bonus: on deep model

hw6/Model.py

@@ -1,5 +1,6 @@
 import numpy as np
-from keras.layers import Input, Embedding, Reshape, Dense, Dropout, Lambda
+from keras.layers import Input, Embedding, Dense, Dropout
+from keras.layers import Reshape, Flatten, Lambda
 from keras.layers.merge import concatenate, dot, add
 from keras.models import Model
 from keras import backend as K
@@ -43,18 +44,26 @@ def build_cf_model(n_users, n_movies, dim, isBest=False):
 
 
 def build_deep_model(n_users, n_movies, dim, dropout=0.1):
- u_input = Input(shape=(1,))
+ u_input = Input(shape=(4,))
  u = Embedding(n_users, dim)(u_input)
- u = Reshape((dim,))(u)
+ # u = Reshape((dim,))(u)
+ u = Flatten()(u)
 
- m_input = Input(shape=(1,))
+ m_input = Input(shape=(19,))
  m = Embedding(n_movies, dim)(m_input)
- m = Reshape((dim,))(m)
+ # m = Reshape((dim,))(m)
+ m = Flatten()(m)
 
  out = concatenate([u, m])
  out = Dropout(dropout)(out)
- out = Dense(dim, activation='relu')(out)
+ out = Dense(256, activation='relu')(out)
+ out = Dropout(dropout)(out)
+ out = Dense(128, activation='relu')(out)
  out = Dropout(dropout)(out)
+ out = Dense(64, activation='relu')(out)
+ out = Dropout(0.15)(out)
+ out = Dense(dim, activation='relu')(out)
+ out = Dropout(0.2)(out)
  out = Dense(1, activation='relu')(out)
 
  model = Model(inputs=[u_input, m_input], outputs=out)

hw6/test_bonus.py

@@ -0,0 +1,95 @@
+import os
+import sys
+import argparse
+import numpy as np
+import pandas as pd
+from Model import build_cf_model, build_deep_model, rate
+
+classes = ["Adventure", "Western", "Comedy", "Thriller", "Horror", "Mystery", "Crime", "Film-Noir", "Sci-Fi", "Fantasy", "Drama", "Musical", "War", "Documentary", "Children's", "Animation", "Action", "Romance"]
+
+
+def parse_args():
+ parser = argparse.ArgumentParser(description='HW6: Matrix Factorization')
+ parser.add_argument('data_dir', type=str)
+ parser.add_argument('output', type=str)
+ return parser.parse_args()
+
+def make_users(row, matrix):
+ matrix[row['UserID']] = [row['UserID'], row['Gender'], row['Age'], row['Occupation']]
+ return row
+
+def categorize_movie(row, matrix, idx_map):
+ x = [0] * len(classes)
+ for g in row['Genres'].split('|'):
+ x[idx_map[g]] = 1
+ matrix[row['movieID']] = [row['movieID']] + x
+
+def predict_rating(trained_model, userid, movieid):
+ return rate(trained_model, userid, movieid)
+
+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):
+ users = pd.read_csv(USERS_CSV, sep='::', engine='python',
+ usecols=['UserID', 'Gender', 'Age', 'Occupation', 'Zip-code'])
+ users['UserID'] -= 1
+ users['Gender'][users['Gender'] == 'F'] = 0
+ users['Gender'][users['Gender'] == 'M'] = 1
+ users_mx = {}
+ users.apply(lambda x: make_users(x, users_mx), axis=1)
+ print('{} description of {} users loaded'.format(len(users), max_userid))
+
+ movies = pd.read_csv(MOVIES_CSV, sep='::', engine='python',
+ usecols=['movieID', 'Title', 'Genres'])
+ movies['movieID'] -= 1
+ movies_mx = {}
+ classes_idx = {}
+ for i, c in enumerate(classes):
+ classes_idx[c] = i
+ movies.apply(lambda x: categorize_movie(x, movies_mx, classes_idx), axis=1)
+ 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 = build_deep_model(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, users_mx[x['UserID']-1], movies_mx[x['MovieID']-1]), axis=1)
+ # print(recommendations)
+
+ ensure_dir(args.output)
+ recommendations.to_csv(args.output, index=False, columns=['TestDataID', 'Rating'])
+
+
+if __name__ == '__main__':
+ args = parse_args()
+
+ MODEL_DIR = './model'
+ MAX_CSV = 'max_bonus.csv'
+ TEST_CSV = 'test.csv'
+ USERS_CSV = 'users.csv'
+ MOVIES_CSV = 'movies.csv'
+ MODEL_WEIGHTS_FILE = 'weights_bonus.h5'
+
+ DATA_DIR = args.data_dir
+ TEST_CSV = os.path.join(DATA_DIR, TEST_CSV)
+ USERS_CSV = os.path.join(DATA_DIR, USERS_CSV)
+ MOVIES_CSV = os.path.join(DATA_DIR, MOVIES_CSV)
+
+ MODEL_WEIGHTS_FILE = os.path.join(MODEL_DIR, MODEL_WEIGHTS_FILE)
+ MAX_CSV = os.path.join(MODEL_DIR, MAX_CSV)
+ info = pd.read_csv(MAX_CSV)
+ DIM = list(info['dim'])[0]
+ max_userid = list(info['max_userid'])[0]
+ max_movieid = list(info['max_movieid'])[0]
+
+ main(args)

