├── src ├── README.md ├── batched_inv_joblib.py ├── content_wmf.py ├── WMF_ML20M.ipynb ├── rec_eval.py ├── cofacto.py ├── preprocess_ML20M.ipynb └── Cofactorization_ML20M.ipynb ├── README.md ├── .gitignore └── LICENSE /src/README.md: -------------------------------------------------------------------------------- 1 | # Experimental Results 2 | By getting the data and running the following notebooks, you should be able to reproduce the experimental results in the paper. 3 | 4 | ## ML20M 5 | - [preprocess_ML20M.ipynb](./preprocess_ML20M.ipynb): pre-process the data and create the train/test/validation splits. 6 | - [Cofactorization_ML20M.ipynb](./Cofactorization_ML20M.ipynb): train the CoFactor model and evaluate on the heldout test set. 7 | - [WMF_ML20M.ipynb](./WMF_ML20M.ipynb): train the baseline WMF model and evaluate on the heldout test set. 8 | 9 | To get the results for TasteProfile, follow [this notebook](https://github.com/dawenl/expo-mf/blob/master/src/processTasteProfile.ipynb) for data pre-processing and replace the data directory `DATA_DIR` in the above notebooks with the location of the processed TasteProfile. 10 | -------------------------------------------------------------------------------- /README.md: -------------------------------------------------------------------------------- 1 | # CoFactor 2 | 3 | This repository contains the source code to reproduce the experimental results as described in the paper ["Factorization Meets the Item Embedding: Regularizing Matrix Factorization with Item Co-occurrence"](http://dawenl.github.io/publications/LiangACB16-cofactor.pdf) (RecSys'16). 4 | 5 | ## Dependencies 6 | The python module dependencies are: 7 | - numpy/scipy 8 | - scikit.learn 9 | - joblib 10 | - bottleneck 11 | - pandas (needed to run the example for data preprocessing) 12 | 13 | **Note**: The code is mostly written for Python 2.7. For Python 3.x, it is still usable with minor modification. If you run into any problem with Python 3.x, feel free to contact me and I will try to get back to you with a helpful solution. 14 | 15 | ## Datasets 16 | - [Taste Profile Subset](http://labrosa.ee.columbia.edu/millionsong/tasteprofile): the pre-processing is done following [this notebook](https://github.com/dawenl/expo-mf/blob/master/src/processTasteProfile.ipynb). 17 | - [MovieLens-20M](http://grouplens.org/datasets/movielens/20m/) 18 | 19 | We adapted the weighted matrix factorization (WMF) implementation from [content_wmf](https://github.com/dawenl/content_wmf) repository. 20 | 21 | ## Examples 22 | See example notebooks in `src/`. 23 | -------------------------------------------------------------------------------- /.gitignore: -------------------------------------------------------------------------------- 1 | # Byte-compiled / optimized / DLL files 2 | __pycache__/ 3 | *.py[cod] 4 | *$py.class 5 | 6 | # C extensions 7 | *.so 8 | 9 | # Distribution / packaging 10 | .Python 11 | env/ 12 | build/ 13 | develop-eggs/ 14 | dist/ 15 | downloads/ 16 | eggs/ 17 | .eggs/ 18 | lib/ 19 | lib64/ 20 | parts/ 21 | sdist/ 22 | var/ 23 | *.egg-info/ 24 | .installed.cfg 25 | *.egg 26 | 27 | # PyInstaller 28 | # Usually these files are written by a python script from a template 29 | # before PyInstaller builds the exe, so as to inject date/other infos into it. 30 | *.manifest 31 | *.spec 32 | 33 | # Installer logs 34 | pip-log.txt 35 | pip-delete-this-directory.txt 36 | 37 | # Unit test / coverage reports 38 | htmlcov/ 39 | .tox/ 40 | .coverage 41 | .coverage.* 42 | .cache 43 | nosetests.xml 44 | coverage.xml 45 | *,cover 46 | .hypothesis/ 47 | 48 | # Translations 49 | *.mo 50 | *.pot 51 | 52 | # Django stuff: 53 | *.log 54 | local_settings.py 55 | 56 | # Flask stuff: 57 | instance/ 58 | .webassets-cache 59 | 60 | # Scrapy stuff: 61 | .scrapy 62 | 63 | # Sphinx documentation 64 | docs/_build/ 65 | 66 | # PyBuilder 67 | target/ 68 | 69 | # IPython Notebook 70 | .ipynb_checkpoints 71 | 72 | # pyenv 73 | .python-version 74 | 75 | # celery beat schedule file 76 | celerybeat-schedule 77 | 78 | # dotenv 79 | .env 80 | 81 | # virtualenv 82 | venv/ 83 | ENV/ 84 | 85 | # Spyder project settings 86 | .spyderproject 87 | 88 | # Rope project settings 89 | .ropeproject 90 | -------------------------------------------------------------------------------- /src/batched_inv_joblib.py: -------------------------------------------------------------------------------- 1 | import os 2 | os.environ['OPENBLAS_NUM_THREADS'] = '1' 3 | import numpy as np 4 | 5 | from joblib import Parallel, delayed 6 | 7 | 8 | def get_row(S, i): 9 | lo, hi = S.indptr[i], S.indptr[i + 1] 10 | return S.data[lo:hi], S.indices[lo:hi] 11 | 12 | 13 | def solve_sequential(As, Bs): 14 | X_stack = np.empty_like(As, dtype=As.dtype) 15 | 16 | for k in xrange(As.shape[0]): 17 | X_stack[k] = np.linalg.solve(Bs[k], As[k]) 18 | 19 | return X_stack 20 | 21 | 22 | def solve_batch(b, S, Y, X_reg, YTYpR, batch_size, m, f, dtype): 23 | lo = b * batch_size 24 | hi = min((b + 1) * batch_size, m) 25 | current_batch_size = hi - lo 26 | 27 | A_stack = np.empty((current_batch_size, f), dtype=dtype) 28 | B_stack = np.empty((current_batch_size, f, f), dtype=dtype) 29 | 30 | for ib, k in enumerate(xrange(lo, hi)): 31 | s_u, i_u = get_row(S, k) 32 | 33 | Y_u = Y[i_u] # exploit sparsity 34 | A = (s_u + 1).dot(Y_u) 35 | 36 | if X_reg is not None: 37 | A += X_reg[k] 38 | 39 | YTSY = np.dot(Y_u.T, (Y_u * s_u[:, None])) 40 | B = YTSY + YTYpR 41 | 42 | A_stack[ib] = A 43 | B_stack[ib] = B 44 | 45 | X_stack = solve_sequential(A_stack, B_stack) 46 | return X_stack 47 | 48 | 49 | def recompute_factors_batched(Y, S, lambda_reg, X=None, 50 | dtype='float32', batch_size=2000, n_jobs=20): 51 | m = S.shape[0] # m = number of users 52 | f = Y.shape[1] # f = number of factors 53 | 54 | YTY = np.dot(Y.T, Y) # precompute this 55 | YTYpR = YTY + lambda_reg * np.eye(f) 56 | if X is not None: 57 | X_reg = lambda_reg * X 58 | else: 59 | X_reg = None 60 | X_new = np.zeros((m, f), dtype=dtype) 61 | 62 | num_batches = int(np.ceil(m / float(batch_size))) 63 | 64 | res = Parallel(n_jobs=n_jobs)(delayed(solve_batch)(b, S, Y, X_reg, YTYpR, 65 | batch_size, m, f, dtype) 66 | for b in xrange(num_batches)) 67 | X_new = np.concatenate(res, axis=0) 68 | 69 | return X_new 70 | -------------------------------------------------------------------------------- /src/content_wmf.py: -------------------------------------------------------------------------------- 1 | import sys 2 | import time 3 | 4 | import numpy as np 5 | from scipy import weave 6 | 7 | import batched_inv_joblib 8 | import rec_eval 9 | 10 | 11 | def linear_surplus_confidence_matrix(B, alpha): 12 | # To construct the surplus confidence matrix, we need to operate only on 13 | # the nonzero elements. 14 | # This is not possible: S = alpha * B 15 | S = B.copy() 16 | S.data = alpha * S.data 17 | return S 18 | 19 | 20 | def log_surplus_confidence_matrix(B, alpha, epsilon): 21 | # To construct the surplus confidence matrix, we need to operate only on 22 | # the nonzero elements. 23 | # This is not possible: S = alpha * np.log(1 + B / epsilon) 24 | S = B.copy() 25 | S.data = alpha * np.log(1 + S.data / epsilon) 26 | return S 27 | 28 | 29 | def factorize(S, num_factors, X=None, vad_data=None, num_iters=10, init_std=0.01, 30 | lambda_U_reg=1e-2, lambda_V_reg=100, lambda_W_reg=1e-2, 31 | dtype='float32', random_state=None, verbose=False, 32 | recompute_factors=batched_inv_joblib.recompute_factors_batched, 33 | fixed_item_embeddings=False, 34 | V=None, 35 | *args, **kwargs): 36 | 37 | num_users, num_items = S.shape 38 | if X is not None: 39 | assert X.shape == (num_items, num_factors) 40 | 41 | if verbose: 42 | print "Precompute S^T (if necessary)" 43 | start_time = time.time() 44 | 45 | ST = S.T.tocsr() 46 | 47 | if verbose: 48 | print " took %.3f seconds" % (time.time() - start_time) 49 | start_time = time.time() 50 | 51 | if type(random_state) is int: 52 | np.random.seed(random_state) 53 | elif random_state is not None: 54 | np.random.setstate(random_state) 55 | 56 | U = None 57 | if not fixed_item_embeddings and not V: 58 | V = np.random.randn(num_items, num_factors).astype(dtype) * init_std 59 | 60 | old_ndcg = -np.inf 61 | for i in xrange(num_iters): 62 | if verbose: 63 | print("Iteration %d:" % i) 64 | start_t = _write_and_time('\tUpdating user factors...') 65 | U = recompute_factors(V, S, lambda_U_reg, dtype=dtype, *args, **kwargs) 66 | 67 | if verbose: 68 | print('\r\tUpdating user factors: time=%.2f' 69 | % (time.time() - start_t)) 70 | if not fixed_item_embeddings: 71 | start_t = _write_and_time('\tUpdating item factors...') 72 | if not fixed_item_embeddings: 73 | V = recompute_factors(U, ST, lambda_V_reg, X=X, dtype=dtype, 74 | *args, **kwargs) 75 | if verbose and not fixed_item_embeddings: 76 | print('\r\tUpdating item factors: time=%.2f' 77 | % (time.time() - start_t)) 78 | if vad_data is not None and not fixed_item_embeddings: 79 | vad_ndcg = rec_eval.normalized_dcg_at_k(S, vad_data, U, V, 80 | k=100, 81 | batch_users=5000) 82 | if verbose: 83 | print("\tValidation NDCG@k: %.5f" % vad_ndcg) 84 | sys.stdout.flush() 85 | if old_ndcg > vad_ndcg: 86 | break # we will not save the parameter for this iteration 87 | old_ndcg = vad_ndcg 88 | 89 | return U, V, old_ndcg 90 | 91 | 92 | def _pred_loglikeli(U, V, dtype, X_new=None, rows_new=None, cols_new=None): 93 | X_pred = _inner(U, V, rows_new, cols_new, dtype) 94 | pred_ll = np.mean((X_new.data - X_pred)**2) 95 | return pred_ll 96 | 97 | 98 | def _write_and_time(s): 99 | sys.stdout.write(s) 100 | sys.stdout.flush() 101 | return time.time() 102 | 103 | 104 | def _inner(U, V, rows, cols, dtype): 105 | n_ratings = rows.size 106 | n_components = U.shape[1] 107 | assert V.shape[1] == n_components 108 | data = np.empty(n_ratings, dtype=dtype) 109 | code = r""" 110 | for (int i = 0; i < n_ratings; i++) { 111 | data[i] = 0.0; 112 | for (int j = 0; j < n_components; j++) { 113 | data[i] += U[rows[i] * n_components + j] * V[cols[i] * n_components + j]; 114 | } 115 | } 116 | """ 117 | weave.inline(code, ['data', 'U', 'V', 'rows', 'cols', 'n_ratings', 118 | 'n_components']) 119 | return data 120 | -------------------------------------------------------------------------------- /src/WMF_ML20M.ipynb: -------------------------------------------------------------------------------- 1 | { 2 | "cells": [ 3 | { 4 | "cell_type": "markdown", 5 | "metadata": {}, 6 | "source": [ 7 | "# Fit WMF (weighted matrix factorization) to the binarized ML20M" 8 | ] 9 | }, 10 | { 11 | "cell_type": "code", 12 | "execution_count": 1, 13 | "metadata": { 14 | "collapsed": false 15 | }, 16 | "outputs": [], 17 | "source": [ 18 | "import itertools\n", 19 | "import os\n", 20 | "import sys\n", 21 | "os.environ['OPENBLAS_NUM_THREADS'] = '1'\n", 22 | "\n", 23 | "import numpy as np\n", 24 | "import pandas as pd\n", 25 | "from scipy import sparse" 26 | ] 27 | }, 28 | { 29 | "cell_type": "code", 30 | "execution_count": 2, 31 | "metadata": { 32 | "collapsed": true 33 | }, 34 | "outputs": [], 35 | "source": [ 36 | "import content_wmf\n", 37 | "import batched_inv_joblib\n", 38 | "import rec_eval" 39 | ] 40 | }, 41 | { 42 | "cell_type": "markdown", 43 | "metadata": {}, 44 | "source": [ 45 | "### Load pre-processed data" 46 | ] 47 | }, 48 | { 49 | "cell_type": "markdown", 50 | "metadata": {}, 51 | "source": [ 52 | "Change this to wherever you saved the pre-processed data following [this notebook](./preprocess_ML20M.ipynb)." 