only makes sense if we sample less than vocabulary size.
85 | if num_samples > 0 and num_samples < self.target_vocab_size:
86 | w = tf.get_variable("proj_w", [size, self.target_vocab_size])
87 | w_t = tf.transpose(w)
88 | b = tf.get_variable("proj_b", [self.target_vocab_size])
89 | output_projection = (w, b)
90 |
91 | def sampled_loss(inputs, labels):
92 | labels = tf.reshape(labels, [-1, 1])
93 | return tf.nn.sampled_softmax_loss(w_t, b, inputs, labels, num_samples,
94 | self.target_vocab_size)
95 | softmax_loss_function = sampled_loss
96 |
97 | # Create the internal multi-layer cell for our RNN.
98 | single_cell = tf.nn.rnn_cell.GRUCell(size)
99 | if use_lstm:
100 | single_cell = tf.nn.rnn_cell.BasicLSTMCell(size)
101 | cell = single_cell
102 | cell = tf.nn.rnn_cell.DropoutWrapper(cell, output_keep_prob=0.5)
103 | if num_layers > 1:
104 | cell = tf.nn.rnn_cell.MultiRNNCell([single_cell] * num_layers)
105 |
106 | # The seq2seq function: we use embedding for the input and attention.
107 | def seq2seq_f(encoder_inputs, decoder_inputs, do_decode):
108 | return tf.nn.seq2seq.embedding_attention_seq2seq(
109 | encoder_inputs, decoder_inputs, cell,
110 | num_encoder_symbols=source_vocab_size,
111 | num_decoder_symbols=target_vocab_size,
112 | embedding_size=size,
113 | output_projection=output_projection,
114 | feed_previous=do_decode)
115 |
116 | # Feeds for inputs.
117 | self.encoder_inputs = []
118 | self.decoder_inputs = []
119 | self.target_weights = []
120 | for i in xrange(buckets[-1][0]): # Last bucket is the biggest one.
121 | self.encoder_inputs.append(tf.placeholder(tf.int32, shape=[None],
122 | name="encoder{0}".format(i)))
123 | for i in xrange(buckets[-1][1] + 1):
124 | self.decoder_inputs.append(tf.placeholder(tf.int32, shape=[None],
125 | name="decoder{0}".format(i)))
126 | self.target_weights.append(tf.placeholder(tf.float32, shape=[None],
127 | name="weight{0}".format(i)))
128 |
129 | # Our targets are decoder inputs shifted by one.
130 | targets = [self.decoder_inputs[i + 1]
131 | for i in xrange(len(self.decoder_inputs) - 1)]
132 |
133 | # Training outputs and losses.
134 | if forward_only:
135 | self.outputs, self.losses = tf.nn.seq2seq.model_with_buckets(
136 | self.encoder_inputs, self.decoder_inputs, targets,
137 | self.target_weights, buckets, lambda x, y: seq2seq_f(x, y, True),
138 | softmax_loss_function=softmax_loss_function)
139 | # If we use output projection, we need to project outputs for decoding.
140 | if output_projection is not None:
141 | for b in xrange(len(buckets)):
142 | self.outputs[b] = [
143 | tf.matmul(output, output_projection[0]) + output_projection[1]
144 | for output in self.outputs[b]
145 | ]
146 | else:
147 | self.outputs, self.losses = tf.nn.seq2seq.model_with_buckets(
148 | self.encoder_inputs, self.decoder_inputs, targets,
149 | self.target_weights, buckets,
150 | lambda x, y: seq2seq_f(x, y, False),
151 | softmax_loss_function=softmax_loss_function)
152 |
153 | # Gradients and SGD update operation for training the model.
154 | params = tf.trainable_variables()
155 | if not forward_only:
156 | self.gradient_norms = []
157 | self.updates = []
158 | opt = tf.train.AdamOptimizer()
159 | for b in xrange(len(buckets)):
160 | gradients = tf.gradients(self.losses[b], params)
161 | clipped_gradients, norm = tf.clip_by_global_norm(gradients,
162 | max_gradient_norm)
163 | self.gradient_norms.append(norm)
164 | self.updates.append(opt.apply_gradients(
165 | zip(clipped_gradients, params), global_step=self.global_step))
166 |
167 | self.saver = tf.train.Saver(tf.global_variables())
168 |
169 | def step(self, session, encoder_inputs, decoder_inputs, target_weights,
170 | bucket_id, forward_only):
171 | """Run a step of the model feeding the given inputs.
172 |
173 | Args:
174 | session: tensorflow session to use.
175 | encoder_inputs: list of numpy int vectors to feed as encoder inputs.
176 | decoder_inputs: list of numpy int vectors to feed as decoder inputs.
177 | target_weights: list of numpy float vectors to feed as target weights.
178 | bucket_id: which bucket of the model to use.
179 | forward_only: whether to do the backward step or only forward.
180 |
181 | Returns:
182 | A triple consisting of gradient norm (or None if we did not do backward),
183 | average perplexity, and the outputs.
184 |
185 | Raises:
186 | ValueError: if length of encoder_inputs, decoder_inputs, or
187 | target_weights disagrees with bucket size for the specified bucket_id.
