report 1: normailization

hw6/Model.py

@@ -12,13 +12,13 @@ def build_cf_model(n_users, n_movies, dim):
  m = Embedding(n_movies, dim)(m_input)
  m = Reshape((dim,))(m)
 
- u_bias = Embedding(n_users, 1)(u_input)
- u_bias = Reshape((1,))(u_bias)
- m_bias = Embedding(n_movies, 1)(m_input)
- m_bias = Reshape((1,))(m_bias)
+ # u_bias = Embedding(n_users, 1)(u_input)
+ # u_bias = Reshape((1,))(u_bias)
+ # m_bias = Embedding(n_movies, 1)(m_input)
+ # m_bias = Reshape((1,))(m_bias)
 
  out = dot([u, m], -1)
- out = add([out, u_bias, m_bias])
+ # out = add([out, u_bias, m_bias])
 
  model = Model(inputs=[u_input, m_input], outputs=out)
  return model

hw6/test.py

@@ -42,7 +42,7 @@ def main(args):
  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)
+ recommendations['Rating'] = test_data.apply(lambda x: predict_rating(trained_model, x['UserID'], x['MovieID']) * std + mean, axis=1)
  # print(recommendations)
 
  ensure_dir(args.output)
@@ -70,5 +70,7 @@ def main(args):
  DIM = list(info['dim'])[0]
  max_userid = list(info['max_userid'])[0]
  max_movieid = list(info['max_movieid'])[0]
+ mean = list(info['mean'])[0]
+ std = list(info['std'])[0]
 
  main(args)

hw6/train.py

@@ -17,7 +17,7 @@ def parse_args():
 
 
 def rmse(y_true, y_pred):
- y_pred = K.clip(y_pred, 1., 5.)
+ # y_pred = K.clip(y_pred, 1., 5.)
  return K.sqrt(K.mean(K.square((y_true - y_pred))))
 
 def main(args):
@@ -29,20 +29,25 @@ def main(args):
  ratings['Movie_emb_id'] = ratings['MovieID'] - 1
  print('{} ratings loaded.'.format(ratings.shape[0]))
 
- 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))
+ mean = Ratings.mean()
+ std = Ratings.std()
+ Ratings = (Ratings - mean) / (std + 1e-100)
+
+ maximum = {}
+ maximum['max_userid'] = [max_userid]
+ maximum['max_movieid'] = [max_movieid]
+ maximum['dim'] = [DIM]
+ maximum['mean'] = [mean]
+ maximum['std'] = [std]
+ pd.DataFrame(data=maximum).to_csv(MAX_FILE, index=False)
+ print('max info save to {}'.format(MAX_FILE))
 
  model = build_cf_model(max_userid, max_movieid, DIM)
  model.compile(loss='mse', optimizer='adamax', metrics=[rmse])