53 | ] 54 | }, 55 | { 56 | "cell_type": "code", 57 | "execution_count": 3, 58 | "metadata": { 59 | "collapsed": true 60 | }, 61 | "outputs": [], 62 | "source": [ 63 | "DATA_DIR = '/hdd2/dawen/data/ml-20m/pro/'" 64 | ] 65 | }, 66 | { 67 | "cell_type": "code", 68 | "execution_count": 4, 69 | "metadata": { 70 | "collapsed": true 71 | }, 72 | "outputs": [], 73 | "source": [ 74 | "unique_uid = list()\n", 75 | "with open(os.path.join(DATA_DIR, 'unique_uid.txt'), 'r') as f:\n", 76 | " for line in f:\n", 77 | " unique_uid.append(line.strip())\n", 78 | " \n", 79 | "unique_sid = list()\n", 80 | "with open(os.path.join(DATA_DIR, 'unique_sid.txt'), 'r') as f:\n", 81 | " for line in f:\n", 82 | " unique_sid.append(line.strip())" 83 | ] 84 | }, 85 | { 86 | "cell_type": "code", 87 | "execution_count": 5, 88 | "metadata": { 89 | "collapsed": false 90 | }, 91 | "outputs": [ 92 | { 93 | "name": "stdout", 94 | "output_type": "stream", 95 | "text": [ 96 | "111148 11711\n" 97 | ] 98 | } 99 | ], 100 | "source": [ 101 | "n_items = len(unique_sid)\n", 102 | "n_users = len(unique_uid)\n", 103 | "\n", 104 | "print n_users, n_items" 105 | ] 106 | }, 107 | { 108 | "cell_type": "code", 109 | "execution_count": 6, 110 | "metadata": { 111 | "collapsed": false 112 | }, 113 | "outputs": [], 114 | "source": [ 115 | "def load_data(csv_file, shape=(n_users, n_items)):\n", 116 | " tp = pd.read_csv(csv_file)\n", 117 | " timestamps, rows, cols = np.array(tp['timestamp']), np.array(tp['uid']), np.array(tp['sid'])\n", 118 | " seq = np.concatenate((rows[:, None], cols[:, None], np.ones((rows.size, 1), dtype='int'), timestamps[:, None]), axis=1)\n", 119 | " data = sparse.csr_matrix((np.ones_like(rows), (rows, cols)), dtype=np.int16, shape=shape)\n", 120 | " return data, seq" 121 | ] 122 | }, 123 | { 124 | "cell_type": "code", 125 | "execution_count": 7, 126 | "metadata": { 127 | "collapsed": false 128 | }, 129 | "outputs": [], 130 | "source": [ 131 | "train_data, train_raw = load_data(os.path.join(DATA_DIR, 'train.csv'))" 132 | ] 133 | }, 134 | { 135 | "cell_type": "code", 136 | "execution_count": 8, 137 | "metadata": { 138 | "collapsed": true 139 | }, 140 | "outputs": [], 141 | "source": [ 142 | "vad_data, vad_raw = load_data(os.path.join(DATA_DIR, 'validation.csv'))" 143 | ] 144 | }, 145 | { 146 | "cell_type": "markdown", 147 | "metadata": {}, 148 | "source": [ 149 | "### Train the model" 150 | ] 151 | }, 152 | { 153 | "cell_type": "code", 154 | "execution_count": 9, 155 | "metadata": { 156 | "collapsed": true 157 | }, 158 | "outputs": [], 159 | "source": [ 160 | "num_factors = 100\n", 161 | "num_iters = 50\n", 162 | "batch_size = 1000\n", 163 | "\n", 164 | "n_jobs = 4\n", 165 | "lam_theta = lam_beta = 1e-5" 166 | ] 167 | }, 168 | { 169 | "cell_type": "code", 170 | "execution_count": null, 171 | "metadata": { 172 | "collapsed": false 173 | }, 174 | "outputs": [], 175 | "source": [ 176 | "best_ndcg = -np.inf\n", 177 | "U_best = None\n", 178 | "V_best = None\n", 179 | "best_alpha = 0\n", 180 | "\n", 181 | "for alpha in [2, 5, 10, 30, 50]: \n", 182 | " S = content_wmf.linear_surplus_confidence_matrix(train_data, alpha=alpha)\n", 183 | "\n", 184 | " U, V, vad_ndcg = content_wmf.factorize(S, num_factors, vad_data=vad_data, num_iters=num_iters, \n", 185 | " init_std=0.01, lambda_U_reg=lam_theta, lambda_V_reg=lam_beta, \n", 186 | " dtype='float32', random_state=98765, verbose=False, \n", 187 | " recompute_factors=batched_inv_joblib.recompute_factors_batched, \n", 188 | " batch_size=batch_size, n_jobs=n_jobs)\n", 189 | " if vad_ndcg > best_ndcg:\n", 190 | " best_ndcg = vad_ndcg\n", 191 | " U_best = U.copy()\n", 192 | " V_best = V.copy()\n", 193 | " best_alpha = alpha" 194 | ] 195 | }, 196 | { 197 | "cell_type": "code", 198 | "execution_count": 11, 199 | "metadata": { 200 | "collapsed": false 201 | }, 202 | "outputs": [ 203 | { 204 | "name": "stdout", 205 | "output_type": "stream", 206 | "text": [ 207 | "10 0.35510611042\n" 208 | ] 209 | } 210 | ], 211 | "source": [ 212 | "print best_alpha, best_ndcg" 213 | ] 214 | }, 215 | { 216 | "cell_type": "code", 217 | "execution_count": 12, 218 | "metadata": { 219 | "collapsed": false 220 | }, 221 | "outputs": [], 222 | "source": [ 223 | "test_data, _ = load_data(os.path.join(DATA_DIR, 'test.csv'))\n", 224 | "test_data.data = np.ones_like(test_data.data)" 225 | ] 226 | }, 227 | { 228 | "cell_type": "code", 229 | "execution_count": 13, 230 | "metadata": { 231 | "collapsed": false 232 | }, 233 | "outputs": [ 234 | { 235 | "name": "stdout", 236 | "output_type": "stream", 237 | "text": [ 238 | "Test Recall@20: 0.1333\n", 239 | "Test Recall@50: 0.1647\n", 240 | "Test NDCG@100: 0.1602\n", 241 | "Test MAP@100: 0.0473\n" 242 | ] 243 | } 244 | ], 245 | "source": [ 246 | "# alpha = 10 gives the best validation performance\n", 247 | "print 'Test Recall@20: %.4f' % rec_eval.recall_at_k(train_data, test_data, U_best, V_best, k=20, vad_data=vad_data)\n", 248 | "print 'Test Recall@50: %.4f' % rec_eval.recall_at_k(train_data, test_data, U_best, V_best, k=50, vad_data=vad_data)\n", 249 | "print 'Test NDCG@100: %.4f' % rec_eval.normalized_dcg_at_k(train_data, test_data, U_best, V_best, k=100, vad_data=vad_data)\n", 250 | "print 'Test MAP@100: %.4f' % rec_eval.map_at_k(train_data, test_data, U_best, V_best, k=100, vad_data=vad_data)" 251 | ] 252 | }, 253 | { 254 | "cell_type": "code", 255 | "execution_count": 14, 256 | "metadata": { 257 | "collapsed": true 258 | }, 259 | "outputs": [], 260 | "source": [ 261 | "np.savez('WMF_K100_ML20M.npz', U=U_best, V=V_best)" 262 | ] 263 | }, 264 | { 265 | "cell_type": "code", 266 | "execution_count": null, 267 | "metadata": { 268 | "collapsed": true 269 | }, 270 | "outputs": [], 271 | "source": [] 272 | } 273 | ], 274 | "metadata": { 275 | "kernelspec": { 276 | "display_name": "Python 2", 277 | "language": "python", 278 | "name": "python2" 279 | }, 280 | "language_info": { 281 | "codemirror_mode": { 282 | "name": "ipython", 283 | "version": 2 284 | }, 285 | "file_extension": ".py", 286 | "mimetype": "text/x-python", 287 | "name": "python", 288 | "nbconvert_exporter": "python", 289 | "pygments_lexer": "ipython2", 290 | "version": "2.7.6" 291 | } 292 | }, 293 | "nbformat": 4, 294 | "nbformat_minor": 0 295 | } 296 | -------------------------------------------------------------------------------- /LICENSE: -------------------------------------------------------------------------------- 1 | Apache License 2 | Version 2.0, January 2004 3 | http://www.apache.org/licenses/ 4 | 5 | TERMS AND CONDITIONS FOR USE, REPRODUCTION, AND DISTRIBUTION 6 | 7 | 1. Definitions. 8 | 9 | "License" shall mean the terms and conditions for use, reproduction, 10 | and distribution as defined by Sections 1 through 9 of this document. 11 | 12 | "Licensor" shall mean the copyright owner or entity authorized by 13 | the copyright owner that is granting the License. 14 | 15 | "Legal Entity" shall mean the union of the acting entity and all 16 | other entities that control, are controlled by, or are under common 17 | control with that entity. For the purposes of this definition, 18 | "control" means (i) the power, direct or indirect, to cause the 19 | direction or management of such entity, whether by contract or 20 | otherwise, or (ii) ownership of fifty percent (50%) or more of the 21 | outstanding shares, or (iii) beneficial ownership of such entity. 22 | 23 | "You" (or "Your") shall mean an individual or Legal Entity 24 | exercising permissions granted by this License. 25 | 26 | "Source" form shall mean the preferred form for making modifications, 27 | including but not limited to software source code, documentation 28 | source, and configuration files. 29 | 30 | "Object" form shall mean any form resulting from mechanical 31 | transformation or translation of a Source form, including but 32 | not limited to compiled object code, generated documentation, 33 | and conversions to other media types. 34 | 35 | "Work" shall mean the work of authorship, whether in Source or 36 | Object form, made available under the License, as indicated by a 37 | copyright notice that is included in or attached to the work 38 | (an example is provided in the Appendix below). 39 | 40 | "Derivative Works" shall mean any work, whether in Source or Object 41 | form, that is based on (or derived from) the Work and for which the 42 | editorial revisions, annotations, elaborations, or other modifications 43 | represent, as a whole, an original work of authorship. For the purposes 44 | of this License, Derivative Works shall not include works that remain 45 | separable from, or merely link (or bind by name) to the interfaces of, 46 | the Work and Derivative Works thereof. 47 | 48 | "Contribution" shall mean any work of authorship, including 49 | the original version of the Work and any modifications or additions 50 | to that Work or Derivative Works thereof, that is intentionally 51 | submitted to Licensor for inclusion in the Work by the copyright owner 52 | or by an individual or Legal Entity authorized to submit on behalf of 53 | the copyright owner. For the purposes of this definition, "submitted" 54 | means any form of electronic, verbal, or written communication sent 55 | to the Licensor or its representatives, including but not limited to 56 | communication on electronic mailing lists, source code control systems, 57 | and issue tracking systems that are managed by, or on behalf of, the 58 | Licensor for the purpose of discussing and improving the Work, but 59 | excluding communication that is conspicuously marked or otherwise 60 | designated in writing by the copyright owner as "Not a Contribution." 61 | 62 | "Contributor" shall mean Licensor and any individual or Legal Entity 63 | on behalf of whom a Contribution has been received by Licensor and 64 | subsequently incorporated within the Work. 65 | 66 | 2. Grant of Copyright License. Subject to the terms and conditions of 67 | this License, each Contributor hereby grants to You a perpetual, 68 | worldwide, non-exclusive, no-charge, royalty-free, irrevocable 69 | copyright license to reproduce, prepare Derivative Works of, 70 | publicly display, publicly perform, sublicense, and distribute the 71 | Work and such Derivative Works in Source or Object form. 72 | 73 | 3. Grant of Patent License. Subject to the terms and conditions of 74 | this License, each Contributor hereby grants to You a perpetual, 75 | worldwide, non-exclusive, no-charge, royalty-free, irrevocable 76 | (except as stated in this section) patent license to make, have made, 77 | use, offer to sell, sell, import, and otherwise transfer the Work, 78 | where such license applies only to those patent claims licensable 79 | by such Contributor that are necessarily infringed by their 80 | Contribution(s) alone or by combination of their Contribution(s) 81 | with the Work to which such Contribution(s) was submitted. If You 82 | institute patent litigation against any entity (including a 83 | cross-claim or counterclaim in a lawsuit) alleging that the Work 84 | or a Contribution incorporated within the Work constitutes direct 85 | or contributory patent infringement, then any patent licenses 86 | granted to You under this License for that Work shall terminate 87 | as of the date such litigation is filed. 88 | 89 | 4. Redistribution. You may reproduce and distribute copies of the 90 | Work or Derivative Works thereof in any medium, with or without 91 | modifications, and in Source or Object form, provided that You 92 | meet the following conditions: 93 | 94 | (a) You must give any other recipients of the Work or 95 | Derivative Works a copy of this License; and 96 | 97 | (b) You must cause any modified files to carry prominent notices 98 | stating that You changed the files; and 99 | 100 | (c) You must retain, in the Source form of any Derivative Works 101 | that You distribute, all copyright, patent, trademark, and 102 | attribution notices from the Source form of the Work, 103 | excluding those notices that do not pertain to any part of 104 | the Derivative Works; and 105 | 106 | (d) If the Work includes a "NOTICE" text file as part of its 107 | distribution, then any Derivative Works that You distribute must 108 | include a readable copy of the attribution notices contained 109 | within such NOTICE file, excluding those notices that do not 110 | pertain to any part of the Derivative Works, in at least one 111 | of the following places: within a NOTICE text file distributed 112 | as part of the Derivative Works; within the Source form or 113 | documentation, if provided along with the Derivative Works; or, 114 | within a display generated by the Derivative Works, if and 115 | wherever such third-party notices normally appear. The contents 116 | of the NOTICE file are for informational purposes only and 117 | do not modify the License. You may add Your own attribution 118 | notices within Derivative Works that You distribute, alongside 119 | or as an addendum to the NOTICE text from the Work, provided 120 | that such additional attribution notices cannot be construed 121 | as modifying the License. 122 | 123 | You may add Your own copyright statement to Your modifications and 124 | may provide additional or different license terms and conditions 125 | for use, reproduction, or distribution of Your modifications, or 126 | for any such Derivative Works as a whole, provided Your use, 127 | reproduction, and distribution of the Work otherwise complies with 128 | the conditions stated in this License. 129 | 130 | 5. Submission of Contributions. Unless You explicitly state otherwise, 131 | any Contribution intentionally submitted for inclusion in the Work 132 | by You to the Licensor shall be under the terms and conditions of 133 | this License, without any additional terms or conditions. 134 | Notwithstanding the above, nothing herein shall supersede or modify 135 | the terms of any separate license agreement you may have executed 136 | with Licensor regarding such Contributions. 137 | 138 | 6. Trademarks. This License does not grant permission to use the trade 139 | names, trademarks, service marks, or product names of the Licensor, 140 | except as required for reasonable and customary use in describing the 141 | origin of the Work and reproducing the content of the NOTICE file. 142 | 143 | 7. Disclaimer of Warranty. Unless required by applicable law or 144 | agreed to in writing, Licensor provides the Work (and each 145 | Contributor provides its Contributions) on an "AS IS" BASIS, 146 | WITHOUT WARRANTIES OR CONDITIONS OF ANY KIND, either express or 147 | implied, including, without limitation, any warranties or conditions 148 | of TITLE, NON-INFRINGEMENT, MERCHANTABILITY, or FITNESS FOR A 149 | PARTICULAR PURPOSE. You are solely responsible for determining the 150 | appropriateness of using or redistributing the Work and assume any 151 | risks associated with Your exercise of permissions under this License. 