188 | """
189 | # Check if the sizes match.
190 | encoder_size, decoder_size = self.buckets[bucket_id]
191 | if len(encoder_inputs) != encoder_size:
192 | raise ValueError("Encoder length must be equal to the one in bucket,"
193 | " %d != %d." % (len(encoder_inputs), encoder_size))
194 | if len(decoder_inputs) != decoder_size:
195 | raise ValueError("Decoder length must be equal to the one in bucket,"
196 | " %d != %d." % (len(decoder_inputs), decoder_size))
197 | if len(target_weights) != decoder_size:
198 | raise ValueError("Weights length must be equal to the one in bucket,"
199 | " %d != %d." % (len(target_weights), decoder_size))
200 |
201 | # Input feed: encoder inputs, decoder inputs, target_weights, as provided.
202 | input_feed = {}
203 | for l in xrange(encoder_size):
204 | input_feed[self.encoder_inputs[l].name] = encoder_inputs[l]
205 | for l in xrange(decoder_size):
206 | input_feed[self.decoder_inputs[l].name] = decoder_inputs[l]
207 | input_feed[self.target_weights[l].name] = target_weights[l]
208 |
209 | # Since our targets are decoder inputs shifted by one, we need one more.
210 | last_target = self.decoder_inputs[decoder_size].name
211 | input_feed[last_target] = np.zeros([self.batch_size], dtype=np.int32)
212 |
213 | # Output feed: depends on whether we do a backward step or not.
214 | if not forward_only:
215 | output_feed = [self.updates[bucket_id], # Update Op that does SGD.
216 | self.gradient_norms[bucket_id], # Gradient norm.
217 | self.losses[bucket_id]] # Loss for this batch.
218 | else:
219 | output_feed = [self.losses[bucket_id]] # Loss for this batch.
220 | for l in xrange(decoder_size): # Output logits.
221 | output_feed.append(self.outputs[bucket_id][l])
222 |
223 | outputs = session.run(output_feed, input_feed)
224 | if not forward_only:
225 | return outputs[1], outputs[2], None # Gradient norm, loss, no outputs.
226 | else:
227 | return None, outputs[0], outputs[1:] # No gradient norm, loss, outputs.
228 |
229 | def get_batch(self, data, bucket_id):
230 | """Get a random batch of data from the specified bucket, prepare for step.
231 |
232 | To feed data in step(..) it must be a list of batch-major vectors, while
233 | data here contains single length-major cases. So the main logic of this
234 | function is to re-index data cases to be in the proper format for feeding.
235 |
236 | Args:
237 | data: a tuple of size len(self.buckets) in which each element contains
238 | lists of pairs of input and output data that we use to create a batch.
239 | bucket_id: integer, which bucket to get the batch for.
240 |
241 | Returns:
242 | The triple (encoder_inputs, decoder_inputs, target_weights) for
243 | the constructed batch that has the proper format to call step(...) later.
244 | """
245 | encoder_size, decoder_size = self.buckets[bucket_id]
246 | encoder_inputs, decoder_inputs = [], []
247 |
248 | # Get a random batch of encoder and decoder inputs from data,
249 | # pad them if needed, reverse encoder inputs and add GO to decoder.
250 | for _ in xrange(self.batch_size):
251 | encoder_input, decoder_input = random.choice(data[bucket_id])
252 |
253 | # Encoder inputs are padded and then reversed.
254 | encoder_pad = [data_utils.PAD_ID] * (encoder_size - len(encoder_input))
255 | encoder_inputs.append(list(reversed(encoder_input + encoder_pad)))
256 |
257 | # Decoder inputs get an extra "GO" symbol, and are padded then.
258 | decoder_pad_size = decoder_size - len(decoder_input) - 1
259 | decoder_inputs.append([data_utils.GO_ID] + decoder_input +
260 | [data_utils.PAD_ID] * decoder_pad_size)
261 |
262 | # Now we create batch-major vectors from the data selected above.
263 | batch_encoder_inputs, batch_decoder_inputs, batch_weights = [], [], []
264 |
265 | # Batch encoder inputs are just re-indexed encoder_inputs.
266 | for length_idx in xrange(encoder_size):
267 | batch_encoder_inputs.append(
268 | np.array([encoder_inputs[batch_idx][length_idx]
269 | for batch_idx in xrange(self.batch_size)], dtype=np.int32))
270 |
271 | # Batch decoder inputs are re-indexed decoder_inputs, we create weights.
272 | for length_idx in xrange(decoder_size):
273 | batch_decoder_inputs.append(
274 | np.array([decoder_inputs[batch_idx][length_idx]
275 | for batch_idx in xrange(self.batch_size)], dtype=np.int32))
276 |
277 | # Create target_weights to be 0 for targets that are padding.
278 | batch_weight = np.ones(self.batch_size, dtype=np.float32)
279 | for batch_idx in xrange(self.batch_size):
280 | # We set weight to 0 if the corresponding target is a PAD symbol.
281 | # The corresponding target is decoder_input shifted by 1 forward.
