version 1.0 on DeepModel

hw6/CFModel.py

@@ -0,0 +1,47 @@
+# CFModel.py
+#
+# A simple implementation of matrix factorization for collaborative filtering
+# expressed as a Keras Sequential model. This code is based on the approach
+# outlined in [Alkahest](http:https://www.fenris.org/)'s blog post
+# [Collaborative Filtering in Keras](http:https://www.fenris.org/2016/03/07/collaborative-filtering-in-keras).
+#
+# License: MIT. See the LICENSE file for the copyright notice.
+#
+
+import numpy as np
+from keras.layers import Embedding, Reshape, Merge, Dropout, Dense
+from keras.models import Sequential
+
+class CFModel(Sequential):
+
+ def __init__(self, n_users, m_items, k_factors, **kwargs):
+ P = Sequential()
+ P.add(Embedding(n_users, k_factors, input_length=1))
+ P.add(Reshape((k_factors,)))
+ Q = Sequential()
+ Q.add(Embedding(m_items, k_factors, input_length=1))
+ Q.add(Reshape((k_factors,)))
+ super(CFModel, self).__init__(**kwargs)
+ self.add(Merge([P, Q], mode='dot', dot_axes=1))
+
+ def rate(self, user_id, item_id):
+ return self.predict([np.array([user_id]), np.array([item_id])])[0][0]
+
+class DeepModel(Sequential):
+
+ def __init__(self, n_users, m_items, k_factors, p_dropout=0.1, **kwargs):
+ P = Sequential()
+ P.add(Embedding(n_users, k_factors, input_length=1))
+ P.add(Reshape((k_factors,)))
+ Q = Sequential()
+ Q.add(Embedding(m_items, k_factors, input_length=1))
+ Q.add(Reshape((k_factors,)))
+ super(DeepModel, self).__init__(**kwargs)
+ self.add(Merge([P, Q], mode='concat'))
+ self.add(Dropout(p_dropout))
+ self.add(Dense(k_factors, activation='relu'))
+ self.add(Dropout(p_dropout))
+ self.add(Dense(1, activation='linear'))
+
+ def rate(self, user_id, item_id):
+ return self.predict([np.array([user_id]), np.array([item_id])])[0][0]

hw6/test.py

@@ -0,0 +1,67 @@
+import os
+import sys
+import argparse
+import numpy as np
+import pandas as pd
+from CFModel import CFModel, DeepModel
+
+def parse_args():
+ parser = argparse.ArgumentParser(description='HW6: Matrix Factorization')
+ parser.add_argument('train', type=str)
+ parser.add_argument('output', type=str)
+ return parser.parse_args()
+
+
+def predict_rating(trained_model, userid, movieid):
+ return trained_model.rate(userid - 1, movieid - 1)
+
+
+def ensure_dir(file_path):
+ directory = os.path.dirname(file_path)
+ if len(directory) == 0: return
+ if not os.path.exists(directory):
+ os.makedirs(directory)
+
+
+def main(args):
+ ratings = pd.read_csv(args.train, usecols=['UserID', 'MovieID', 'Rating'])
+ max_userid = ratings['UserID'].drop_duplicates().max()
+ max_movieid = ratings['MovieID'].drop_duplicates().max()
+ print('{} ratings loaded.'.format(ratings.shape[0]))
+
+ users = pd.read_csv(USERS_CSV, sep='::', engine='python',
+ usecols=['UserID', 'Gender', 'Age', 'Occupation', 'Zip-code'])
+ print('{} description of {} users loaded'.format(len(users), max_userid))
+
+ movies = pd.read_csv(MOVIES_CSV, sep='::', engine='python',
+ usecols=['movieID', 'Title', 'Genres'])
+ print('{} descriptions of {} movies loaded'.format(len(movies), max_movieid))
+
+ test_data = pd.read_csv(TEST_CSV, usecols=['UserID', 'MovieID'])
+ print('{} testing data loaded.'.format(test_data.shape[0]))
+
+ trained_model = DeepModel(max_userid, max_movieid, DIM)
+ print('Loading model weights...')
+ trained_model.load_weights(MODEL_WEIGHTS_FILE)
+ print('Loading model done!!!')
+
+ recommendations = pd.read_csv(TEST_CSV, usecols=['TestDataID'])
+ recommendations['Rating'] = test_data.apply(lambda x: predict_rating(trained_model, x['UserID'], x['MovieID']), axis=1)
+ # print(recommendations)
+
+ ensure_dir(args.output)
+ recommendations.to_csv(args.output, index=False, columns=['TestDataID', 'Rating'])
+
+
+if __name__ == '__main__':
+ args = parse_args()
+
+ TEST_CSV = 'data/test.csv'
+ USERS_CSV = 'data/users.csv'
+ MOVIES_CSV = 'data/movies.csv'
+ MODEL_WEIGHTS_FILE = 'weights.h5'
+
+ DIM = 120
+ TEST_USER = 3000
+
+ main(args)

hw6/train.py

@@ -0,0 +1,47 @@
+import os
+import sys
+import argparse
+import numpy as np
+import pandas as pd
+from keras.callbacks import Callback, EarlyStopping, ModelCheckpoint
+from CFModel import CFModel, DeepModel
+
+def parse_args():
+ parser = argparse.ArgumentParser(description='HW6: Matrix Factorization')
+ parser.add_argument('train', type=str)
+ parser.add_argument('test', type=str)
+ return parser.parse_args()
+
+
+def main(args):
+ ratings = pd.read_csv(args.train,
+ usecols=['UserID', 'MovieID', 'Rating'])
+ max_userid = ratings['UserID'].drop_duplicates().max()
+ max_movieid = ratings['MovieID'].drop_duplicates().max()
+ ratings['User_emb_id'] = ratings['UserID'] - 1
+ ratings['Movie_emb_id'] = ratings['MovieID'] - 1
+ print('{} ratings loaded.'.format(ratings.shape[0]))
+
+ ratings = ratings.sample(frac=1)
+ Users = ratings['User_emb_id'].values
+ print('Users: {}, shape = {}'.format(Users, Users.shape))
+ Movies = ratings['Movie_emb_id'].values
+ print('Movies: {}, shape = {}'.format(Movies, Movies.shape))
+ Ratings = ratings['Rating'].values
+ print('Ratings: {}, shape = {}'.format(Ratings, Ratings.shape))
+
+ model = DeepModel(max_userid, max_movieid, DIM)
+ model.compile(loss='mse', optimizer='adamax')
+
+ callbacks = [EarlyStopping('val_loss', patience=2),
+ ModelCheckpoint(MODEL_WEIGHTS_FILE, save_best_only=True)]
+ history = model.fit([Users, Movies], Ratings, epochs=100, validation_split=.1, verbose=1, callbacks=callbacks)
+
+
+if __name__ == '__main__':
+ args = parse_args()
+
+ DIM = 120
+ MODEL_WEIGHTS_FILE = 'weights.h5'
+
+ main(args)