152 | 153 | 8. Limitation of Liability. In no event and under no legal theory, 154 | whether in tort (including negligence), contract, or otherwise, 155 | unless required by applicable law (such as deliberate and grossly 156 | negligent acts) or agreed to in writing, shall any Contributor be 157 | liable to You for damages, including any direct, indirect, special, 158 | incidental, or consequential damages of any character arising as a 159 | result of this License or out of the use or inability to use the 160 | Work (including but not limited to damages for loss of goodwill, 161 | work stoppage, computer failure or malfunction, or any and all 162 | other commercial damages or losses), even if such Contributor 163 | has been advised of the possibility of such damages. 164 | 165 | 9. Accepting Warranty or Additional Liability. While redistributing 166 | the Work or Derivative Works thereof, You may choose to offer, 167 | and charge a fee for, acceptance of support, warranty, indemnity, 168 | or other liability obligations and/or rights consistent with this 169 | License. However, in accepting such obligations, You may act only 170 | on Your own behalf and on Your sole responsibility, not on behalf 171 | of any other Contributor, and only if You agree to indemnify, 172 | defend, and hold each Contributor harmless for any liability 173 | incurred by, or claims asserted against, such Contributor by reason 174 | of your accepting any such warranty or additional liability. 175 | 176 | END OF TERMS AND CONDITIONS 177 | 178 | APPENDIX: How to apply the Apache License to your work. 179 | 180 | To apply the Apache License to your work, attach the following 181 | boilerplate notice, with the fields enclosed by brackets "{}" 182 | replaced with your own identifying information. (Don't include 183 | the brackets!) The text should be enclosed in the appropriate 184 | comment syntax for the file format. We also recommend that a 185 | file or class name and description of purpose be included on the 186 | same "printed page" as the copyright notice for easier 187 | identification within third-party archives. 188 | 189 | Copyright {yyyy} {name of copyright owner} 190 | 191 | Licensed under the Apache License, Version 2.0 (the "License"); 192 | you may not use this file except in compliance with the License. 193 | You may obtain a copy of the License at 194 | 195 | http://www.apache.org/licenses/LICENSE-2.0 196 | 197 | Unless required by applicable law or agreed to in writing, software 198 | distributed under the License is distributed on an "AS IS" BASIS, 199 | WITHOUT WARRANTIES OR CONDITIONS OF ANY KIND, either express or implied. 200 | See the License for the specific language governing permissions and 201 | limitations under the License. 202 | -------------------------------------------------------------------------------- /src/rec_eval.py: -------------------------------------------------------------------------------- 1 | import bottleneck as bn 2 | import numpy as np 3 | 4 | from scipy import sparse 5 | 6 | 7 | """ 8 | All the data should be in the shape of (n_users, n_items) 9 | All the latent factors should in the shape of (n_users/n_items, n_components) 10 | 11 | 1. train_data refers to the data that was used to train the model 12 | 2. heldout_data refers to the data that was used for evaluation (could be test 13 | set or validation set) 14 | 3. vad_data refers to the data that should be excluded as validation set, which 15 | should only be used when calculating test scores 16 | 17 | """ 18 | 19 | 20 | def prec_at_k(train_data, heldout_data, U, V, batch_users=5000, k=20, 21 | mu=None, vad_data=None, agg=np.nanmean): 22 | n_users = train_data.shape[0] 23 | res = list() 24 | for user_idx in user_idx_generator(n_users, batch_users): 25 | res.append(precision_at_k_batch(train_data, heldout_data, 26 | U, V.T, user_idx, k=k, 27 | mu=mu, vad_data=vad_data)) 28 | mn_prec = np.hstack(res) 29 | if callable(agg): 30 | return agg(mn_prec) 31 | return mn_prec 32 | 33 | 34 | def recall_at_k(train_data, heldout_data, U, V, batch_users=5000, k=20, 35 | mu=None, vad_data=None, agg=np.nanmean): 36 | n_users = train_data.shape[0] 37 | res = list() 38 | for user_idx in user_idx_generator(n_users, batch_users): 39 | res.append(recall_at_k_batch(train_data, heldout_data, 40 | U, V.T, user_idx, k=k, 41 | mu=mu, vad_data=vad_data)) 42 | mn_recall = np.hstack(res) 43 | if callable(agg): 44 | return agg(mn_recall) 45 | return mn_recall 46 | 47 | 48 | def ric_rank_at_k(train_data, heldout_data, U, V, batch_users=5000, k=5, 49 | mu=None, vad_data=None): 50 | n_users = train_data.shape[0] 51 | res = list() 52 | for user_idx in user_idx_generator(n_users, batch_users): 53 | res.append(mean_rrank_at_k_batch(train_data, heldout_data, 54 | U, V.T, user_idx, k=k, 55 | mu=mu, vad_data=vad_data)) 56 | mrrank = np.hstack(res) 57 | return mrrank[mrrank > 0].mean() 58 | 59 | 60 | def mean_perc_rank(train_data, heldout_data, U, V, batch_users=5000, 61 | mu=None, vad_data=None): 62 | n_users = train_data.shape[0] 63 | mpr = 0 64 | for user_idx in user_idx_generator(n_users, batch_users): 65 | mpr += mean_perc_rank_batch(train_data, heldout_data, U, V.T, user_idx, 66 | mu=mu, vad_data=vad_data) 67 | mpr /= heldout_data.sum() 68 | return mpr 69 | 70 | 71 | def normalized_dcg(train_data, heldout_data, U, V, batch_users=5000, 72 | mu=None, vad_data=None, agg=np.nanmean): 73 | n_users = train_data.shape[0] 74 | res = list() 75 | for user_idx in user_idx_generator(n_users, batch_users): 76 | res.append(NDCG_binary_batch(train_data, heldout_data, U, V.T, 77 | user_idx, mu=mu, vad_data=vad_data)) 78 | ndcg = np.hstack(res) 79 | if callable(agg): 80 | return agg(ndcg) 81 | return ndcg 82 | 83 | 84 | def normalized_dcg_at_k(train_data, heldout_data, U, V, batch_users=5000, 85 | k=100, mu=None, vad_data=None, agg=np.nanmean): 86 | 87 | n_users = train_data.shape[0] 88 | res = list() 89 | for user_idx in user_idx_generator(n_users, batch_users): 90 | res.append(NDCG_binary_at_k_batch(train_data, heldout_data, U, V.T, 91 | user_idx, k=k, mu=mu, 92 | vad_data=vad_data)) 93 | ndcg = np.hstack(res) 94 | if callable(agg): 95 | return agg(ndcg) 96 | return ndcg 97 | 98 | 99 | def map_at_k(train_data, heldout_data, U, V, batch_users=5000, k=100, mu=None, 100 | vad_data=None, agg=np.nanmean): 101 | 102 | n_users = train_data.shape[0] 103 | res = list() 104 | for user_idx in user_idx_generator(n_users, batch_users): 105 | res.append(MAP_at_k_batch(train_data, heldout_data, U, V.T, user_idx, 106 | k=k, mu=mu, vad_data=vad_data)) 107 | map = np.hstack(res) 108 | if callable(agg): 109 | return agg(map) 110 | return map 111 | 112 | 113 | # helper functions # 114 | 115 | def user_idx_generator(n_users, batch_users): 116 | ''' helper function to generate the user index to loop through the dataset 117 | ''' 118 | for start in xrange(0, n_users, batch_users): 119 | end = min(n_users, start + batch_users) 120 | yield slice(start, end) 121 | 122 | 123 | def _make_prediction(train_data, Et, Eb, user_idx, batch_users, mu=None, 124 | vad_data=None): 125 | n_songs = train_data.shape[1] 126 | # exclude examples from training and validation (if any) 127 | item_idx = np.zeros((batch_users, n_songs), dtype=bool) 128 | item_idx[train_data[user_idx].nonzero()] = True 129 | if vad_data is not None: 130 | item_idx[vad_data[user_idx].nonzero()] = True 131 | X_pred = Et[user_idx].dot(Eb) 132 | if mu is not None: 133 | if isinstance(mu, np.ndarray): 134 | assert mu.size == n_songs # mu_i 135 | X_pred *= mu 136 | elif isinstance(mu, dict): # func(mu_ui) 137 | params, func = mu['params'], mu['func'] 138 | args = [params[0][user_idx], params[1]] 139 | if len(params) > 2: # for bias term in document or length-scale 140 | args += [params[2][user_idx]] 141 | if not callable(func): 142 | raise TypeError("expecting a callable function") 143 | X_pred *= func(*args) 144 | else: 145 | raise ValueError("unsupported mu type") 146 | X_pred[item_idx] = -np.inf 147 | return X_pred 148 | 149 | 150 | def precision_at_k_batch(train_data, heldout_data, Et, Eb, user_idx, 151 | k=20, normalize=True, mu=None, vad_data=None): 152 | batch_users = user_idx.stop - user_idx.start 153 | 154 | X_pred = _make_prediction(train_data, Et, Eb, user_idx, 155 | batch_users, mu=mu, vad_data=vad_data) 156 | idx = bn.argpartition(-X_pred, k, axis=1) 157 | X_pred_binary = np.zeros_like(X_pred, dtype=bool) 158 | X_pred_binary[np.arange(batch_users)[:, np.newaxis], idx[:, :k]] = True 159 | 160 | X_true_binary = (heldout_data[user_idx] > 0).toarray() 161 | tmp = (np.logical_and(X_true_binary, X_pred_binary).sum(axis=1)).astype( 162 | np.float32) 163 | 164 | if normalize: 165 | precision = tmp / np.minimum(k, X_true_binary.sum(axis=1)) 166 | else: 167 | precision = tmp / k 168 | return precision 169 | 170 | 171 | def recall_at_k_batch(train_data, heldout_data, Et, Eb, user_idx, 172 | k=20, normalize=True, mu=None, vad_data=None): 173 | batch_users = user_idx.stop - user_idx.start 174 | 175 | X_pred = _make_prediction(train_data, Et, Eb, user_idx, 176 | batch_users, mu=mu, vad_data=vad_data) 177 | idx = bn.argpartition(-X_pred, k, axis=1) 178 | X_pred_binary = np.zeros_like(X_pred, dtype=bool) 179 | X_pred_binary[np.arange(batch_users)[:, np.newaxis], idx[:, :k]] = True 180 | 181 | X_true_binary = (heldout_data[user_idx] > 0).toarray() 182 | tmp = (np.logical_and(X_true_binary, X_pred_binary).sum(axis=1)).astype( 183 | np.float32) 184 | recall = tmp / np.minimum(k, X_true_binary.sum(axis=1)) 185 | return recall 186 | 187 | 188 | def mean_rrank_at_k_batch(train_data, heldout_data, Et, Eb, 189 | user_idx, k=5, mu=None, vad_data=None): 190 | ''' 191 | mean reciprocal rank@k: For each user, make predictions and rank for 192 | all the items. Then calculate the mean reciprocal rank for the top K that 193 | are in the held-out set. 194 | ''' 195 | batch_users = user_idx.stop - user_idx.start 196 | 197 | X_pred = _make_prediction(train_data, Et, Eb, user_idx, 198 | batch_users, mu=mu, vad_data=vad_data) 199 | all_rrank = 1. / (np.argsort(np.argsort(-X_pred, axis=1), axis=1) + 1) 200 | X_true_binary = (heldout_data[user_idx] > 0).toarray() 201 | 202 | heldout_rrank = X_true_binary * all_rrank 203 | top_k = bn.partsort(-heldout_rrank, k, axis=1) 204 | return -top_k[:, :k].mean(axis=1) 205 | 206 | 207 | def NDCG_binary_batch(train_data, heldout_data, Et, Eb, user_idx, 208 | mu=None, vad_data=None): 209 | ''' 210 | normalized discounted cumulative gain for binary relevance 211 | ''' 212 | batch_users = user_idx.stop - user_idx.start 213 | n_items = train_data.shape[1] 214 | 215 | X_pred = _make_prediction(train_data, Et, Eb, user_idx, 216 | batch_users, mu=mu, vad_data=vad_data) 217 | all_rank = np.argsort(np.argsort(-X_pred, axis=1), axis=1) 218 | # build the discount template 219 | tp = 1. / np.log2(np.arange(2, n_items + 2)) 220 | all_disc = tp[all_rank] 221 | 222 | X_true_binary = (heldout_data[user_idx] > 0).tocoo() 223 | disc = sparse.csr_matrix((all_disc[X_true_binary.row, X_true_binary.col], 224 | (X_true_binary.row, X_true_binary.col)), 225 | shape=all_disc.shape) 226 | DCG = np.array(disc.sum(axis=1)).ravel() 227 | IDCG = np.array([tp[:n].sum() 228 | for n in heldout_data[user_idx].getnnz(axis=1)]) 229 | return DCG / IDCG 230 | 231 | 232 | def NDCG_binary_at_k_batch(train_data, heldout_data, Et, Eb, user_idx, 233 | mu=None, k=100, vad_data=None): 234 | ''' 235 | normalized discounted cumulative gain@k for binary relevance 236 | ASSUMPTIONS: all the 0's in heldout_data indicate 0 relevance 237 | ''' 238 | batch_users = user_idx.stop - user_idx.start 239 | 240 | X_pred = _make_prediction(train_data, Et, Eb, user_idx, 241 | batch_users, mu=mu, vad_data=vad_data) 242 | idx_topk_part = bn.argpartition(-X_pred, k, axis=1) 243 | topk_part = X_pred[np.arange(batch_users)[:, np.newaxis], 244 | idx_topk_part[:, :k]] 245 | idx_part = np.argsort(-topk_part, axis=1) 246 | # X_pred[np.arange(batch_users)[:, np.newaxis], idx_topk] is the sorted 247 | # topk predicted score 248 | idx_topk = idx_topk_part[np.arange(batch_users)[:, np.newaxis], idx_part] 249 | # build the discount template 250 | tp = 1. / np.log2(np.arange(2, k + 2)) 251 | 252 | heldout_batch = heldout_data[user_idx] 253 | DCG = (heldout_batch[np.arange(batch_users)[:, np.newaxis], 254 | idx_topk].toarray() * tp).sum(axis=1) 255 | IDCG = np.array([(tp[:min(n, k)]).sum() 256 | for n in heldout_batch.getnnz(axis=1)]) 257 | return DCG / IDCG 258 | 259 | 260 | def MAP_at_k_batch(train_data, heldout_data, Et, Eb, user_idx, mu=None, k=100, 