282 | if length_idx < decoder_size - 1:
283 | target = decoder_inputs[batch_idx][length_idx + 1]
284 | if length_idx == decoder_size - 1 or target == data_utils.PAD_ID:
285 | batch_weight[batch_idx] = 0.0
286 | batch_weights.append(batch_weight)
287 | return batch_encoder_inputs, batch_decoder_inputs, batch_weights
288 |
--------------------------------------------------------------------------------
/seq2seq_serve.ini:
--------------------------------------------------------------------------------
1 | [strings]
2 | # Mode : train, test, serve
3 | mode = serve
4 | train_enc = data/train.enc
5 | train_dec = data/train.dec
6 | test_enc = data/test.enc
7 | test_dec = data/test.enc
8 | # folder where checkpoints, vocabulary, temporary data will be stored
9 | working_directory = working_dir/
10 | [ints]
11 | # vocabulary size
12 | # 20,000 is a reasonable size
13 | enc_vocab_size = 20000
14 | dec_vocab_size = 20000
15 | # number of LSTM layers : 1/2/3
16 | num_layers = 3
17 | # typical options : 128, 256, 512, 1024
18 | layer_size = 256
19 | # dataset size limit; typically none : no limit
20 | max_train_data_size = 0
21 | batch_size = 64
22 | # steps per checkpoint
23 | # Note : At a checkpoint, models parameters are saved, model is evaluated
24 | # and results are printed
25 | steps_per_checkpoint = 300
26 | [floats]
27 | learning_rate = 0.5
28 | learning_rate_decay_factor = 0.99
29 | max_gradient_norm = 5.0
30 | ##############################################################################
31 | # Note : Edit the bucket sizes at line47 of execute.py (_buckets)
32 | #
33 | # Learn more about the configurations from this link
34 | # https://www.tensorflow.org/versions/r0.9/tutorials/seq2seq/index.html
35 | ##############################################################################
36 |
--------------------------------------------------------------------------------
/ui/.gitignore:
--------------------------------------------------------------------------------
1 | .DS_Store
2 | *.pyc
3 | *.pyo
4 | env
5 | env*
6 | dist
7 | build
8 | *.egg
9 | *.egg-info
10 | _mailinglist
11 | .tox
12 | .cache/
13 | .idea/
14 |
--------------------------------------------------------------------------------
/ui/README.md:
--------------------------------------------------------------------------------
1 | # Flask Chatbox
2 |
3 | A simple flask application that provides an UI for chat and handles the model internally.
4 |
--------------------------------------------------------------------------------
/ui/__init.py__:
--------------------------------------------------------------------------------
https://raw.githubusercontent.com/suriyadeepan/easy_seq2seq/c05b0d02f15e9bd1947ff79ddcd0b761a214e2a9/ui/__init.py__
--------------------------------------------------------------------------------
/ui/app.py:
--------------------------------------------------------------------------------
1 | from flask import Flask, render_template, request
2 | from flask import jsonify
3 |
4 | app = Flask(__name__,static_url_path="/static")
5 |
6 | #############
7 | # Routing
8 | #
9 | @app.route('/message', methods=['POST'])
10 | def reply():
11 | return jsonify( { 'text': execute.decode_line(sess, model, enc_vocab, rev_dec_vocab, request.form['msg'] ) } )
12 |
13 | @app.route("/")
14 | def index():
15 | return render_template("index.html")
16 | #############
17 |
18 | '''
19 | Init seq2seq model
20 |
21 | 1. Call main from execute.py
22 | 2. Create decode_line function that takes message as input
23 | '''
24 | #_________________________________________________________________
25 | import sys
26 | import os.path
27 |
28 | sys.path.append(os.path.abspath(os.path.join(os.path.dirname(__file__), os.path.pardir)))
29 | import tensorflow as tf
30 | import execute
31 |
32 | sess = tf.Session()
33 | sess, model, enc_vocab, rev_dec_vocab = execute.init_session(sess, conf='seq2seq_serve.ini')