hw6/train_bonus.py

@@ -0,0 +1,102 @@
+import os
+import sys
+import argparse
+import numpy as np
+import pandas as pd
+from keras import backend as K
+from keras.callbacks import Callback, EarlyStopping, ModelCheckpoint
+from Model import build_cf_model, build_deep_model, rate
+
+classes = ["Adventure", "Western", "Comedy", "Thriller", "Horror", "Mystery", "Crime", "Film-Noir", "Sci-Fi", "Fantasy", "Drama", "Musical", "War", "Documentary", "Children's", "Animation", "Action", "Romance"]
+
+
+def parse_args():
+ parser = argparse.ArgumentParser(description='HW6: Matrix Factorization')
+ parser.add_argument('train', type=str)
+ parser.add_argument('users', type=str)
+ parser.add_argument('movies', type=str)
+ parser.add_argument('test', type=str)
+ parser.add_argument('--dim', type=int, default=15)
+ return parser.parse_args()
+
+def make_users(row, matrix):
+ matrix[row['UserID']] = [row['UserID'], row['Gender'], row['Age'], row['Occupation']]
+ return row
+
+def categorize_movie(row, matrix, idx_map):
+ x = [0] * len(classes)
+ for g in row['Genres'].split('|'):
+ x[idx_map[g]] = 1
+ matrix[row['movieID']] = [row['movieID']] + x
+
+def rmse(y_true, y_pred):
+ y_pred = K.clip(y_pred, 1., 5.)
+ return K.sqrt(K.mean(K.square((y_true - y_pred))))
+
+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]))
+
+ users = pd.read_csv(args.users, sep='::', engine='python',
+ usecols=['UserID', 'Gender', 'Age', 'Occupation'])
+ users['UserID'] -= 1
+ users['Gender'][users['Gender'] == 'F'] = 0
+ users['Gender'][users['Gender'] == 'M'] = 1
+ users_mx = {}
+ users.apply(lambda x: make_users(x, users_mx), axis=1)
+ print('{} description of {} users loaded'.format(len(users), max_userid))
+
+ movies = pd.read_csv(args.movies, sep='::', engine='python',
+ usecols=['movieID', 'Genres'])
+ movies['movieID'] -= 1
+ movies_mx = {}
+ classes_idx = {}
+ for i, c in enumerate(classes):
+ classes_idx[c] = i
+ movies.apply(lambda x: categorize_movie(x, movies_mx, classes_idx), axis=1)
+ print('{} descriptions of {} movies loaded'.format(len(movies), max_movieid))
+
+ maximum = {}
+ maximum['max_userid'] = [max_userid]
+ maximum['max_movieid'] = [max_movieid]
+ maximum['dim'] = [DIM]
+ pd.DataFrame(data=maximum).to_csv(MAX_FILE, index=False)
+ print('max info save to {}'.format(MAX_FILE))
+
+ 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))
+
+ new_Users = np.array(list(map(users_mx.get, Users)))
+ new_Movies = np.array(list(map(movies_mx.get, Movies)))
+
+ model = build_deep_model(max_userid, max_movieid, DIM)
+ model.compile(loss='mse', optimizer='adamax', metrics=[rmse])
+
+ callbacks = [EarlyStopping('val_rmse', patience=2),
+ ModelCheckpoint(MODEL_WEIGHTS_FILE, save_best_only=True)]
+ history = model.fit([new_Users, new_Movies], Ratings, epochs=1000, batch_size=256, validation_split=.1, verbose=1, callbacks=callbacks)
+
+
+if __name__ == '__main__':
+ args = parse_args()
+
+ MODEL_DIR = './model'
+ DIM = args.dim
+ MODEL_WEIGHTS_FILE = 'weights_bonus.h5'
+ MAX_FILE = 'max_bonus.csv'
+
+ if not os.path.exists(MODEL_DIR):
+ os.makedirs(MODEL_DIR)
+ MODEL_WEIGHTS_FILE = os.path.join(MODEL_DIR, MODEL_WEIGHTS_FILE)
+ MAX_FILE = os.path.join(MODEL_DIR, MAX_FILE)
+ main(args)