261 | vad_data=None): 262 | ''' 263 | mean average precision@k 264 | ''' 265 | batch_users = user_idx.stop - user_idx.start 266 | 267 | X_pred = _make_prediction(train_data, Et, Eb, user_idx, batch_users, mu=mu, 268 | vad_data=vad_data) 269 | idx_topk_part = bn.argpartition(-X_pred, k, axis=1) 270 | topk_part = X_pred[np.arange(batch_users)[:, np.newaxis], 271 | idx_topk_part[:, :k]] 272 | idx_part = np.argsort(-topk_part, axis=1) 273 | # X_pred[np.arange(batch_users)[:, np.newaxis], idx_topk] is the sorted 274 | # topk predicted score 275 | idx_topk = idx_topk_part[np.arange(batch_users)[:, np.newaxis], idx_part] 276 | 277 | aps = np.zeros(batch_users) 278 | for i, idx in enumerate(xrange(user_idx.start, user_idx.stop)): 279 | actual = heldout_data[idx].nonzero()[1] 280 | if len(actual) > 0: 281 | predicted = idx_topk[i] 282 | aps[i] = apk(actual, predicted, k=k) 283 | else: 284 | aps[i] = np.nan 285 | return aps 286 | 287 | 288 | def mean_perc_rank_batch(train_data, heldout_data, Et, Eb, user_idx, 289 | mu=None, vad_data=None): 290 | ''' 291 | mean percentile rank for a batch of users 292 | MPR of the full set is the sum of batch MPR's divided by the sum of all the 293 | feedbacks. (Eq. 8 in Hu et al.) 294 | This metric not necessarily constrains the data to be binary 295 | ''' 296 | batch_users = user_idx.stop - user_idx.start 297 | 298 | X_pred = _make_prediction(train_data, Et, Eb, user_idx, batch_users, 299 | mu=mu, vad_data=vad_data) 300 | all_perc = np.argsort(np.argsort(-X_pred, axis=1), axis=1) / \ 301 | np.isfinite(X_pred).sum(axis=1, keepdims=True).astype(np.float32) 302 | perc_batch = (all_perc[heldout_data[user_idx].nonzero()] * 303 | heldout_data[user_idx].data).sum() 304 | return perc_batch 305 | 306 | 307 | ## steal from https://github.com/benhamner/Metrics/blob/master/Python/ml_metrics/average_precision.py 308 | def apk(actual, predicted, k=100): 309 | """ 310 | Computes the average precision at k. 311 | This function computes the average prescision at k between two lists of 312 | items. 313 | Parameters 314 | ---------- 315 | actual : list 316 | A list of elements that are to be predicted (order doesn't matter) 317 | predicted : list 318 | A list of predicted elements (order does matter) 319 | k : int, optional 320 | The maximum number of predicted elements 321 | Returns 322 | ------- 323 | score : double 324 | The average precision at k over the input lists 325 | """ 326 | if len(predicted)>k: 327 | predicted = predicted[:k] 328 | 329 | score = 0.0 330 | num_hits = 0.0 331 | 332 | for i,p in enumerate(predicted): 333 | if p in actual: #and p not in predicted[:i]: # not necessary for us since we will not make duplicated recs 334 | num_hits += 1.0 335 | score += num_hits / (i+1.0) 336 | 337 | # we handle this part before making the function call 338 | #if not actual: 339 | # return np.nan 340 | 341 | return score / min(len(actual), k) 342 | -------------------------------------------------------------------------------- /src/cofacto.py: -------------------------------------------------------------------------------- 1 | ''' 2 | 3 | Co-factorize the user click matrix and item co-occurrence matrix 4 | 5 | CREATED: 2016-03-22 13:21:35 by Dawen Liang 6 | 7 | ''' 8 | 9 | import os 10 | import sys 11 | import time 12 | 13 | import numpy as np 14 | from numpy import linalg as LA 15 | 16 | from joblib import Parallel, delayed 17 | from sklearn.base import BaseEstimator, TransformerMixin 18 | 19 | import rec_eval 20 | 21 | 22 | class CoFacto(BaseEstimator, TransformerMixin): 23 | def __init__(self, n_components=100, max_iter=10, batch_size=1000, 24 | init_std=0.01, dtype='float32', n_jobs=8, random_state=None, 25 | save_params=False, save_dir='.', early_stopping=False, 26 | verbose=False, **kwargs): 27 | ''' 28 | CoFacto 29 | 30 | Parameters 31 | --------- 32 | n_components : int 33 | Number of latent factors 34 | max_iter : int 35 | Maximal number of iterations to perform 36 | batch_size: int 37 | Batch size to perform parallel update 38 | init_std: float 39 | The latent factors will be initialized as Normal(0, init_std**2) 40 | dtype: str or type 41 | Data-type for the parameters, default 'float32' (np.float32) 42 | n_jobs: int 43 | Number of parallel jobs to update latent factors 44 | random_state : int or RandomState 45 | Pseudo random number generator used for sampling 46 | save_params: bool 47 | Whether to save parameters after each iteration 48 | save_dir: str 49 | The directory to save the parameters 50 | early_stopping: bool 51 | Whether to early stop the training by monitoring performance on 52 | validation set 53 | verbose : bool 54 | Whether to show progress during model fitting 55 | **kwargs: dict 56 | Model hyperparameters 57 | ''' 58 | self.n_components = n_components 59 | self.max_iter = max_iter 60 | self.batch_size = batch_size 61 | self.init_std = init_std 62 | self.dtype = dtype 63 | self.n_jobs = n_jobs 64 | self.random_state = random_state 65 | self.save_params = save_params 66 | self.save_dir = save_dir 67 | self.early_stopping = early_stopping 68 | self.verbose = verbose 69 | 70 | if type(self.random_state) is int: 71 | np.random.seed(self.random_state) 72 | elif self.random_state is not None: 73 | np.random.setstate(self.random_state) 74 | 75 | self._parse_kwargs(**kwargs) 76 | 77 | def _parse_kwargs(self, **kwargs): 78 | ''' Model hyperparameters 79 | 80 | Parameters 81 | --------- 82 | lambda_theta, lambda_beta, lambda_gamma: float 83 | Regularization parameter for user (lambda_theta), item factors ( 84 | lambda_beta), and context factors (lambda_gamma). 85 | c0, c1: float 86 | Confidence for 0 and 1 in Hu et al., c0 must be less than c1 87 | ''' 88 | self.lam_theta = float(kwargs.get('lambda_theta', 1e-5)) 89 | self.lam_beta = float(kwargs.get('lambda_beta', 1e-5)) 90 | self.lam_gamma = float(kwargs.get('lambda_gamma', 1e+0)) 91 | self.c0 = float(kwargs.get('c0', 0.01)) 92 | self.c1 = float(kwargs.get('c1', 1.0)) 93 | assert self.c0 < self.c1, "c0 must be smaller than c1" 94 | 95 | def _init_params(self, n_users, n_items): 96 | ''' Initialize all the latent factors and biases ''' 97 | self.theta = self.init_std * \ 98 | np.random.randn(n_users, self.n_components).astype(self.dtype) 99 | self.beta = self.init_std * \ 100 | np.random.randn(n_items, self.n_components).astype(self.dtype) 101 | self.gamma = self.init_std * \ 102 | np.random.randn(n_items, self.n_components).astype(self.dtype) 103 | # bias for beta and gamma 104 | self.bias_b = np.zeros(n_items, dtype=self.dtype) 105 | self.bias_g = np.zeros(n_items, dtype=self.dtype) 106 | # global bias 107 | self.alpha = 0.0 108 | 109 | def fit(self, X, M, F=None, vad_data=None, **kwargs): 110 | '''Fit the model to the data in X. 111 | 112 | Parameters 113 | ---------- 114 | X : scipy.sparse.csr_matrix, shape (n_users, n_items) 115 | Training click matrix. 116 | 117 | M : scipy.sparse.csr_matrix, shape (n_items, n_items) 118 | Training co-occurrence matrix. 119 | 120 | F : scipy.sparse.csr_matrix, shape (n_items, n_items) 121 | The weight for the co-occurrence matrix. If not provided, 122 | weight by default is 1. 123 | 124 | vad_data: scipy.sparse.csr_matrix, shape (n_users, n_items) 125 | Validation click data. 126 | 127 | **kwargs: dict 128 | Additional keywords to evaluation function call on validation data 129 | 130 | Returns 131 | ------- 132 | self: object 133 | Returns the instance itself. 134 | ''' 135 | n_users, n_items = X.shape 136 | assert M.shape == (n_items, n_items) 137 | 138 | self._init_params(n_users, n_items) 139 | self._update(X, M, F, vad_data, **kwargs) 140 | return self 141 | 142 | def transform(self, X): 143 | pass 144 | 145 | def _update(self, X, M, F, vad_data, **kwargs): 146 | '''Model training and evaluation on validation set''' 147 | XT = X.T.tocsr() # pre-compute this 148 | self.vad_ndcg = -np.inf 149 | for i in xrange(self.max_iter): 150 | if self.verbose: 151 | print('ITERATION #%d' % i) 152 | self._update_factors(X, XT, M, F) 153 | self._update_biases(M, F) 154 | if vad_data is not None: 155 | vad_ndcg = self._validate(X, vad_data, **kwargs) 156 | if self.early_stopping and self.vad_ndcg > vad_ndcg: 157 | break # we will not save the parameter for this iteration 158 | self.vad_ndcg = vad_ndcg 159 | if self.save_params: 160 | self._save_params(i) 161 | pass 162 | 163 | def _update_factors(self, X, XT, M, F): 164 | if self.verbose: 165 | start_t = _writeline_and_time('\tUpdating user factors...') 166 | self.theta = update_theta(self.beta, X, self.c0, 167 | self.c1, self.lam_theta, 168 | self.n_jobs, 169 | batch_size=self.batch_size) 170 | if self.verbose: 171 | print('\r\tUpdating user factors: time=%.2f' 172 | % (time.time() - start_t)) 173 | start_t = _writeline_and_time('\tUpdating item factors...') 174 | self.beta = update_beta(self.theta, self.gamma, 175 | self.bias_b, self.bias_g, self.alpha, 176 | XT, M, F, self.c0, self.c1, self.lam_beta, 177 | self.n_jobs, 178 | batch_size=self.batch_size) 179 | if self.verbose: 180 | print('\r\tUpdating item factors: time=%.2f' 181 | % (time.time() - start_t)) 182 | start_t = _writeline_and_time('\tUpdating context factors...') 183 | # here it really should be M^T and F^T, but both are symmetric 184 | self.gamma = update_gamma(self.beta, self.bias_b, self.bias_g, 185 | self.alpha, M, F, self.lam_gamma, 186 | self.n_jobs, 187 | batch_size=self.batch_size) 188 | if self.verbose: 189 | print('\r\tUpdating context factors: time=%.2f' 190 | % (time.time() - start_t)) 191 | pass 192 | 193 | def _update_biases(self, M, F): 194 | if self.verbose: 195 | start_t = _writeline_and_time('\tUpdating bias terms...') 196 | self.bias_b = update_bias(self.beta, self.gamma, 197 | self.bias_g, self.alpha, M, F, 198 | self.n_jobs, batch_size=self.batch_size) 199 | # here it really should be M^T and F^T, but both are symmetric 200 | self.bias_g = update_bias(self.gamma, self.beta, 201 | self.bias_b, self.alpha, M, F, 202 | self.n_jobs, batch_size=self.batch_size) 203 | self.alpha = update_alpha(self.beta, self.gamma, 204 | self.bias_b, self.bias_g, M, F, 205 | self.n_jobs, batch_size=self.batch_size) 206 | if self.verbose: 207 | print('\r\tUpdating bias terms: time=%.2f' 208 | % (time.time() - start_t)) 209 | pass 210 | 211 | def _validate(self, X, vad_data, **kwargs): 212 | vad_ndcg = rec_eval.normalized_dcg_at_k(X, vad_data, 213 | self.theta, 214 | self.beta, 215 | **kwargs) 216 | if self.verbose: 217 | print('\tValidation NDCG@k: %.5f' % vad_ndcg) 218 | return vad_ndcg 219 | 220 | def _save_params(self, iter): 221 | '''Save the parameters''' 222 | if not os.path.exists(self.save_dir): 223 | os.makedirs(self.save_dir) 224 | filename = 'CoFacto_K%d_iter%d.npz' % (self.n_components, iter) 225 | np.savez(os.path.join(self.save_dir, filename), U=self.theta, 226 | V=self.beta) 227 | 228 | 229 | # Utility functions # 230 | def _writeline_and_time(s): 231 | sys.stdout.write(s) 232 | sys.stdout.flush() 233 | return time.time() 234 | 235 | 236 | def get_row(Y, i): 237 | '''Given a scipy.sparse.csr_matrix Y, get the values and indices of the 238 | non-zero values in i_th row''' 239 | lo, hi = Y.indptr[i], Y.indptr[i + 1] 240 | return Y.data[lo:hi], Y.indices[lo:hi] 241 | 242 | 243 | def update_theta(beta, X, c0, c1, lam_theta, n_jobs, batch_size=1000): 244 | '''Update user latent factors''' 245 | m, n = X.shape # m: number of users, n: number of items 246 | f = beta.shape[1] # f: number of factors 247 | 248 | BTB = c0 * np.dot(beta.T, beta) # precompute this 249 | BTBpR = BTB + lam_theta * np.eye(f, dtype=beta.dtype) 250 | 251 | start_idx = range(0, m, batch_size) 252 | end_idx = start_idx[1:] + [m] 253 | res = Parallel(n_jobs=n_jobs)( 254 | delayed(_solve_weighted_factor)( 255 | lo, hi, beta, X, BTBpR, c0, c1, f, lam_theta) 256 | for lo, hi in zip(start_idx, end_idx)) 257 | theta = np.vstack(res) 258 | return theta 259 | 260 | 261 | def _solve_weighted_factor(lo, hi, beta, X, BTBpR, c0, c1, f, lam_theta): 262 | theta_batch = np.empty((hi - lo, f), dtype=beta.dtype) 263 | for ib, u in enumerate(xrange(lo, hi)): 264 | x_u, idx_u = get_row(X, u) 265 | B_u = beta[idx_u] 266 | a = x_u.dot(c1 * B_u) 267 | B = BTBpR + B_u.T.dot((c1 - c0) * B_u) 268 | theta_batch[ib] = LA.solve(B, a) 269 | return theta_batch 270 | 271 | 272 | def update_beta(theta, gamma, bias_b, bias_g, alpha, XT, M, F, c0, c1, 273 | lam_beta, n_jobs, batch_size=1000): 274 | '''Update item latent factors/embeddings''' 275 | n, m = XT.shape # m: number of users, n: number of items 276 | f = theta.shape[1] 277 | assert theta.shape[0] == m 278 | assert gamma.shape == (n, f) 279 | 280 | TTT = c0 * np.dot(theta.T, theta) # precompute this 281 | TTTpR = TTT + lam_beta * np.eye(f, dtype=theta.dtype) 282 | 283 | start_idx = range(0, n, batch_size) 284 | end_idx = start_idx[1:] + [n] 285 | res = Parallel(n_jobs=n_jobs)( 286 | delayed(_solve_weighted_cofactor)( 287 | lo, hi, theta, gamma, bias_b, bias_g, alpha, XT, M, F, TTTpR, c0, 288 | c1, f, lam_beta) 