34 | #_________________________________________________________________
35 |
36 | # start app
37 | if (__name__ == "__main__"):
38 | app.run(port = 5000)
39 |
--------------------------------------------------------------------------------
/ui/requirements.txt:
--------------------------------------------------------------------------------
1 | click==6.6
2 | Flask==0.11.1
3 | itsdangerous==0.24
4 | Jinja2==2.8
5 | MarkupSafe==0.23
6 | Werkzeug==0.11.10
7 |
--------------------------------------------------------------------------------
/ui/setup.sh:
--------------------------------------------------------------------------------
1 | sudo -H pip install --upgrade pip
2 | sudo -H pip install -r requirements.txt
3 |
--------------------------------------------------------------------------------
/ui/static/css/normalize.css:
--------------------------------------------------------------------------------
1 | /*! normalize.css v4.0.0 | MIT License | github.com/necolas/normalize.css */html{font-family:sans-serif;-ms-text-size-adjust:100%;-webkit-text-size-adjust:100%}body{margin:0}article,aside,details,figcaption,figure,footer,header,main,menu,nav,section,summary{display:block}audio,canvas,progress,video{display:inline-block}audio:not([controls]){display:none;height:0}progress{vertical-align:baseline}template,[hidden]{display:none}a{background-color:transparent}a:active,a:hover{outline-width:0}abbr[title]{border-bottom:none;text-decoration:underline;text-decoration:underline dotted}b,strong{font-weight:inherit}b,strong{font-weight:bolder}dfn{font-style:italic}h1{font-size:2em;margin:0.67em 0}mark{background-color:#ff0;color:#000}small{font-size:80%}sub,sup{font-size:75%;line-height:0;position:relative;vertical-align:baseline}sub{bottom:-0.25em}sup{top:-0.5em}img{border-style:none}svg:not(:root){overflow:hidden}code,kbd,pre,samp{font-family:monospace, monospace;font-size:1em}figure{margin:1em 40px}hr{-webkit-box-sizing:content-box;-moz-box-sizing:content-box;box-sizing:content-box;height:0;overflow:visible}button,input,select,textarea{font:inherit;margin:0}optgroup{font-weight:bold}button,input,select{overflow:visible}button,select{text-transform:none}button,[type="button"],[type="reset"],[type="submit"]{cursor:pointer}[disabled]{cursor:default}button,html [type="button"],[type="reset"],[type="submit"]{-webkit-appearance:button}button::-moz-focus-inner,input::-moz-focus-inner{border:0;padding:0}button:-moz-focusring,input:-moz-focusring{outline:1px dotted ButtonText}fieldset{border:1px solid #c0c0c0;margin:0 2px;padding:0.35em 0.625em 0.75em}legend{-webkit-box-sizing:border-box;-moz-box-sizing:border-box;box-sizing:border-box;color:inherit;display:table;max-width:100%;padding:0;white-space:normal}textarea{overflow:auto}[type="checkbox"],[type="radio"]{-webkit-box-sizing:border-box;-moz-box-sizing:border-box;box-sizing:border-box;padding:0}[type="number"]::-webkit-inner-spin-button,[type="number"]::-webkit-outer-spin-button{height:auto}[type="search"]{-webkit-appearance:textfield}[type="search"]::-webkit-search-cancel-button,[type="search"]::-webkit-search-decoration{-webkit-appearance:none}
2 |
--------------------------------------------------------------------------------
/ui/static/css/style.css:
--------------------------------------------------------------------------------
1 | /*--------------------
2 | Mixins
3 | --------------------*/
4 | /*--------------------
5 | Body
6 | --------------------*/
7 | *,
8 | *::before,
9 | *::after {
10 | box-sizing: border-box;
11 | }
12 |
13 | html,
14 | body {
15 | height: 100%;
16 | }
17 |
18 | body {
19 | background: -webkit-linear-gradient(315deg, #044f48, #2a7561);
20 | background: linear-gradient(135deg, #044f48, #2a7561);
21 | background-size: cover;
22 | font-family: 'Open Sans', sans-serif;
23 | font-size: 12px;
24 | line-height: 1.3;
25 | overflow: hidden;
26 | }
27 |
28 | .bg {
29 | width: 100%;
30 | height: 100%;
31 | top: 0;
32 | left: 0;
33 | z-index: 1;
34 | background: url("/static/res/hd2.jpg") no-repeat 0 0;
35 | -webkit-filter: blur(80px);
36 | filter: blur(80px);
37 | -webkit-transform: scale(1.2);
38 | transform: scale(1.2);
39 | }
40 |
41 | /*--------------------
42 | Chat
43 | --------------------*/
44 | .chat {
45 | position: absolute;
46 | top: 50%;
47 | left: 50%;
48 | -webkit-transform: translate(-50%, -50%);
49 | transform: translate(-50%, -50%);
50 | width: 300px;
51 | height: 80vh;
52 | max-height: 500px;
53 | z-index: 2;
54 | overflow: hidden;
55 | box-shadow: 0 5px 30px rgba(0, 0, 0, 0.2);
56 | background: rgba(0, 0, 0, 0.5);
57 | border-radius: 20px;
58 | display: -webkit-box;
59 | display: -webkit-flex;
60 | display: -ms-flexbox;
61 | display: flex;
62 | -webkit-box-pack: justify;
63 | -webkit-justify-content: space-between;
64 | -ms-flex-pack: justify;
65 | justify-content: space-between;
66 | -webkit-box-orient: vertical;
67 | -webkit-box-direction: normal;
68 | -webkit-flex-direction: column;
69 | -ms-flex-direction: column;
70 | flex-direction: column;
71 | }
72 |
73 | /*--------------------
74 | Chat Title
75 | --------------------*/
76 | .chat-title {
77 | -webkit-box-flex: 0;
78 | -webkit-flex: 0 1 45px;
79 | -ms-flex: 0 1 45px;
80 | flex: 0 1 45px;
81 | position: relative;
82 | z-index: 2;
83 | background: rgba(0, 0, 0, 0.2);
84 | color: #fff;
85 | text-transform: uppercase;
86 | text-align: left;
87 | padding: 10px 10px 10px 50px;
88 | }
89 | .chat-title h1, .chat-title h2 {
90 | font-weight: normal;
91 | font-size: 10px;
92 | margin: 0;
93 | padding: 0;
94 | }
95 | .chat-title h2 {
96 | color: rgba(255, 255, 255, 0.5);
97 | font-size: 8px;
98 | letter-spacing: 1px;
99 | }
100 | .chat-title .avatar {
101 | position: absolute;
102 | z-index: 1;
103 | top: 8px;
104 | left: 9px;
105 | border-radius: 30px;
106 | width: 30px;
107 | height: 30px;
108 | overflow: hidden;
109 | margin: 0;
110 | padding: 0;
111 | border: 2px solid rgba(255, 255, 255, 0.24);
112 | }
113 | .chat-title .avatar img {
114 | width: 100%;
115 | height: auto;
116 | }
117 |
118 | /*--------------------
119 | Messages
120 | --------------------*/
121 | .messages {
122 | -webkit-box-flex: 1;
123 | -webkit-flex: 1 1 auto;
124 | -ms-flex: 1 1 auto;
125 | flex: 1 1 auto;
126 | color: rgba(255, 255, 255, 0.5);
127 | overflow: hidden;
128 | position: relative;
129 | width: 100%;
130 | }
131 | .messages .messages-content {
132 | position: absolute;
133 | top: 0;
134 | left: 0;
135 | height: 101%;
136 | width: 100%;
137 | }
138 | .messages .message {
139 | clear: both;
140 | float: left;
141 | padding: 6px 10px 7px;
142 | border-radius: 10px 10px 10px 0;
143 | background: rgba(0, 0, 0, 0.3);
144 | margin: 8px 0;
145 | font-size: 11px;
146 | line-height: 1.4;
147 | margin-left: 35px;
148 | position: relative;
149 | text-shadow: 0 1px 1px rgba(0, 0, 0, 0.2);
150 | }
151 | .messages .message .timestamp {
152 | position: absolute;
153 | bottom: -15px;
154 | font-size: 9px;
155 | color: rgba(255, 255, 255, 0.3);
156 | }
157 | .messages .message::before {
158 | content: '';
159 | position: absolute;
160 | bottom: -6px;
161 | border-top: 6px solid rgba(0, 0, 0, 0.3);
162 | left: 0;
163 | border-right: 7px solid transparent;
164 | }
165 | .messages .message .avatar {
166 | position: absolute;
167 | z-index: 1;
168 | bottom: -15px;
169 | left: -35px;
170 | border-radius: 30px;
171 | width: 30px;
172 | height: 30px;
173 | overflow: hidden;
174 | margin: 0;
175 | padding: 0;
176 | border: 2px solid rgba(255, 255, 255, 0.24);
177 | }
178 | .messages .message .avatar img {
179 | width: 100%;
180 | height: auto;
181 | }
182 | .messages .message.message-personal {
183 | float: right;
184 | color: #fff;
185 | text-align: right;
186 | background: -webkit-linear-gradient(330deg, #248A52, #257287);
187 | background: linear-gradient(120deg, #248A52, #257287);
188 | border-radius: 10px 10px 0 10px;
189 | }