289 | for lo, hi in zip(start_idx, end_idx)) 290 | beta = np.vstack(res) 291 | return beta 292 | 293 | 294 | def _solve_weighted_cofactor(lo, hi, theta, gamma, bias_b, bias_g, alpha, XT, 295 | M, F, TTTpR, c0, c1, f, lam_beta): 296 | beta_batch = np.empty((hi - lo, f), dtype=theta.dtype) 297 | for ib, i in enumerate(xrange(lo, hi)): 298 | x_i, idx_x_i = get_row(XT, i) 299 | T_i = theta[idx_x_i] 300 | 301 | m_i, idx_m_i = get_row(M, i) 302 | G_i = gamma[idx_m_i] 303 | 304 | rsd = m_i - bias_b[i] - bias_g[idx_m_i] - alpha 305 | 306 | if F is not None: 307 | f_i, _ = get_row(F, i) 308 | GTG = G_i.T.dot(G_i * f_i[:, np.newaxis]) 309 | rsd *= f_i 310 | else: 311 | GTG = G_i.T.dot(G_i) 312 | 313 | B = TTTpR + T_i.T.dot((c1 - c0) * T_i) + GTG 314 | a = x_i.dot(c1 * T_i) + np.dot(rsd, G_i) 315 | beta_batch[ib] = LA.solve(B, a) 316 | return beta_batch 317 | 318 | 319 | def update_gamma(beta, bias_b, bias_g, alpha, MT, FT, lam_gamma, 320 | n_jobs, batch_size=1000): 321 | '''Update context latent factors''' 322 | n, f = beta.shape # n: number of items, f: number of factors 323 | 324 | start_idx = range(0, n, batch_size) 325 | end_idx = start_idx[1:] + [n] 326 | res = Parallel(n_jobs=n_jobs)( 327 | delayed(_solve_factor)( 328 | lo, hi, beta, bias_b, bias_g, alpha, MT, FT, f, lam_gamma) 329 | for lo, hi in zip(start_idx, end_idx)) 330 | gamma = np.vstack(res) 331 | return gamma 332 | 333 | 334 | def _solve_factor(lo, hi, beta, bias_b, bias_g, alpha, MT, FT, f, lam_gamma, 335 | BTBpR=None): 336 | gamma_batch = np.empty((hi - lo, f), dtype=beta.dtype) 337 | for ib, j in enumerate(xrange(lo, hi)): 338 | m_j, idx_j = get_row(MT, j) 339 | rsd = m_j - bias_b[idx_j] - bias_g[j] - alpha 340 | B_j = beta[idx_j] 341 | if FT is not None: 342 | f_j, _ = get_row(FT, j) 343 | BTB = B_j.T.dot(B_j * f_j[:, np.newaxis]) 344 | rsd *= f_j 345 | else: 346 | BTB = B_j.T.dot(B_j) 347 | 348 | B = BTB + lam_gamma * np.eye(f, dtype=beta.dtype) 349 | a = np.dot(rsd, B_j) 350 | gamma_batch[ib] = LA.solve(B, a) 351 | return gamma_batch 352 | 353 | 354 | def update_bias(beta, gamma, bias_g, alpha, M, F, n_jobs, batch_size=1000): 355 | ''' Update the per-item (or context) bias term. 356 | ''' 357 | n = beta.shape[0] 358 | 359 | start_idx = range(0, n, batch_size) 360 | end_idx = start_idx[1:] + [n] 361 | 362 | res = Parallel(n_jobs=n_jobs)( 363 | delayed(_solve_bias)(lo, hi, beta, gamma, bias_g, alpha, M, F) 364 | for lo, hi in zip(start_idx, end_idx)) 365 | bias_b = np.hstack(res) 366 | return bias_b 367 | 368 | 369 | def _solve_bias(lo, hi, beta, gamma, bias_g, alpha, M, F): 370 | bias_b_batch = np.empty(hi - lo, dtype=beta.dtype) 371 | for ib, i in enumerate(xrange(lo, hi)): 372 | m_i, idx_i = get_row(M, i) 373 | m_i_hat = gamma[idx_i].dot(beta[i]) + bias_g[idx_i] + alpha 374 | rsd = m_i - m_i_hat 375 | 376 | if F is not None: 377 | f_i, _ = get_row(F, i) 378 | rsd *= f_i 379 | 380 | if rsd.size > 0: 381 | bias_b_batch[ib] = rsd.mean() 382 | else: 383 | bias_b_batch[ib] = 0. 384 | return bias_b_batch 385 | 386 | 387 | def update_alpha(beta, gamma, bias_b, bias_g, M, F, n_jobs, batch_size=1000): 388 | ''' Update the global bias term 389 | ''' 390 | n = beta.shape[0] 391 | assert beta.shape == gamma.shape 392 | assert bias_b.shape == bias_g.shape 393 | 394 | start_idx = range(0, n, batch_size) 395 | end_idx = start_idx[1:] + [n] 396 | 397 | res = Parallel(n_jobs=n_jobs)( 398 | delayed(_solve_alpha)(lo, hi, beta, gamma, bias_b, bias_g, M, F) 399 | for lo, hi in zip(start_idx, end_idx)) 400 | 401 | return np.sum(res) / M.data.size 402 | 403 | 404 | def _solve_alpha(lo, hi, beta, gamma, bias_b, bias_g, M, F): 405 | res = 0. 406 | for ib, i in enumerate(xrange(lo, hi)): 407 | m_i, idx_i = get_row(M, i) 408 | m_i_hat = gamma[idx_i].dot(beta[i]) + bias_b[i] + bias_g[idx_i] 409 | rsd = m_i - m_i_hat 410 | 411 | if F is not None: 412 | f_i, _ = get_row(F, i) 413 | rsd *= f_i 414 | res += rsd.sum() 415 | return res 416 | -------------------------------------------------------------------------------- /src/preprocess_ML20M.ipynb: -------------------------------------------------------------------------------- 1 | { 2 | "cells": [ 3 | { 4 | "cell_type": "markdown", 5 | "metadata": {}, 6 | "source": [ 7 | "# Preprocess MovieLens-20M " 8 | ] 9 | }, 10 | { 11 | "cell_type": "code", 12 | "execution_count": 1, 13 | "metadata": { 14 | "collapsed": false 15 | }, 16 | "outputs": [], 17 | "source": [ 18 | "import datetime\n", 19 | "import json\n", 20 | "import os\n", 21 | "import time\n", 22 | "\n", 23 | "import numpy as np\n", 24 | "\n", 25 | "import matplotlib\n", 26 | "matplotlib.use('Agg')\n", 27 | "import matplotlib.pyplot as plt\n", 28 | "%matplotlib inline\n", 29 | "\n", 30 | "import pandas as pd\n", 31 | "import scipy.sparse\n", 32 | "\n", 33 | "import seaborn as sns\n", 34 | "sns.set(context=\"paper\", font_scale=1.5, rc={\"lines.linewidth\": 2}, font='DejaVu Serif')" 35 | ] 36 | }, 37 | { 38 | "cell_type": "code", 39 | "execution_count": 2, 40 | "metadata": { 41 | "collapsed": true 42 | }, 43 | "outputs": [], 44 | "source": [ 45 | "DATA_DIR = '/hdd2/dawen/data/ml-20m/'" 46 | ] 47 | }, 48 | { 49 | "cell_type": "code", 50 | "execution_count": 3, 51 | "metadata": { 52 | "collapsed": true 53 | }, 54 | "outputs": [], 55 | "source": [ 56 | "def timestamp_to_date(timestamp):\n", 57 | " return datetime.datetime.fromtimestamp(timestamp).strftime('%Y-%m-%d %H:%M:%S')" 58 | ] 59 | }, 60 | { 61 | "cell_type": "code", 62 | "execution_count": 4, 63 | "metadata": { 64 | "collapsed": true 65 | }, 66 | "outputs": [], 67 | "source": [ 68 | "raw_data = pd.read_csv(os.path.join(DATA_DIR, 'ratings.csv'), header=0)" 69 | ] 70 | }, 71 | { 72 | "cell_type": "code", 73 | "execution_count": 5, 74 | "metadata": { 75 | "collapsed": false 76 | }, 77 | "outputs": [], 78 | "source": [ 79 | "# binarize the data (only keep ratings >= 4)\n", 80 | "raw_data = raw_data[raw_data['rating'] > 3.5]" 81 | ] 82 | }, 83 | { 84 | "cell_type": "code", 85 | "execution_count": 6, 86 | "metadata": { 87 | "collapsed": true 88 | }, 89 | "outputs": [], 90 | "source": [ 91 | "# sort the raw data accorindg to timestamp\n", 92 | "raw_data = raw_data.sort_index(by=['timestamp'])" 93 | ] 94 | }, 95 | { 96 | "cell_type": "code", 97 | "execution_count": 7, 98 | "metadata": { 99 | "collapsed": false 100 | }, 101 | "outputs": [ 102 | { 103 | "data": { 104 | "text/html": [ 105 | "
\n", 106 | "\n", 107 | " \n", 108 | " \n", 109 | " \n", 110 | " \n", 111 | " \n", 112 | " \n", 113 | " \n", 114 | " \n", 115 | " \n", 116 | " \n", 117 | " \n", 118 | " \n", 119 | " \n", 120 | " \n", 121 | " \n", 122 | " \n", 123 | " \n", 124 | " \n", 125 | " \n", 126 | " \n", 127 | " \n", 128 | " \n", 129 | " \n", 130 | " \n", 131 | " \n", 132 | " \n", 133 | " \n", 134 | " \n", 135 | " \n", 136 | " \n", 137 | " \n", 138 | " \n", 139 | " \n", 140 | " \n", 141 | " \n", 142 | " \n", 143 | " \n", 144 | " \n", 145 | " \n", 146 | " \n", 147 | " \n", 148 | " \n", 149 | " \n", 150 | " \n", 151 | " \n", 152 | " \n", 153 | " \n", 154 | " \n", 155 | " \n", 156 | " \n", 157 | " \n", 158 | " \n", 159 | " \n", 160 | " \n", 161 | " \n", 162 | " \n", 163 | " \n", 164 | " \n", 165 | " \n", 166 | " \n", 167 | " \n", 168 | " \n", 169 | " \n", 170 | " \n", 171 | " \n", 172 | " \n", 173 | " \n", 174 | " \n", 175 | " \n", 176 | " \n", 177 | " \n", 178 | " \n", 179 | " \n", 180 | " \n", 181 | " \n", 182 | " \n", 183 | " \n", 184 | " \n", 185 | " \n", 186 | " \n", 187 | " \n", 188 | " \n", 189 | " \n", 190 | " \n", 191 | " \n", 192 | " \n", 193 | " \n", 194 | " \n", 195 | " \n", 196 | " \n", 197 | " \n", 198 | " \n", 199 | " \n", 200 | " \n", 201 | " \n", 202 | " \n", 203 | " \n", 204 | " \n", 205 | " \n", 206 | " \n", 207 | " \n", 208 | " \n", 209 | " \n", 210 | " \n", 211 | " \n", 212 | " \n", 213 | " \n", 214 | " \n", 215 | " \n", 216 | " \n", 217 | " \n", 218 | " \n", 219 | " \n", 220 | " \n", 221 | " \n", 222 | " \n", 223 | " \n", 224 | " \n", 225 | " \n", 226 | " \n", 227 | " \n", 228 | " \n", 229 | " \n", 230 | " \n", 231 | " \n", 232 | " \n", 233 | " \n", 234 | " \n", 235 | " \n", 236 | " \n", 237 | " \n", 238 | " \n", 239 | " \n", 240 | " \n", 241 | " \n", 242 | " \n", 243 | " \n", 244 | " \n", 245 | " \n", 246 | " \n", 247 | " \n", 248 | " \n", 249 | " \n", 250 | " \n", 251 | " \n", 252 | " \n", 253 | " \n", 254 | " \n", 255 | " \n", 256 | " \n", 257 | " \n", 258 | " \n", 259 | " \n", 260 | " \n", 261 | " \n", 262 | " \n", 263 | " \n", 264 | " \n", 265 | " \n", 266 | " \n", 267 | " \n", 268 | " \n", 269 | " \n", 270 | " \n", 271 | " \n", 272 | " \n", 273 | " \n", 274 | " \n", 275 | " \n", 276 | " \n", 277 | " \n", 278 | " \n", 279 | " \n", 280 | " \n", 281 | " \n", 282 | " \n", 283 | " \n", 284 | " \n", 285 | " \n", 286 | " \n", 287 | " \n", 288 | " \n", 289 | " \n", 290 | " \n", 291 | " \n", 292 | " \n", 293 | " \n", 294 | " \n", 295 | " \n", 296 | " \n", 297 | " \n", 298 | " \n", 299 | " \n", 300 | " \n", 301 | " \n", 302 | " \n", 303 | " \n", 304 | " \n", 305 | " \n", 306 | " \n", 307 | " \n", 308 | " \n", 309 | " \n", 310 | " \n", 311 | " \n", 312 | " \n", 313 | " \n", 314 | " \n", 315 | " \n", 316 | " \n", 317 | " \n", 318 | " \n", 319 | " \n", 320 | " \n", 321 | " \n", 322 | " \n", 323 | " \n", 324 | " \n", 325 | " \n", 326 | " \n", 327 | " \n", 328 | " \n", 329 | " \n", 330 | " \n", 331 | " \n", 332 | " \n", 333 | " \n", 334 | " \n", 335 | " \n", 336 | " \n", 337 | " \n", 338 | " \n", 339 | " \n", 340 | " \n", 341 | " \n", 342 | " \n", 343 | " \n", 344 | " \n", 345 | " \n", 346 | " \n", 347 | " \n", 348 | " \n", 349 | " \n", 350 | " \n", 351 | " \n", 352 | " \n", 353 | " \n", 354 | " \n", 355 | " \n", 356 | " \n", 357 | " \n", 358 | " \n", 359 | " \n", 360 | " \n", 361 | " \n", 362 | " \n", 363 | " \n", 364 | " \n", 365 | " \n", 366 | " \n", 367 | " \n", 368 | " \n", 369 | " \n", 370 | " \n", 371 | " \n", 372 | " \n", 373 | " \n", 374 | " \n", 375 | " \n", 376 | " \n", 377 | " \n", 378 | " \n", 379 | " \n", 380 | " \n", 381 | " \n", 382 | " \n", 383 | " \n", 384 | " \n", 385 | " \n", 386 | " \n", 387 | " \n", 388 | " \n", 389 | " \n", 390 | " \n", 391 | " \n", 392 | " \n", 393 | " \n", 394 | " \n", 395 | " \n", 396 | " \n", 397 | " \n", 398 | " \n", 399 | " \n", 400 | " \n", 401 | " \n", 402 | " \n", 403 | " \n", 404 | " \n", 405 | " \n", 406 | " \n", 407 | " \n", 408 | " \n", 409 | " \n", 410 | " \n", 411 | " \n", 412 | " \n", 413 | " \n", 414 | " \n", 415 | " \n", 416 | " \n", 417 | " \n", 418 | " \n", 419 | " \n", 420 | " \n", 421 | " \n", 422 | " \n", 423 | " \n", 424 | " \n", 425 | " \n", 426 | " \n", 427 | " \n", 428 | " \n", 429 | " \n", 430 | " \n", 431 | " \n", 432 | " \n", 433 | " \n", 434 | " \n", 435 | " \n", 436 | " \n", 437 | " \n", 438 | " \n", 439 | " \n", 440 | " \n", 441 | " \n", 442 | " \n", 443 | " \n", 444 | " \n", 445 | " \n", 446 | " \n", 447 | " \n", 448 | " \n", 449 | " \n", 450 | " \n", 451 | " \n", 452 | " \n", 453 | " \n", 454 | " \n", 455 | " \n", 456 | " \n", 457 | " \n", 458 | " \n", 459 | " \n", 460 | " \n", 461 | " \n", 462 | " \n", 463 | " \n", 464 | " \n", 465 | " \n", 466 | " \n", 467 | " \n", 468 | " \n", 469 | " \n", 470 | " \n", 471 | " \n", 472 | " \n", 473 | " \n", 474 | " \n", 475 | " \n", 476 | " \n", 477 | " \n", 478 | " \n", 479 | " \n", 480 | " \n", 481 | " \n", 482 | " \n", 483 | " \n", 484 | " \n", 485 | " \n", 486 | " \n", 487 | " \n", 488 | " \n", 489 | " \n", 490 | " \n", 491 | " \n", 492 | " \n", 493 | " \n", 494 | " \n", 495 | " \n", 496 | " \n", 497 | " \n", 498 | " \n", 499 | " \n", 500 | " \n", 501 | " \n", 502 | " \n", 503 | " \n", 504 | " \n", 505 | " \n", 506 | " \n", 507 | " \n", 508 | " \n", 509 | " \n", 510 | " \n", 511 | " \n", 512 | " \n", 513 | " \n", 514 | " \n", 515 | " \n", 516 | " \n", 517 | " \n", 518 | " \n", 519 | " \n", 520 | " \n", 521 | " \n", 522 | " \n", 523 | " \n", 524 | " \n", 525 | " \n", 526 | " \n", 527 | " \n", 528 | " \n", 529 | " \n", 530 | " \n", 531 | " \n", 532 | " \n", 533 | " \n", 534 | " \n", 535 | " \n", 536 | " \n", 537 | " \n", 538 | " \n", 539 | " \n", 540 | " \n", 541 | " \n", 542 | " \n", 543 | " \n", 544 | " \n", 545 | "
userIdmovieIdratingtimestamp
41824212850711764.0789652004
18950936131160475.0789652009
15688196108467574.0822873600
1234118685252704.0822873600
144525019985114.0822873600
1445251799851585.0822873600
1445251699851554.0822873600
1445251599851524.0822873600
1445251499851505.0822873600
1445251399851475.0822873600
1445251299851454.0822873600
1445250999851395.0822873600
1445250799851325.0822873600
1445250699851315.0822873600
1445250599851215.0822873600
305663920821325.0822873600
15688194108467114.0822873600
1445250499851194.0822873600
1445250399851184.0822873600
1234118485252604.0822873600
1445250299851104.0822873600
1234118185252505.0822873600
19424622134445114.0822873600
19424624134445215.0822873600
19424626134445455.0822873600
19424627134445585.0822873600
123411598525224.0822873600
123411618525275.0822873600
1234116285252105.0822873600
1234116585252175.0822873600
...............