190 | .messages .message.message-personal::before {
191 | left: auto;
192 | right: 0;
193 | border-right: none;
194 | border-left: 5px solid transparent;
195 | border-top: 4px solid #257287;
196 | bottom: -4px;
197 | }
198 | .messages .message:last-child {
199 | margin-bottom: 30px;
200 | }
201 | .messages .message.new {
202 | -webkit-transform: scale(0);
203 | transform: scale(0);
204 | -webkit-transform-origin: 0 0;
205 | transform-origin: 0 0;
206 | -webkit-animation: bounce 500ms linear both;
207 | animation: bounce 500ms linear both;
208 | }
209 | .messages .message.loading::before {
210 | position: absolute;
211 | top: 50%;
212 | left: 50%;
213 | -webkit-transform: translate(-50%, -50%);
214 | transform: translate(-50%, -50%);
215 | content: '';
216 | display: block;
217 | width: 3px;
218 | height: 3px;
219 | border-radius: 50%;
220 | background: rgba(255, 255, 255, 0.5);
221 | z-index: 2;
222 | margin-top: 4px;
223 | -webkit-animation: ball 0.45s cubic-bezier(0, 0, 0.15, 1) alternate infinite;
224 | animation: ball 0.45s cubic-bezier(0, 0, 0.15, 1) alternate infinite;
225 | border: none;
226 | -webkit-animation-delay: .15s;
227 | animation-delay: .15s;
228 | }
229 | .messages .message.loading span {
230 | display: block;
231 | font-size: 0;
232 | width: 20px;
233 | height: 10px;
234 | position: relative;
235 | }
236 | .messages .message.loading span::before {
237 | position: absolute;
238 | top: 50%;
239 | left: 50%;
240 | -webkit-transform: translate(-50%, -50%);
241 | transform: translate(-50%, -50%);
242 | content: '';
243 | display: block;
244 | width: 3px;
245 | height: 3px;
246 | border-radius: 50%;
247 | background: rgba(255, 255, 255, 0.5);
248 | z-index: 2;
249 | margin-top: 4px;
250 | -webkit-animation: ball 0.45s cubic-bezier(0, 0, 0.15, 1) alternate infinite;
251 | animation: ball 0.45s cubic-bezier(0, 0, 0.15, 1) alternate infinite;
252 | margin-left: -7px;
253 | }
254 | .messages .message.loading span::after {
255 | position: absolute;
256 | top: 50%;
257 | left: 50%;
258 | -webkit-transform: translate(-50%, -50%);
259 | transform: translate(-50%, -50%);
260 | content: '';
261 | display: block;
262 | width: 3px;
263 | height: 3px;
264 | border-radius: 50%;
265 | background: rgba(255, 255, 255, 0.5);
266 | z-index: 2;
267 | margin-top: 4px;
268 | -webkit-animation: ball 0.45s cubic-bezier(0, 0, 0.15, 1) alternate infinite;
269 | animation: ball 0.45s cubic-bezier(0, 0, 0.15, 1) alternate infinite;
270 | margin-left: 7px;
271 | -webkit-animation-delay: .3s;
272 | animation-delay: .3s;
273 | }
274 |
275 | /*--------------------
276 | Message Box
277 | --------------------*/
278 | .message-box {
279 | -webkit-box-flex: 0;
280 | -webkit-flex: 0 1 40px;
281 | -ms-flex: 0 1 40px;
282 | flex: 0 1 40px;
283 | width: 100%;
284 | background: rgba(0, 0, 0, 0.3);
285 | padding: 10px;
286 | position: relative;
287 | }
288 | .message-box .message-input {
289 | background: none;
290 | border: none;
291 | outline: none !important;
292 | resize: none;
293 | color: rgba(255, 255, 255, 0.7);
294 | font-size: 11px;
295 | height: 17px;
296 | margin: 0;
297 | padding-right: 20px;
298 | width: 265px;
299 | }
300 | .message-box textarea:focus:-webkit-placeholder {
301 | color: transparent;
302 | }
303 | .message-box .message-submit {
304 | position: absolute;
305 | z-index: 1;
306 | top: 9px;
307 | right: 10px;
308 | color: #fff;
309 | border: none;
310 | background: #248A52;
311 | font-size: 10px;
312 | text-transform: uppercase;
313 | line-height: 1;
314 | padding: 6px 10px;
315 | border-radius: 10px;
316 | outline: none !important;
317 | -webkit-transition: background .2s ease;
318 | transition: background .2s ease;
319 | }
320 | .message-box .message-submit:hover {
321 | background: #1D7745;
322 | }
323 |
324 | /*--------------------
325 | Custom Srollbar
326 | --------------------*/
327 | .mCSB_scrollTools {
328 | margin: 1px -3px 1px 0;
329 | opacity: 0;
330 | }
331 |
332 | .mCSB_inside > .mCSB_container {
333 | margin-right: 0px;
334 | padding: 0 10px;
335 | }
336 |
337 | .mCSB_scrollTools .mCSB_dragger .mCSB_dragger_bar {
338 | background-color: rgba(0, 0, 0, 0.5) !important;
339 | }
340 |
341 | /*--------------------
342 | Bounce
343 | --------------------*/
344 | @-webkit-keyframes bounce {
345 | 0% {
346 | -webkit-transform: matrix3d(0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 1, 0, 0, 0, 0, 1);
347 | transform: matrix3d(0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 1, 0, 0, 0, 0, 1);
348 | }
349 | 4.7% {
350 | -webkit-transform: matrix3d(0.45, 0, 0, 0, 0, 0.45, 0, 0, 0, 0, 1, 0, 0, 0, 0, 1);
351 | transform: matrix3d(0.45, 0, 0, 0, 0, 0.45, 0, 0, 0, 0, 1, 0, 0, 0, 0, 1);
352 | }
353 | 9.41% {
354 | -webkit-transform: matrix3d(0.883, 0, 0, 0, 0, 0.883, 0, 0, 0, 0, 1, 0, 0, 0, 0, 1);
355 | transform: matrix3d(0.883, 0, 0, 0, 0, 0.883, 0, 0, 0, 0, 1, 0, 0, 0, 0, 1);
356 | }
357 | 14.11% {
358 | -webkit-transform: matrix3d(1.141, 0, 0, 0, 0, 1.141, 0, 0, 0, 0, 1, 0, 0, 0, 0, 1);
359 | transform: matrix3d(1.141, 0, 0, 0, 0, 1.141, 0, 0, 0, 0, 1, 0, 0, 0, 0, 1);
360 | }
361 | 18.72% {
362 | -webkit-transform: matrix3d(1.212, 0, 0, 0, 0, 1.212, 0, 0, 0, 0, 1, 0, 0, 0, 0, 1);
363 | transform: matrix3d(1.212, 0, 0, 0, 0, 1.212, 0, 0, 0, 0, 1, 0, 0, 0, 0, 1);
364 | }
365 | 24.32% {
366 | -webkit-transform: matrix3d(1.151, 0, 0, 0, 0, 1.151, 0, 0, 0, 0, 1, 0, 0, 0, 0, 1);
367 | transform: matrix3d(1.151, 0, 0, 0, 0, 1.151, 0, 0, 0, 0, 1, 0, 0, 0, 0, 1);
368 | }
369 | 29.93% {
370 | -webkit-transform: matrix3d(1.048, 0, 0, 0, 0, 1.048, 0, 0, 0, 0, 1, 0, 0, 0, 0, 1);
371 | transform: matrix3d(1.048, 0, 0, 0, 0, 1.048, 0, 0, 0, 0, 1, 0, 0, 0, 0, 1);
372 | }
373 | 35.54% {
374 | -webkit-transform: matrix3d(0.979, 0, 0, 0, 0, 0.979, 0, 0, 0, 0, 1, 0, 0, 0, 0, 1);
375 | transform: matrix3d(0.979, 0, 0, 0, 0, 0.979, 0, 0, 0, 0, 1, 0, 0, 0, 0, 1);
376 | }
377 | 41.04% {
378 | -webkit-transform: matrix3d(0.961, 0, 0, 0, 0, 0.961, 0, 0, 0, 0, 1, 0, 0, 0, 0, 1);
379 | transform: matrix3d(0.961, 0, 0, 0, 0, 0.961, 0, 0, 0, 0, 1, 0, 0, 0, 0, 1);
380 | }
381 | 52.15% {
382 | -webkit-transform: matrix3d(0.991, 0, 0, 0, 0, 0.991, 0, 0, 0, 0, 1, 0, 0, 0, 0, 1);
383 | transform: matrix3d(0.991, 0, 0, 0, 0, 0.991, 0, 0, 0, 0, 1, 0, 0, 0, 0, 1);
384 | }
385 | 63.26% {
386 | -webkit-transform: matrix3d(1.007, 0, 0, 0, 0, 1.007, 0, 0, 0, 0, 1, 0, 0, 0, 0, 1);
387 | transform: matrix3d(1.007, 0, 0, 0, 0, 1.007, 0, 0, 0, 0, 1, 0, 0, 0, 0, 1);
388 | }
389 | 85.49% {
390 | -webkit-transform: matrix3d(0.999, 0, 0, 0, 0, 0.999, 0, 0, 0, 0, 1, 0, 0, 0, 0, 1);
391 | transform: matrix3d(0.999, 0, 0, 0, 0, 0.999, 0, 0, 0, 0, 1, 0, 0, 0, 0, 1);
392 | }
393 | 100% {
394 | -webkit-transform: matrix3d(1, 0, 0, 0, 0, 1, 0, 0, 0, 0, 1, 0, 0, 0, 0, 1);
395 | transform: matrix3d(1, 0, 0, 0, 0, 1, 0, 0, 0, 0, 1, 0, 0, 0, 0, 1);
396 | }
397 | }
398 | @keyframes bounce {
399 | 0% {
400 | -webkit-transform: matrix3d(0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 1, 0, 0, 0, 0, 1);
401 | transform: matrix3d(0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 1, 0, 0, 0, 0, 1);
402 | }
403 | 4.7% {
404 | -webkit-transform: matrix3d(0.45, 0, 0, 0, 0, 0.45, 0, 0, 0, 0, 1, 0, 0, 0, 0, 1);
405 | transform: matrix3d(0.45, 0, 0, 0, 0, 0.45, 0, 0, 0, 0, 1, 0, 0, 0, 0, 1);
406 | }
407 | 9.41% {
408 | -webkit-transform: matrix3d(0.883, 0, 0, 0, 0, 0.883, 0, 0, 0, 0, 1, 0, 0, 0, 0, 1);
409 | transform: matrix3d(0.883, 0, 0, 0, 0, 0.883, 0, 0, 0, 0, 1, 0, 0, 0, 0, 1);
410 | }
411 | 14.11% {
412 | -webkit-transform: matrix3d(1.141, 0, 0, 0, 0, 1.141, 0, 0, 0, 0, 1, 0, 0, 0, 0, 1);
413 | transform: matrix3d(1.141, 0, 0, 0, 0, 1.141, 0, 0, 0, 0, 1, 0, 0, 0, 0, 1);
414 | }
415 | 18.72% {
416 | -webkit-transform: matrix3d(1.212, 0, 0, 0, 0, 1.212, 0, 0, 0, 0, 1, 0, 0, 0, 0, 1);
417 | transform: matrix3d(1.212, 0, 0, 0, 0, 1.212, 0, 0, 0, 0, 1, 0, 0, 0, 0, 1);