19742824136690483945.01427775557
1974280513669011365.01427775558
197428311366901048414.51427775561
154808021070737455.01427776814
1548082210707359715.01427776816
1548080510707311484.01427776833
15480828107073922594.01427776892
1548082310707360165.01427777118
154808041070738585.01427777123
15480826107073585594.01427777129
1548081910707349935.01427777155
1548082110707359525.01427777157
1548081810707343065.01427777158
1548082510707371535.01427777166
1548081610707337934.01427777169
154807981070735275.01427777203
178777481236131092434.01427779965
83784515781473614.01427780465
83784525781474385.01427780468
8378505578141089795.01427780517
83784095781415274.01427780519
83784075781412745.01427780571
83784135781417484.51427780617
83784435781462835.01427780623
8378393578149244.51427780631
83784235781435274.01427780657
837846857814487744.01427780663
83784055781412405.01427781001
83784155781423114.51427781083
1289852789081524584.01427782288
\n", 546 | "

9995410 rows × 4 columns

\n", 547 | "
" 548 | ], 549 | "text/plain": [ 550 | " userId movieId rating timestamp\n", 551 | "4182421 28507 1176 4.0 789652004\n", 552 | "18950936 131160 47 5.0 789652009\n", 553 | "15688196 108467 57 4.0 822873600\n", 554 | "12341186 85252 70 4.0 822873600\n", 555 | "14452501 99851 1 4.0 822873600\n", 556 | "14452517 99851 58 5.0 822873600\n", 557 | "14452516 99851 55 4.0 822873600\n", 558 | "14452515 99851 52 4.0 822873600\n", 559 | "14452514 99851 50 5.0 822873600\n", 560 | "14452513 99851 47 5.0 822873600\n", 561 | "14452512 99851 45 4.0 822873600\n", 562 | "14452509 99851 39 5.0 822873600\n", 563 | "14452507 99851 32 5.0 822873600\n", 564 | "14452506 99851 31 5.0 822873600\n", 565 | "14452505 99851 21 5.0 822873600\n", 566 | "3056639 20821 32 5.0 822873600\n", 567 | "15688194 108467 11 4.0 822873600\n", 568 | "14452504 99851 19 4.0 822873600\n", 569 | "14452503 99851 18 4.0 822873600\n", 570 | "12341184 85252 60 4.0 822873600\n", 571 | "14452502 99851 10 4.0 822873600\n", 572 | "12341181 85252 50 5.0 822873600\n", 573 | "19424622 134445 11 4.0 822873600\n", 574 | "19424624 134445 21 5.0 822873600\n", 575 | "19424626 134445 45 5.0 822873600\n", 576 | "19424627 134445 58 5.0 822873600\n", 577 | "12341159 85252 2 4.0 822873600\n", 578 | "12341161 85252 7 5.0 822873600\n", 579 | "12341162 85252 10 5.0 822873600\n", 580 | "12341165 85252 17 5.0 822873600\n", 581 | "... ... ... ... ...\n", 582 | "19742824 136690 48394 5.0 1427775557\n", 583 | "19742805 136690 1136 5.0 1427775558\n", 584 | "19742831 136690 104841 4.5 1427775561\n", 585 | "15480802 107073 745 5.0 1427776814\n", 586 | "15480822 107073 5971 5.0 1427776816\n", 587 | "15480805 107073 1148 4.0 1427776833\n", 588 | "15480828 107073 92259 4.0 1427776892\n", 589 | "15480823 107073 6016 5.0 1427777118\n", 590 | "15480804 107073 858 5.0 1427777123\n", 591 | "15480826 107073 58559 4.0 1427777129\n", 592 | "15480819 107073 4993 5.0 1427777155\n", 593 | "15480821 107073 5952 5.0 1427777157\n", 594 | "15480818 107073 4306 5.0 1427777158\n", 595 | "15480825 107073 7153 5.0 1427777166\n", 596 | "15480816 107073 3793 4.0 1427777169\n", 597 | "15480798 107073 527 5.0 1427777203\n", 598 | "17877748 123613 109243 4.0 1427779965\n", 599 | "8378451 57814 7361 4.0 1427780465\n", 600 | "8378452 57814 7438 5.0 1427780468\n", 601 | "8378505 57814 108979 5.0 1427780517\n", 602 | "8378409 57814 1527 4.0 1427780519\n", 603 | "8378407 57814 1274 5.0 1427780571\n", 604 | "8378413 57814 1748 4.5 1427780617\n", 605 | "8378443 57814 6283 5.0 1427780623\n", 606 | "8378393 57814 924 4.5 1427780631\n", 607 | "8378423 57814 3527 4.0 1427780657\n", 608 | "8378468 57814 48774 4.0 1427780663\n", 609 | "8378405 57814 1240 5.0 1427781001\n", 610 | "8378415 57814 2311 4.5 1427781083\n", 611 | "12898527 89081 52458 4.0 1427782288\n", 612 | "\n", 613 | "[9995410 rows x 4 columns]" 614 | ] 615 | }, 616 | "execution_count": 7, 617 | "metadata": {}, 618 | "output_type": "execute_result" 619 | } 620 | ], 621 | "source": [ 622 | "raw_data" 623 | ] 624 | }, 625 | { 626 | "cell_type": "code", 627 | "execution_count": 8, 628 | "metadata": { 629 | "collapsed": false 630 | }, 631 | "outputs": [], 632 | "source": [ 633 | "tstamp = np.array(raw_data['timestamp'])" 634 | ] 635 | }, 636 | { 637 | "cell_type": "code", 638 | "execution_count": 9, 639 | "metadata": { 640 | "collapsed": false 641 | }, 642 | "outputs": [ 643 | { 644 | "name": "stdout", 645 | "output_type": "stream", 646 | "text": [ 647 | "Time span of the dataset: From 1995-01-09 06:46:44 to 2015-03-31 02:11:28\n" 648 | ] 649 | } 650 | ], 651 | "source": [ 652 | "print(\"Time span of the dataset: From %s to %s\" % \n", 653 | " (timestamp_to_date(np.min(tstamp)), timestamp_to_date(np.max(tstamp))))" 654 | ] 655 | }, 656 | { 657 | "cell_type": "code", 658 | "execution_count": 10, 659 | "metadata": { 660 | "collapsed": true 661 | }, 662 | "outputs": [], 663 | "source": [ 664 | "# apparently the timestamps are ordered, check to make sure\n", 665 | "\n", 666 | "for i in xrange(tstamp.size - 1):\n", 667 | " if tstamp[i] > tstamp[i + 1]:\n", 668 | " print(\"not ordered\")" 669 | ] 670 | }, 671 | { 672 | "cell_type": "markdown", 673 | "metadata": {}, 674 | "source": [ 675 | "Confirmed the timestamps are ordered" 676 | ] 677 | }, 678 | { 679 | "cell_type": "code", 680 | "execution_count": 11, 681 | "metadata": { 682 | "collapsed": false 683 | }, 684 | "outputs": [ 685 | { 686 | "data": { 687 | "image/png": "iVBORw0KGgoAAAANSUhEUgAAAaYAAAFBCAYAAAA8MAs5AAAABHNCSVQICAgIfAhkiAAAAAlwSFlz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\nKUxCCCF0RQqTEEIIXfn/AI1snaA4I+GAAAAAAElFTkSuQmCC\n", 688 | "text/plain": [ 689 | "" 690 | ] 691 | }, 692 | "metadata": {}, 693 | "output_type": "display_data" 694 | } 695 | ], 696 | "source": [ 697 | "plt.hist(tstamp, bins=50)\n", 698 | "xticks = np.linspace(tstamp[0], tstamp[-1], 10)\n", 699 | "plt.xticks(xticks, map(lambda x: timestamp_to_date(x)[:7], xticks), rotation=90)\n", 700 | "pass" 701 | ] 702 | }, 703 | { 704 | "cell_type": "markdown", 705 | "metadata": {}, 706 | "source": [ 707 | "Now we select the data from 1995-01-01 to the last day as the dataset (i.e., all the dataset)" 708 | ] 709 | }, 710 | { 711 | "cell_type": "code", 712 | "execution_count": 12, 713 | "metadata": { 714 | "collapsed": false 715 | }, 716 | "outputs": [], 717 | "source": [ 718 | "start_t = time.mktime(datetime.datetime.strptime(\"1995-01-01\", \"%Y-%m-%d\").timetuple())" 719 | ] 720 | }, 721 | { 722 | "cell_type": "code", 723 | "execution_count": 42, 724 | "metadata": { 725 | "collapsed": false 726 | }, 727 | "outputs": [], 728 | "source": [ 729 | "raw_data = raw_data[raw_data['timestamp'] >= start_t]" 730 | ] 731 | }, 732 | { 733 | "cell_type": "markdown", 734 | "metadata": {}, 735 | "source": [ 736 | "Take the first 80% of the data as train and validation set" 737 | ] 738 | }, 739 | { 740 | "cell_type": "code", 741 | "execution_count": 96, 742 | "metadata": { 743 | "collapsed": true 744 | }, 745 | "outputs": [], 746 | "source": [ 747 | "tr_vd_raw_data = raw_data[:int(0.8 * raw_data.shape[0])]" 748 | ] 749 | }, 750 | { 751 | "cell_type": "code", 752 | "execution_count": 97, 753 | "metadata": { 754 | "collapsed": true 755 | }, 756 | "outputs": [], 757 | "source": [ 758 | "def get_count(tp, id):\n", 759 | " playcount_groupbyid = tp[[id]].groupby(id, as_index=False)\n", 760 | " count = playcount_groupbyid.size()\n", 761 | " return count" 762 | ] 763 | }, 764 | { 765 | "cell_type": "code", 766 | "execution_count": 98, 767 | "metadata": { 768 | "collapsed": true 769 | }, 770 | "outputs": [], 771 | "source": [ 772 | "def filter_triplets(tp, min_uc=5, min_sc=0):\n", 773 | " # Only keep the triplets for songs which were listened to by at least min_sc users. \n", 774 | " if min_sc > 0:\n", 775 | " songcount = get_count(tp, 'movieId')\n", 776 | " tp = tp[tp['movieId'].isin(songcount.index[songcount >= min_sc])]\n", 777 | " \n", 778 | " # Only keep the triplets for users who listened to at least min_uc songs\n", 779 | " # After doing this, some of the songs will have less than min_uc users, but should only be a small proportion\n", 780 | " if min_uc > 0:\n", 781 | " usercount = get_count(tp, 'userId')\n", 782 | " tp = tp[tp['userId'].isin(usercount.index[usercount >= min_uc])]\n", 783 | " \n", 784 | " # Update both usercount and songcount after filtering\n", 785 | " usercount, songcount = get_count(tp, 'userId'), get_count(tp, 'movieId') \n", 786 | " return tp, usercount, songcount" 787 | ] 788 | }, 789 | { 790 | "cell_type": "code", 791 | "execution_count": 99, 792 | "metadata": { 793 | "collapsed": true 794 | }, 795 | "outputs": [], 796 | "source": [ 797 | "tr_vd_raw_data, user_activity, item_popularity = filter_triplets(tr_vd_raw_data)" 798 | ] 799 | }, 800 | { 801 | "cell_type": "code", 802 | "execution_count": 100, 803 | "metadata": { 804 | "collapsed": false 805 | }, 806 | "outputs": [ 807 | { 808 | "name": "stdout", 809 | "output_type": "stream", 810 | "text": [ 811 | "After filtering, there are 7992863 watching events from 111148 users and 11711 movies (sparsity: 0.614%)\n" 812 | ] 813 | } 814 | ], 815 | "source": [ 816 | "sparsity = 1. * tr_vd_raw_data.shape[0] / (user_activity.shape[0] * item_popularity.shape[0])\n", 817 | "\n", 818 | "print(\"After filtering, there are %d watching events from %d users and %d movies (sparsity: %.3f%%)\" % \n", 819 | " (tr_vd_raw_data.shape[0], user_activity.shape[0], item_popularity.shape[0], sparsity * 100))" 820 | ] 821 | }, 822 | { 823 | "cell_type": "code", 824 | "execution_count": 101, 825 | "metadata": { 826 | "collapsed": false 827 | }, 828 | "outputs": [], 829 | "source": [ 830 | "unique_uid = user_activity.index\n", 831 | "unique_sid = item_popularity.index" 832 | ] 833 | }, 834 | { 835 | "cell_type": "code", 836 | "execution_count": 102, 837 | "metadata": { 838 | "collapsed": true 839 | }, 840 | "outputs": [], 841 | "source": [ 842 | "song2id = dict((sid, i) for (i, sid) in enumerate(unique_sid))\n", 843 | "user2id = dict((uid, i) for (i, uid) in enumerate(unique_uid))" 844 | ] 845 | }, 846 | { 847 | "cell_type": "code", 848 | "execution_count": 103, 849 | "metadata": { 850 | "collapsed": true 851 | }, 852 | "outputs": [], 853 | "source": [ 854 | "with open(os.path.join(DATA_DIR, 'pro', 'unique_uid.txt'), 'w') as f:\n", 855 | " for uid in unique_uid:\n", 856 | " f.write('%s\\n' % uid)" 857 | ] 858 | }, 859 | { 860 | "cell_type": "code", 861 | "execution_count": 104, 862 | "metadata": { 863 | "collapsed": true 864 | }, 865 | "outputs": [], 866 | "source": [ 867 | "with open(os.path.join(DATA_DIR, 'pro', 'unique_sid.txt'), 'w') as f:\n", 868 | " for sid in unique_sid:\n", 869 | " f.write('%s\\n' % sid)" 870 | ] 871 | }, 872 | { 873 | "cell_type": "markdown", 874 | "metadata": {}, 875 | "source": [ 876 | "Split 12.5% (10% of the total ratings) as validation set" 877 | ] 878 | }, 879 | { 880 | "cell_type": "code", 881 | "execution_count": 105, 882 | "metadata": { 883 | "collapsed": true 884 | }, 885 | "outputs": [], 886 | "source": [ 887 | "np.random.seed(13579)\n", 888 | "n_ratings = tr_vd_raw_data.shape[0]\n", 889 | "vad = np.random.choice(n_ratings, size=int(0.125 * n_ratings), replace=False)" 890 | ] 891 | }, 892 | { 893 | "cell_type": "code", 894 | "execution_count": 106, 895 | "metadata": { 896 | "collapsed": true 897 | }, 898 | "outputs": [], 899 | "source": [ 900 | "vad_idx = np.zeros(n_ratings, dtype=bool)\n", 901 | "vad_idx[vad] = True\n", 902 | "\n", 903 | "vad_raw_data = tr_vd_raw_data[vad_idx]\n", 904 | "train_raw_data = tr_vd_raw_data[~vad_idx]" 905 | ] 906 | }, 907 | { 908 | "cell_type": "markdown", 909 | "metadata": {}, 910 | "source": [ 911 | "Make sure there is no empty users/items" 912 | ] 913 | }, 914 | { 915 | "cell_type": "code", 916 | "execution_count": 107, 917 | "metadata": { 918 | "collapsed": false 919 | }, 920 | "outputs": [ 921 | { 922 | "name": "stdout", 923 | "output_type": "stream", 924 | "text": [ 925 | "There are total of 111148 unique users in the training set and 111148 unique users in the entire dataset\n" 926 | ] 927 | } 928 | ], 929 | "source": [ 930 | "print \"There are total of %d unique users in the training set and %d unique users in the entire dataset\" % \\\n", 931 | "(len(pd.unique(train_raw_data['userId'])), len(unique_uid))" 932 | ] 933 | }, 934 | { 935 | "cell_type": "code", 936 | "execution_count": 108, 937 | "metadata": { 938 | "collapsed": false 939 | }, 940 | "outputs": [ 941 | { 942 | "name": "stdout", 943 | "output_type": "stream", 944 | "text": [ 945 | "There are total of 11612 unique items in the training set and 11711 unique items in the entire dataset\n" 946 | ] 947 | } 948 | ], 949 | "source": [ 950 | "print \"There are total of %d unique items in the training set and %d unique items in the entire dataset\" % \\\n", 951 | "(len(pd.unique(train_raw_data['movieId'])), len(unique_sid))" 952 | ] 953 | }, 954 | { 955 | "cell_type": "code", 956 | "execution_count": 109, 957 | "metadata": { 958 | "collapsed": true 959 | }, 960 | "outputs": [], 961 | "source": [ 962 | "train_sid = set(pd.unique(train_raw_data['movieId']))" 963 | ] 964 | }, 965 | { 966 | "cell_type": "code", 967 | "execution_count": 110, 968 | "metadata": { 969 | "collapsed": true 970 | }, 971 | "outputs": [], 972 | "source": [ 973 | "left_sid = list()\n", 974 | "for i, sid in enumerate(unique_sid):\n", 975 | " if sid not in train_sid:\n", 976 | " left_sid.append(sid)" 977 | ] 978 | }, 979 | { 980 | "cell_type": "code", 981 | "execution_count": 111, 982 | "metadata": { 983 | "collapsed": true 984 | }, 985 | "outputs": [], 986 | "source": [ 987 | "move_idx = vad_raw_data['movieId'].isin(left_sid)" 988 | ] 989 | }, 990 | { 991 | "cell_type": "code", 992 | "execution_count": 112, 993 | "metadata": { 994 | "collapsed": false 995 | }, 996 | "outputs": [], 997 | "source": [ 998 | "train_raw_data = train_raw_data.append(vad_raw_data[move_idx])\n", 999 | "vad_raw_data = vad_raw_data[~move_idx]" 1000 | ] 1001 | }, 1002 | { 1003 | "cell_type": "code", 1004 | "execution_count": 113, 1005 | "metadata": { 1006 | "collapsed": false 1007 | }, 1008 | "outputs": [ 1009 | { 1010 | "name": "stdout", 1011 | "output_type": "stream", 1012 | "text": [ 1013 | "There are total of 11711 unique items in the training set and 11711 unique items in the entire dataset\n" 1014 | ] 1015 | } 1016 | ], 1017 | "source": [ 1018 | "print \"There are total of %d unique items in the training set and %d unique items in the entire dataset\" % \\\n", 1019 | "(len(pd.unique(train_raw_data['movieId'])), len(unique_sid))" 1020 | ] 1021 | }, 1022 | { 1023 | "cell_type": "markdown", 1024 | "metadata": {}, 1025 | "source": [ 1026 | "For test data, only keep the users and items that appear in the training/validation sets" 1027 | ] 1028 | }, 1029 | { 1030 | "cell_type": "code", 1031 | "execution_count": 114, 1032 | "metadata": { 1033 | "collapsed": true 1034 | }, 1035 | "outputs": [], 1036 | "source": [ 1037 | "test_raw_data = raw_data[int(0.8 * len(raw_data)):]" 1038 | ] 1039 | }, 1040 | { 1041 | "cell_type": "code", 1042 | "execution_count": 115, 1043 | "metadata": { 1044 | "collapsed": true 1045 | }, 1046 | "outputs": [], 1047 | "source": [ 1048 | "test_raw_data = test_raw_data[test_raw_data['movieId'].isin(unique_sid)]\n", 1049 | "test_raw_data = test_raw_data[test_raw_data['userId'].isin(unique_uid)]" 1050 | ] 1051 | }, 1052 | { 1053 | "cell_type": "code", 1054 | "execution_count": 116, 1055 | "metadata": { 1056 | "collapsed": false 1057 | }, 1058 | "outputs": [ 1059 | { 1060 | "name": "stdout", 1061 | "output_type": "stream", 1062 | "text": [ 1063 | "6993860 999003 207161\n" 1064 | ] 1065 | } 1066 | ], 1067 | "source": [ 1068 | "print len(train_raw_data), len(vad_raw_data), len(test_raw_data)" 1069 | ] 1070 | }, 1071 | { 1072 | "cell_type": "markdown", 1073 | "metadata": {}, 1074 | "source": [ 1075 | "Basic data information: what's the timespan for train/test?" 1076 | ] 1077 | }, 1078 | { 1079 | "cell_type": "code", 1080 | "execution_count": 117, 1081 | "metadata": { 1082 | "collapsed": false 1083 | }, 1084 | "outputs": [ 1085 | { 1086 | "name": "stdout", 1087 | "output_type": "stream", 1088 | "text": [ 1089 | "train: from 1995-01-09 06:46:44 to 2009-10-19 06:51:15\n", 1090 | "test: from 2009-10-19 06:51:53 to 2015-03-31 02:11:28\n" 1091 | ] 1092 | } 1093 | ], 1094 | "source": [ 1095 | "train_timestamp = np.asarray(tr_vd_raw_data['timestamp'])\n", 1096 | "print(\"train: from %s to %s\" % (timestamp_to_date(train_timestamp[0]), \n", 1097 | " timestamp_to_date(train_timestamp[-1])))\n", 1098 | "\n", 1099 | "test_timestamp = np.asarray(test_raw_data['timestamp'])\n", 1100 | "print(\"test: from %s to %s\" % (timestamp_to_date(test_timestamp[0]), \n", 1101 | " timestamp_to_date(test_timestamp[-1])))" 1102 | ] 1103 | }, 1104 | { 1105 | "cell_type": "markdown", 1106 | "metadata": {}, 1107 | "source": [ 1108 | "### Numerize the data into (timestamp, user_index, item_index) format" 1109 | ] 1110 | }, 1111 | { 1112 | "cell_type": "code", 1113 | "execution_count": 118, 1114 | "metadata": { 1115 | "collapsed": false 1116 | }, 1117 | "outputs": [], 1118 | "source": [ 1119 | "def numerize(tp):\n", 1120 | " uid = map(lambda x: user2id[x], tp['userId'])\n", 1121 | " sid = map(lambda x: song2id[x], tp['movieId'])\n", 1122 | " tp['uid'] = uid\n", 1123 | " tp['sid'] = sid\n", 1124 | " return tp[['timestamp', 'uid', 'sid']]" 1125 | ] 1126 | }, 1127 | { 1128 | "cell_type": "code", 1129 | "execution_count": 119, 1130 | "metadata": { 1131 | "collapsed": false 1132 | }, 1133 | "outputs": [], 1134 | "source": [ 1135 | "train_data = numerize(train_raw_data)\n", 1136 | "train_data.to_csv(os.path.join(DATA_DIR, 'pro', 'train.csv'), index=False)" 1137 | ] 1138 | }, 1139 | { 1140 | "cell_type": "code", 1141 | "execution_count": 120, 1142 | "metadata": { 1143 | "collapsed": true 1144 | }, 1145 | "outputs": [], 1146 | "source": [ 1147 | "vad_data = numerize(vad_raw_data)\n", 1148 | "vad_data.to_csv(os.path.join(DATA_DIR, 'pro', 'validation.csv'), index=False)" 1149 | ] 1150 | }, 1151 | { 1152 | "cell_type": "code", 1153 | "execution_count": 121, 1154 | "metadata": { 1155 | "collapsed": true 1156 | }, 1157 | "outputs": [], 1158 | "source": [ 1159 | "test_data = numerize(test_raw_data)\n", 1160 | "test_data.to_csv(os.path.join(DATA_DIR, 'pro', 'test.csv'), index=False)" 1161 | ] 1162 | }, 1163 | { 1164 | "cell_type": "code", 1165 | "execution_count": null, 1166 | "metadata": { 1167 | "collapsed": true 1168 | }, 1169 | "outputs": [], 1170 | "source": [] 1171 | } 1172 | ], 1173 | "metadata": { 1174 | "kernelspec": { 1175 | "display_name": "Python 2", 1176 | "language": "python", 1177 | "name": "python2" 1178 | }, 1179 | "language_info": { 1180 | "codemirror_mode": { 1181 | "name": "ipython", 1182 | "version": 2 1183 | }, 1184 | "file_extension": ".py", 1185 | "mimetype": "text/x-python", 1186 | "name": "python", 1187 | "nbconvert_exporter": "python", 1188 | "pygments_lexer": "ipython2", 1189 | "version": "2.7.6" 1190 | } 1191 | }, 1192 | "nbformat": 4, 1193 | "nbformat_minor": 0 1194 | } 1195 | -------------------------------------------------------------------------------- /src/Cofactorization_ML20M.ipynb: -------------------------------------------------------------------------------- 1 | { 2 | "cells": [ 3 | { 4 | "cell_type": "markdown", 5 | "metadata": {}, 6 | "source": [ 7 | "# Fit CoFactor model to the binarized ML20M" 8 | ] 9 | }, 10 | { 11 | "cell_type": "code", 12 | "execution_count": 1, 13 | "metadata": { 14 | "collapsed": false 15 | }, 16 | "outputs": [], 17 | "source": [ 18 | "import itertools\n", 19 | "import glob\n", 20 | "import os\n", 21 | "import sys\n", 22 | "os.environ['OPENBLAS_NUM_THREADS'] = '1'\n", 23 | "\n", 24 | "import numpy as np\n", 25 | "import matplotlib\n", 26 | "matplotlib.use('Agg')\n", 27 | "import matplotlib.pyplot as plt\n", 28 | "%matplotlib inline\n", 29 | "\n", 30 | "import pandas as pd\n", 31 | "from scipy import sparse\n", 32 | "import seaborn as sns\n", 33 | "sns.set(context=\"paper\", font_scale=1.5, rc={\"lines.linewidth\": 2}, font='DejaVu Serif')" 34 | ] 35 | }, 36 | { 37 | "cell_type": "code", 38 | "execution_count": 2, 39 | "metadata": { 40 | "collapsed": true 41 | }, 42 | "outputs": [], 43 | "source": [ 44 | "import cofacto\n", 45 | "import rec_eval" 46 | ] 47 | }, 48 | { 49 | "cell_type": "markdown", 50 | "metadata": {}, 51 | "source": [ 52 | "### Construct the positive pairwise mutual information (PPMI) matrix" 53 | ] 54 | }, 55 | { 56 | "cell_type": "markdown", 57 | "metadata": {}, 58 | "source": [ 59 | "Change this to wherever you saved the pre-processed data following [this notebook](./preprocess_ML20M.ipynb)." 60 | ] 61 | }, 62 | { 63 | "cell_type": "code", 64 | "execution_count": 3, 65 | "metadata": { 66 | "collapsed": true 67 | }, 68 | "outputs": [], 69 | "source": [ 70 | "DATA_DIR = '/hdd2/dawen/data/ml-20m/pro/'" 71 | ] 72 | }, 73 | { 74 | "cell_type": "code", 75 | "execution_count": 4, 76 | "metadata": { 77 | "collapsed": false 78 | }, 79 | "outputs": [], 80 | "source": [ 81 | "unique_uid = list()\n", 82 | "with open(os.path.join(DATA_DIR, 'unique_uid.txt'), 'r') as f:\n", 83 | " for line in f:\n", 84 | " unique_uid.append(line.strip())\n", 85 | " \n", 86 | "unique_sid = list()\n", 87 | "with open(os.path.join(DATA_DIR, 'unique_sid.txt'), 'r') as f:\n", 88 | " for line in f:\n", 89 | " unique_sid.append(line.strip())" 90 | ] 91 | }, 92 | { 93 | "cell_type": "code", 94 | "execution_count": 5, 95 | "metadata": { 96 | "collapsed": false 97 | }, 98 | "outputs": [ 99 | { 100 | "name": "stdout", 101 | "output_type": "stream", 102 | "text": [ 103 | "111148 11711\n" 104 | ] 105 | } 106 | ], 107 | "source": [ 108 | "n_items = len(unique_sid)\n", 109 | "n_users = len(unique_uid)\n", 110 | "\n", 111 | "print n_users, n_items" 112 | ] 113 | }, 114 | { 115 | "cell_type": "code", 116 | "execution_count": 6, 117 | "metadata": { 118 | "collapsed": true 119 | }, 120 | "outputs": [], 121 | "source": [ 122 | "def load_data(csv_file, shape=(n_users, n_items)):\n", 123 | " tp = pd.read_csv(csv_file)\n", 124 | " timestamps, rows, cols = np.array(tp['timestamp']), np.array(tp['uid']), np.array(tp['sid'])\n", 125 | " seq = np.concatenate((rows[:, None], cols[:, None], np.ones((rows.size, 1), dtype='int'), timestamps[:, None]), axis=1)\n", 126 | " data = sparse.csr_matrix((np.ones_like(rows), (rows, cols)), dtype=np.int16, shape=shape)\n", 127 | " return data, seq" 128 | ] 129 | }, 130 | { 131 | "cell_type": "code", 132 | "execution_count": 7, 133 | "metadata": { 134 | "collapsed": false 135 | }, 136 | "outputs": [], 137 | "source": [ 138 | "train_data, train_raw = load_data(os.path.join(DATA_DIR, 'train.csv'))" 139 | ] 140 | }, 141 | { 142 | "cell_type": "code", 143 | "execution_count": 8, 144 | "metadata": { 145 | "collapsed": true 146 | }, 147 | "outputs": [], 148 | "source": [ 149 | "watches_per_movie = np.asarray(train_data.astype('int64').sum(axis=0)).ravel()" 150 | ] 151 | }, 152 | { 153 | "cell_type": "code", 154 | "execution_count": 9, 155 | "metadata": { 156 | "collapsed": false 157 | }, 158 | "outputs": [ 159 | { 160 | "name": "stdout", 161 | "output_type": "stream", 162 | "text": [ 163 | "The mean (median) watches per movie is 597 (48)\n" 164 | ] 165 | } 166 | ], 167 | "source": [ 168 | "print(\"The mean (median) watches per movie is %d (%d)\" % (watches_per_movie.mean(), np.median(watches_per_movie)))" 169 | ] 170 | }, 171 | { 172 | "cell_type": "code", 173 | "execution_count": 10, 174 | "metadata": { 175 | "collapsed": true 176 | }, 177 | "outputs": [], 178 | "source": [ 179 | "user_activity = np.asarray(train_data.sum(axis=1)).ravel()" 180 | ] 181 | }, 182 | { 183 | "cell_type": "code", 184 | "execution_count": 11, 185 | "metadata": { 186 | "collapsed": false 187 | }, 188 | "outputs": [ 189 | { 190 | "name": "stdout", 191 | "output_type": "stream", 192 | "text": [ 193 | "The mean (median) movies each user wathced is 62 (33)\n" 194 | ] 195 | } 196 | ], 197 | "source": [ 198 | "print(\"The mean (median) movies each user wathced is %d (%d)\" % (user_activity.mean(), np.median(user_activity)))" 199 | ] 200 | }, 201 | { 202 | "cell_type": "code", 203 | "execution_count": 12, 204 | "metadata": { 205 | "collapsed": false 206 | }, 207 | "outputs": [], 208 | "source": [ 209 | "vad_data, vad_raw = load_data(os.path.join(DATA_DIR, 'validation.csv'))" 210 | ] 211 | }, 212 | { 213 | "cell_type": "code", 214 | "execution_count": 13, 215 | "metadata": { 216 | "collapsed": false 217 | }, 218 | "outputs": [ 219 | { 220 | "data": { 221 | "image/png": "iVBORw0KGgoAAAANSUhEUgAAAbQAAAE5CAYAAAD4GtVAAAAABHNCSVQICAgIfAhkiAAAAAlwSFlz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\n5AkMNCIi8gQGGhERecInh5yK783j73IAAAAASUVORK5CYII=\n", 222 | "text/plain": [ 223 | "" 224 | ] 225 | }, 226 | "metadata": {}, 227 | "output_type": "display_data" 228 | } 229 | ], 230 | "source": [ 231 | "plt.semilogx(1 + np.arange(n_users), -np.sort(-user_activity), 'o')\n", 232 | "plt.ylabel('Number of items that this user clicked on')\n", 233 | "plt.xlabel('User rank by number of consumed items')\n", 234 | "pass" 235 | ] 236 | }, 237 | { 238 | "cell_type": "code", 239 | "execution_count": 14, 240 | "metadata": { 241 | "collapsed": false 242 | }, 243 | "outputs": [ 244 | { 245 | "data": { 246 | "image/png": "iVBORw0KGgoAAAANSUhEUgAAAbwAAAE5CAYAAADrzZW0AAAABHNCSVQICAgIfAhkiAAAAAlwSFlz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\nAWwLndy7BvRhAAAAAElFTkSuQmCC\n", 247 | "text/plain": [ 248 | "" 249 | ] 250 | }, 251 | "metadata": {}, 252 | "output_type": "display_data" 253 | } 254 | ], 255 | "source": [ 256 | "plt.semilogx(1 + np.arange(n_items), -np.sort(-watches_per_movie), 'o')\n", 257 | "plt.ylabel('Number of users who watched this movie')\n", 258 | "plt.xlabel('Movie rank by number of watches')\n", 259 | "pass" 260 | ] 261 | }, 262 | { 263 | "cell_type": "markdown", 264 | "metadata": {}, 265 | "source": [ 266 | "### Generate co-occurrence matrix based on the user's entire watching history" 267 | ] 268 | }, 269 | { 270 | "cell_type": "code", 271 | "execution_count": 17, 272 | "metadata": { 273 | "collapsed": true 274 | }, 275 | "outputs": [], 276 | "source": [ 277 | "def _coord_batch(lo, hi, train_data):\n", 278 | " rows = []\n", 279 | " cols = []\n", 280 | " for u in xrange(lo, hi):\n", 281 | " for w, c in itertools.permutations(train_data[u].nonzero()[1], 2):\n", 282 | " rows.append(w)\n", 283 | " cols.append(c)\n", 284 | " np.save(os.path.join(DATA_DIR, 'coo_%d_%d.npy' % (lo, hi)),\n", 285 | " np.concatenate([np.array(rows)[:, None], np.array(cols)[:, None]], axis=1))\n", 286 | " pass" 287 | ] 288 | }, 289 | { 290 | "cell_type": "code", 291 | "execution_count": 18, 292 | "metadata": { 293 | "collapsed": false 294 | }, 295 | "outputs": [], 296 | "source": [ 297 | "from joblib import Parallel, delayed\n", 298 | "\n", 299 | "batch_size = 5000\n", 300 | "\n", 301 | "start_idx = range(0, n_users, batch_size)\n", 302 | "end_idx = start_idx[1:] + [n_users]\n", 303 | "\n", 304 | "Parallel(n_jobs=8)(delayed(_coord_batch)(lo, hi, train_data) for lo, hi in zip(start_idx, end_idx))\n", 305 | "pass" 306 | ] 307 | }, 308 | { 309 | "cell_type": "code", 310 | "execution_count": 19, 311 | "metadata": { 312 | "collapsed": false 313 | }, 314 | "outputs": [ 315 | { 316 | "name": "stdout", 317 | "output_type": "stream", 318 | "text": [ 319 | "User 0 to 5000 finished\n", 320 | "User 5000 to 10000 finished\n", 321 | "User 10000 to 15000 finished\n", 