418 | }
419 | 24.32% {
420 | -webkit-transform: matrix3d(1.151, 0, 0, 0, 0, 1.151, 0, 0, 0, 0, 1, 0, 0, 0, 0, 1);
421 | transform: matrix3d(1.151, 0, 0, 0, 0, 1.151, 0, 0, 0, 0, 1, 0, 0, 0, 0, 1);
422 | }
423 | 29.93% {
424 | -webkit-transform: matrix3d(1.048, 0, 0, 0, 0, 1.048, 0, 0, 0, 0, 1, 0, 0, 0, 0, 1);
425 | transform: matrix3d(1.048, 0, 0, 0, 0, 1.048, 0, 0, 0, 0, 1, 0, 0, 0, 0, 1);
426 | }
427 | 35.54% {
428 | -webkit-transform: matrix3d(0.979, 0, 0, 0, 0, 0.979, 0, 0, 0, 0, 1, 0, 0, 0, 0, 1);
429 | transform: matrix3d(0.979, 0, 0, 0, 0, 0.979, 0, 0, 0, 0, 1, 0, 0, 0, 0, 1);
430 | }
431 | 41.04% {
432 | -webkit-transform: matrix3d(0.961, 0, 0, 0, 0, 0.961, 0, 0, 0, 0, 1, 0, 0, 0, 0, 1);
433 | transform: matrix3d(0.961, 0, 0, 0, 0, 0.961, 0, 0, 0, 0, 1, 0, 0, 0, 0, 1);
434 | }
435 | 52.15% {
436 | -webkit-transform: matrix3d(0.991, 0, 0, 0, 0, 0.991, 0, 0, 0, 0, 1, 0, 0, 0, 0, 1);
437 | transform: matrix3d(0.991, 0, 0, 0, 0, 0.991, 0, 0, 0, 0, 1, 0, 0, 0, 0, 1);
438 | }
439 | 63.26% {
440 | -webkit-transform: matrix3d(1.007, 0, 0, 0, 0, 1.007, 0, 0, 0, 0, 1, 0, 0, 0, 0, 1);
441 | transform: matrix3d(1.007, 0, 0, 0, 0, 1.007, 0, 0, 0, 0, 1, 0, 0, 0, 0, 1);
442 | }
443 | 85.49% {
444 | -webkit-transform: matrix3d(0.999, 0, 0, 0, 0, 0.999, 0, 0, 0, 0, 1, 0, 0, 0, 0, 1);
445 | transform: matrix3d(0.999, 0, 0, 0, 0, 0.999, 0, 0, 0, 0, 1, 0, 0, 0, 0, 1);
446 | }
447 | 100% {
448 | -webkit-transform: matrix3d(1, 0, 0, 0, 0, 1, 0, 0, 0, 0, 1, 0, 0, 0, 0, 1);
449 | transform: matrix3d(1, 0, 0, 0, 0, 1, 0, 0, 0, 0, 1, 0, 0, 0, 0, 1);
450 | }
451 | }
452 | @-webkit-keyframes ball {
453 | from {
454 | -webkit-transform: translateY(0) scaleY(0.8);
455 | transform: translateY(0) scaleY(0.8);
456 | }
457 | to {
458 | -webkit-transform: translateY(-10px);
459 | transform: translateY(-10px);
460 | }
461 | }
462 | @keyframes ball {
463 | from {
464 | -webkit-transform: translateY(0) scaleY(0.8);
465 | transform: translateY(0) scaleY(0.8);
466 | }
467 | to {
468 | -webkit-transform: translateY(-10px);
469 | transform: translateY(-10px);
470 | }
471 | }
472 |
--------------------------------------------------------------------------------
/ui/static/js/index.js:
--------------------------------------------------------------------------------
1 | var $messages = $('.messages-content'),
2 | d, h, m,
3 | i = 0;
4 |
5 | $(window).load(function() {
6 | $messages.mCustomScrollbar();
7 | setTimeout(function() {
8 | //fakeMessage();
9 | }, 100);
10 | });
11 |
12 | function updateScrollbar() {
13 | $messages.mCustomScrollbar("update").mCustomScrollbar('scrollTo', 'bottom', {
14 | scrollInertia: 10,
15 | timeout: 0
16 | });
17 | }
18 |
19 | function setDate(){
20 | d = new Date()
21 | if (m != d.getMinutes()) {
22 | m = d.getMinutes();
23 | $('' + d.getHours() + ':' + m + '
').appendTo($('.message:last'));
24 | }
25 | }
26 |
27 | function insertMessage() {
28 | msg = $('.message-input').val();
29 | if ($.trim(msg) == '') {
30 | return false;
31 | }
32 | $('' + msg + '
').appendTo($('.mCSB_container')).addClass('new');
33 | setDate();
34 | $('.message-input').val(null);
35 | updateScrollbar();
36 | interact(msg);
37 | setTimeout(function() {
38 | //fakeMessage();
39 | }, 1000 + (Math.random() * 20) * 100);
40 | }
41 |
42 | $('.message-submit').click(function() {
43 | insertMessage();
44 | });
45 |
46 | $(window).on('keydown', function(e) {
47 | if (e.which == 13) {
48 | insertMessage();
49 | return false;
50 | }
51 | })
52 |
53 |
54 | function interact(message){
55 | // loading message
56 | $('').appendTo($('.mCSB_container'));
57 | // make a POST request [ajax call]
58 | $.post('/message', {
59 | msg: message,
60 | }).done(function(reply) {
61 | // Message Received
62 | // remove loading meassage
63 | $('.message.loading').remove();
64 | // Add message to chatbox
65 | $('
' + reply['text'] + '
18 |
19 |
JANUS
20 |
Ask Me Anything
21 |
22 | 
23 |
24 |
27 |
28 |
29 |
30 |
31 |
32 |