322 | "User 15000 to 20000 finished\n", 323 | "User 20000 to 25000 finished\n", 324 | "User 25000 to 30000 finished\n", 325 | "User 30000 to 35000 finished\n", 326 | "User 35000 to 40000 finished\n", 327 | "User 40000 to 45000 finished\n", 328 | "User 45000 to 50000 finished\n", 329 | "User 50000 to 55000 finished\n", 330 | "User 55000 to 60000 finished\n", 331 | "User 60000 to 65000 finished\n", 332 | "User 65000 to 70000 finished\n", 333 | "User 70000 to 75000 finished\n", 334 | "User 75000 to 80000 finished\n", 335 | "User 80000 to 85000 finished\n", 336 | "User 85000 to 90000 finished\n", 337 | "User 90000 to 95000 finished\n", 338 | "User 95000 to 100000 finished\n", 339 | "User 100000 to 105000 finished\n", 340 | "User 105000 to 110000 finished\n", 341 | "User 110000 to 111148 finished\n" 342 | ] 343 | } 344 | ], 345 | "source": [ 346 | "X = sparse.csr_matrix((n_items, n_items), dtype='float32')\n", 347 | "\n", 348 | "for lo, hi in zip(start_idx, end_idx):\n", 349 | " coords = np.load(os.path.join(DATA_DIR, 'coo_%d_%d.npy' % (lo, hi)))\n", 350 | " \n", 351 | " rows = coords[:, 0]\n", 352 | " cols = coords[:, 1]\n", 353 | " \n", 354 | " tmp = sparse.coo_matrix((np.ones_like(rows), (rows, cols)), shape=(n_items, n_items), dtype='float32').tocsr()\n", 355 | " X = X + tmp\n", 356 | " \n", 357 | " print(\"User %d to %d finished\" % (lo, hi))\n", 358 | " sys.stdout.flush()" 359 | ] 360 | }, 361 | { 362 | "cell_type": "markdown", 363 | "metadata": {}, 364 | "source": [ 365 | "Note: Don't forget to delete all the temporary coo_LO_HI.npy files" 366 | ] 367 | }, 368 | { 369 | "cell_type": "code", 370 | "execution_count": 20, 371 | "metadata": { 372 | "collapsed": true 373 | }, 374 | "outputs": [], 375 | "source": [ 376 | "np.save(os.path.join(DATA_DIR, 'coordinate_co_binary_data.npy'), X.data)\n", 377 | "np.save(os.path.join(DATA_DIR, 'coordinate_co_binary_indices.npy'), X.indices)\n", 378 | "np.save(os.path.join(DATA_DIR, 'coordinate_co_binary_indptr.npy'), X.indptr)" 379 | ] 380 | }, 381 | { 382 | "cell_type": "code", 383 | "execution_count": 21, 384 | "metadata": { 385 | "collapsed": false 386 | }, 387 | "outputs": [ 388 | { 389 | "data": { 390 | "text/plain": [ 391 | "0.38557504805354814" 392 | ] 393 | }, 394 | "execution_count": 21, 395 | "metadata": {}, 396 | "output_type": "execute_result" 397 | } 398 | ], 399 | "source": [ 400 | "float(X.nnz) / np.prod(X.shape)" 401 | ] 402 | }, 403 | { 404 | "cell_type": "markdown", 405 | "metadata": {}, 406 | "source": [ 407 | "### Or load the pre-saved co-occurrence matrix" 408 | ] 409 | }, 410 | { 411 | "cell_type": "code", 412 | "execution_count": 15, 413 | "metadata": { 414 | "collapsed": true 415 | }, 416 | "outputs": [], 417 | "source": [ 418 | "# or co-occurrence matrix from the entire user history\n", 419 | "dir_predix = DATA_DIR" 420 | ] 421 | }, 422 | { 423 | "cell_type": "code", 424 | "execution_count": 16, 425 | "metadata": { 426 | "collapsed": true 427 | }, 428 | "outputs": [], 429 | "source": [ 430 | "data = np.load(os.path.join(dir_predix, 'coordinate_co_binary_data.npy'))\n", 431 | "indices = np.load(os.path.join(dir_predix, 'coordinate_co_binary_indices.npy'))\n", 432 | "indptr = np.load(os.path.join(dir_predix, 'coordinate_co_binary_indptr.npy'))" 433 | ] 434 | }, 435 | { 436 | "cell_type": "code", 437 | "execution_count": 17, 438 | "metadata": { 439 | "collapsed": false 440 | }, 441 | "outputs": [], 442 | "source": [ 443 | "X = sparse.csr_matrix((data, indices, indptr), shape=(n_items, n_items))" 444 | ] 445 | }, 446 | { 447 | "cell_type": "code", 448 | "execution_count": 18, 449 | "metadata": { 450 | "collapsed": false 451 | }, 452 | "outputs": [ 453 | { 454 | "data": { 455 | "text/plain": [ 456 | "0.38557504805354814" 457 | ] 458 | }, 459 | "execution_count": 18, 460 | "metadata": {}, 461 | "output_type": "execute_result" 462 | } 463 | ], 464 | "source": [ 465 | "float(X.nnz) / np.prod(X.shape)" 466 | ] 467 | }, 468 | { 469 | "cell_type": "code", 470 | "execution_count": 19, 471 | "metadata": { 472 | "collapsed": true 473 | }, 474 | "outputs": [], 475 | "source": [ 476 | "def get_row(Y, i):\n", 477 | " lo, hi = Y.indptr[i], Y.indptr[i + 1]\n", 478 | " return lo, hi, Y.data[lo:hi], Y.indices[lo:hi]" 479 | ] 480 | }, 481 | { 482 | "cell_type": "code", 483 | "execution_count": 20, 484 | "metadata": { 485 | "collapsed": true 486 | }, 487 | "outputs": [], 488 | "source": [ 489 | "count = np.asarray(X.sum(axis=1)).ravel()" 490 | ] 491 | }, 492 | { 493 | "cell_type": "code", 494 | "execution_count": 21, 495 | "metadata": { 496 | "collapsed": true 497 | }, 498 | "outputs": [], 499 | "source": [ 500 | "n_pairs = X.data.sum()" 501 | ] 502 | }, 503 | { 504 | "cell_type": "markdown", 505 | "metadata": {}, 506 | "source": [ 507 | "### Construct the SPPMI matrix" 508 | ] 509 | }, 510 | { 511 | "cell_type": "code", 512 | "execution_count": 22, 513 | "metadata": { 514 | "collapsed": false 515 | }, 516 | "outputs": [], 517 | "source": [ 518 | "M = X.copy()\n", 519 | "\n", 520 | "for i in xrange(n_items):\n", 521 | " lo, hi, d, idx = get_row(M, i)\n", 522 | " M.data[lo:hi] = np.log(d * n_pairs / (count[i] * count[idx]))" 523 | ] 524 | }, 525 | { 526 | "cell_type": "code", 527 | "execution_count": 23, 528 | "metadata": { 529 | "collapsed": true 530 | }, 531 | "outputs": [], 532 | "source": [ 533 | "M.data[M.data < 0] = 0\n", 534 | "M.eliminate_zeros()" 535 | ] 536 | }, 537 | { 538 | "cell_type": "code", 539 | "execution_count": 24, 540 | "metadata": { 541 | "collapsed": false 542 | }, 543 | "outputs": [ 544 | { 545 | "name": "stdout", 546 | "output_type": "stream", 547 | "text": [ 548 | "0.237642385093\n" 549 | ] 550 | } 551 | ], 552 | "source": [ 553 | "print float(M.nnz) / np.prod(M.shape)" 554 | ] 555 | }, 556 | { 557 | "cell_type": "markdown", 558 | "metadata": {}, 559 | "source": [ 560 | "Now $M$ is the PPMI matrix. Depending on the number of negative examples $k$, we can obtain the shifted PPMI matrix as $\\max(M_{wc} - \\log k, 0)$" 561 | ] 562 | }, 563 | { 564 | "cell_type": "code", 565 | "execution_count": 25, 566 | "metadata": { 567 | "collapsed": true 568 | }, 569 | "outputs": [], 570 | "source": [ 571 | "# number of negative samples\n", 572 | "k_ns = 1\n", 573 | "\n", 574 | "M_ns = M.copy()\n", 575 | "\n", 576 | "if k_ns > 1:\n", 577 | " offset = np.log(k_ns)\n", 578 | "else:\n", 579 | " offset = 0.\n", 580 | " \n", 581 | "M_ns.data -= offset\n", 582 | "M_ns.data[M_ns.data < 0] = 0\n", 583 | "M_ns.eliminate_zeros()" 584 | ] 585 | }, 586 | { 587 | "cell_type": "code", 588 | "execution_count": 26, 589 | "metadata": { 590 | "collapsed": false 591 | }, 592 | "outputs": [ 593 | { 594 | "data": { 595 | "image/png": "iVBORw0KGgoAAAANSUhEUgAAAZcAAAEdCAYAAADATSqzAAAABHNCSVQICAgIfAhkiAAAAAlwSFlz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\n", 596 | "text/plain": [ 597 | "" 598 | ] 599 | }, 600 | "metadata": {}, 601 | "output_type": "display_data" 602 | } 603 | ], 604 | "source": [ 605 | "plt.hist(M_ns.data, bins=50)\n", 606 | "plt.yscale('log')\n", 607 | "pass" 608 | ] 609 | }, 610 | { 611 | "cell_type": "code", 612 | "execution_count": 27, 613 | "metadata": { 614 | "collapsed": false 615 | }, 616 | "outputs": [ 617 | { 618 | "data": { 619 | "text/plain": [ 620 | "0.23764238509276445" 621 | ] 622 | }, 623 | "execution_count": 27, 624 | "metadata": {}, 625 | "output_type": "execute_result" 626 | } 627 | ], 628 | "source": [ 629 | "float(M_ns.nnz) / np.prod(M_ns.shape)" 630 | ] 631 | }, 632 | { 633 | "cell_type": "markdown", 634 | "metadata": {}, 635 | "source": [ 636 | "### Train the model" 637 | ] 638 | }, 639 | { 640 | "cell_type": "code", 641 | "execution_count": 28, 642 | "metadata": { 643 | "collapsed": true 644 | }, 645 | "outputs": [], 646 | "source": [ 647 | "scale = 0.03\n", 648 | "\n", 649 | "n_components = 100\n", 650 | "max_iter = 20\n", 651 | "n_jobs = 8\n", 652 | "lam_theta = lam_beta = 1e-5 * scale\n", 653 | "lam_gamma = 1e-5\n", 654 | "c0 = 1. * scale\n", 655 | "c1 = 10. * scale\n", 656 | "\n", 657 | "save_dir = os.path.join(DATA_DIR, 'ML20M_ns%d_scale%1.2E' % (k_ns, scale))" 658 | ] 659 | }, 660 | { 661 | "cell_type": "code", 662 | "execution_count": 29, 663 | "metadata": { 664 | "collapsed": true 665 | }, 666 | "outputs": [], 667 | "source": [ 668 | "reload(cofacto)\n", 669 | "coder = cofacto.CoFacto(n_components=n_components, max_iter=max_iter, batch_size=1000, init_std=0.01, n_jobs=n_jobs, \n", 670 | " random_state=98765, save_params=True, save_dir=save_dir, early_stopping=True, verbose=True, \n", 671 | " lam_theta=lam_theta, lam_beta=lam_beta, lam_gamma=lam_gamma, c0=c0, c1=c1)" 672 | ] 673 | }, 674 | { 675 | "cell_type": "code", 676 | "execution_count": 30, 677 | "metadata": { 678 | "collapsed": false 679 | }, 680 | "outputs": [ 681 | { 682 | "name": "stdout", 683 | "output_type": "stream", 684 | "text": [ 685 | "ITERATION #0\n", 686 | "\tUpdating user factors: time=7.40\n", 687 | "\tUpdating item factors: time=6.99\n", 688 | "\tUpdating context factors: time=5.63\n", 689 | "\tUpdating bias terms: time=5.99\n", 690 | "\tValidation NDCG@k: 0.19163\n", 691 | "ITERATION #1\n", 692 | "\tUpdating user factors: time=5.65\n", 693 | "\tUpdating item factors: time=7.21\n", 694 | "\tUpdating context factors: time=5.19\n", 695 | "\tUpdating bias terms: time=5.94\n", 696 | "\tValidation NDCG@k: 0.26963\n", 697 | "ITERATION #2\n", 698 | "\tUpdating user factors: time=5.58\n", 699 | "\tUpdating item factors: time=7.58\n", 700 | "\tUpdating context factors: time=5.86\n", 701 | "\tUpdating bias terms: time=5.94\n", 702 | "\tValidation NDCG@k: 0.34485\n", 703 | "ITERATION #3\n", 704 | "\tUpdating user factors: time=5.93\n", 705 | "\tUpdating item factors: time=7.09\n", 706 | "\tUpdating context factors: time=5.57\n", 707 | "\tUpdating bias terms: time=5.90\n", 708 | "\tValidation NDCG@k: 0.35849\n", 709 | "ITERATION #4\n", 710 | "\tUpdating user factors: time=5.62\n", 711 | "\tUpdating item factors: time=7.35\n", 712 | "\tUpdating context factors: time=5.51\n", 713 | "\tUpdating bias terms: time=5.97\n", 714 | "\tValidation NDCG@k: 0.36298\n", 715 | "ITERATION #5\n", 716 | "\tUpdating user factors: time=5.52\n", 717 | "\tUpdating item factors: time=6.91\n", 718 | "\tUpdating context factors: time=5.61\n", 719 | "\tUpdating bias terms: time=5.96\n", 720 | "\tValidation NDCG@k: 0.36325\n", 721 | "ITERATION #6\n", 722 | "\tUpdating user factors: time=5.50\n", 723 | "\tUpdating item factors: time=7.12\n", 724 | "\tUpdating context factors: time=5.42\n", 725 | "\tUpdating bias terms: time=5.92\n", 726 | "\tValidation NDCG@k: 0.36150\n" 727 | ] 728 | }, 729 | { 730 | "data": { 731 | "text/plain": [ 732 | "CoFacto(batch_size=1000, dtype='float32', early_stopping=True, init_std=0.01,\n", 733 | " max_iter=20, n_components=100, n_jobs=8, random_state=98765,\n", 734 | " save_dir='/hdd2/dawen/data/ml-20m/pro/ML20M_ns1_scale3.00E-02',\n", 735 | " save_params=True, verbose=True)" 736 | ] 737 | }, 738 | "execution_count": 30, 739 | "metadata": {}, 740 | "output_type": "execute_result" 741 | } 742 | ], 743 | "source": [ 744 | "coder.fit(train_data, M_ns, vad_data=vad_data, batch_users=5000, k=100)" 745 | ] 746 | }, 747 | { 748 | "cell_type": "code", 749 | "execution_count": 31, 750 | "metadata": { 751 | "collapsed": true 752 | }, 753 | "outputs": [], 754 | "source": [ 755 | "test_data, _ = load_data(os.path.join(DATA_DIR, 'test.csv'))\n", 756 | "test_data.data = np.ones_like(test_data.data)" 757 | ] 758 | }, 759 | { 760 | "cell_type": "code", 761 | "execution_count": 32, 762 | "metadata": { 763 | "collapsed": true 764 | }, 765 | "outputs": [], 766 | "source": [ 767 | "n_params = len(glob.glob(os.path.join(save_dir, '*.npz')))\n", 768 | "\n", 769 | "params = np.load(os.path.join(save_dir, 'CoFacto_K%d_iter%d.npz' % (n_components, n_params - 1)))\n", 770 | "U, V = params['U'], params['V']" 771 | ] 772 | }, 773 | { 774 | "cell_type": "code", 775 | "execution_count": 33, 776 | "metadata": { 777 | "collapsed": false 778 | }, 779 | "outputs": [ 780 | { 781 | "name": "stdout", 782 | "output_type": "stream", 783 | "text": [ 784 | "Test Recall@20: 0.1448\n", 785 | "Test Recall@50: 0.1765\n", 786 | "Test NDCG@100: 0.1724\n", 787 | "Test MAP@100: 0.0545\n" 788 | ] 789 | } 790 | ], 791 | "source": [ 792 | "print 'Test Recall@20: %.4f' % rec_eval.recall_at_k(train_data, test_data, U, V, k=20, vad_data=vad_data)\n", 793 | "print 'Test Recall@50: %.4f' % rec_eval.recall_at_k(train_data, test_data, U, V, k=50, vad_data=vad_data)\n", 794 | "print 'Test NDCG@100: %.4f' % rec_eval.normalized_dcg_at_k(train_data, test_data, U, V, k=100, vad_data=vad_data)\n", 795 | "print 'Test MAP@100: %.4f' % rec_eval.map_at_k(train_data, test_data, U, V, k=100, vad_data=vad_data)" 796 | ] 797 | }, 798 | { 799 | "cell_type": "code", 800 | "execution_count": 34, 801 | "metadata": { 802 | "collapsed": true 803 | }, 804 | "outputs": [], 805 | "source": [ 806 | "np.savez('CoFactor_K100_ML20M.npz', U=U, V=V)" 807 | ] 808 | }, 809 | { 810 | "cell_type": "code", 811 | "execution_count": null, 812 | "metadata": { 813 | "collapsed": true 814 | }, 815 | "outputs": [], 816 | "source": [] 817 | } 818 | ], 819 | "metadata": { 820 | "kernelspec": { 821 | "display_name": "Python 2", 822 | "language": "python", 823 | "name": "python2" 824 | }, 825 | "language_info": { 826 | "codemirror_mode": { 827 | "name": "ipython", 828 | "version": 2 829 | }, 830 | "file_extension": ".py", 831 | "mimetype": "text/x-python", 832 | "name": "python", 833 | "nbconvert_exporter": "python", 834 | "pygments_lexer": "ipython2", 835 | "version": "2.7.6" 836 | } 837 | }, 838 | "nbformat": 4, 839 | "nbformat_minor": 0 840 | } 841 | --------------------------------------------------------------------------------