def make_records(self): # record all the results' details into files
_, Decoder, l1, l2 = self.sess.run(
[self.cost, self.Decoder, self.pre_cost3, self.pre_cost4],
feed_dict={
self.input_R_U: self.train_R,
self.input_R_I: self.train_R,
self.input_OH_U: self.U_OH_mat,
self.input_OH_I: self.I_OH_mat,
self.input_P_cor: [[0, 0]],
self.input_N_cor: [[0, 0]],
self.row_idx: np.reshape(range(self.num_rows),
(self.num_rows, 1)),
self.col_idx: np.reshape(range(self.num_cols),
(self.num_cols, 1))
})
if self.base == 'i':
[precision, recall, f_score, NDCG,
r_f_table] = utility.test_model_all(Decoder.T, self.vali_R.T,
self.train_R.T)
else:
[precision, recall, f_score, NDCG,
r_f_table] = utility.test_model_all(Decoder, self.vali_R,
self.train_R)
utility.metric_record(precision, recall, f_score, NDCG, self.args,
self.metric_path)
utility.test_model_factor(Decoder, self.vali_R, self.train_R)
print("******** max_epoch ********")
print(self.max_epoch)
return precision, recall, f_score, NDCG
def make_records(self): # record all the results' details into files
_, Decoder = self.sess.run(
[self.cost, self.Decoder],
feed_dict={
self.input_R_U: self.train_R,
self.input_R_U_index: self.U_OH_mat,
self.is_training_ph: 0,
self.anneal_ph: 0.2,
self.row_idx: np.reshape(range(self.num_rows),
(self.num_rows, 1))
})
if self.base == 'i':
[precision, recall, f_score, NDCG,
r_f_table] = utility.test_model_all(Decoder.T, self.vali_R.T,
self.train_R.T)
else:
[precision, recall, f_score, NDCG,
r_f_table] = utility.test_model_all(Decoder, self.vali_R,
self.train_R)
utility.metric_record(precision, recall, f_score, NDCG, self.args,
self.metric_path)
utility.test_model_factor(Decoder, self.vali_R, self.train_R)
print("******** max_epoch ********")
print(self.max_epoch)
return precision, recall, f_score, NDCG
def make_records(self): # record all the results' details into files
print("in make record. we will calculate metrics after here.")
_, Decoder = self.sess.run(
[self.cost, self.Decoder],
feed_dict={
self.input_R_U: self.train_R,
self.input_R_I: self.train_R,
self.input_OH_I: self.I_OH_mat,
self.input_P_cor: [[0, 0]],
self.input_N_cor: [[0, 0]],
self.row_idx: np.reshape(xrange(self.num_rows),
(self.num_rows, 1)),
self.col_idx: np.reshape(xrange(self.num_cols),
(self.num_cols, 1))
})
if self.base == 'i':
[precision, recall, f_score,
NDCG] = utility.test_model_all(Decoder.T, self.vali_R.T,
self.train_R.T, self.price_R.T,
'save_intermediate_results')
else:
[precision, recall, f_score,
NDCG] = utility.test_model_all(Decoder, self.vali_R, self.train_R,
self.price_R,
'save_intermediate_results')
utility.metric_record(precision, recall, f_score, NDCG, self.args,
self.metric_path)
utility.test_model_factor(Decoder, self.vali_R, self.train_R)
return precision, recall, f_score, NDCG
def test_model(
self,
itr): # calculate the cost and rmse of testing set in each epoch
start_time = time.time()
corruption_R = np.random.binomial(1, 1 - self.corruption_ratio,
(self.num_rows, self.num_cols))
_, plot1, plot2, scaling, Decoder = self.sess.run(
[
self.cost, self.plotting, self.plotting2, self.scaling,
self.Decoder
],
feed_dict={
self.input_R_U: self.train_R,
self.input_R_U_index: self.U_OH_mat,
self.corrupted_R: corruption_R,
self.input_P_cor: [[0, 0]],
self.input_N_cor: [[0, 0]],
self.row_idx: np.reshape(range(self.num_rows),
(self.num_rows, 1)),
self.col_idx: np.reshape(range(self.num_cols),
(self.num_cols, 1))
})
if itr % self.display_step == 0:
pre_numerator = np.multiply((Decoder - self.vali_R), self.vali_R)
numerator = np.sum(np.square(pre_numerator))
denominator = self.num_test_ratings
RMSE = np.sqrt(numerator / float(denominator))
if itr % 1 == 0:
if self.base == 'i':
[precision, recall, f_score, NDCG, r_f_table
] = utility.test_model_all(Decoder.T, self.vali_R.T,
self.train_R.T)
else:
[precision, recall, f_score, NDCG, r_f_table
] = utility.test_model_all(Decoder, self.vali_R,
self.train_R)
if self.max_f1_avg < f_score[2]:
self.max_f1_avg = f_score[2]
self.max_r_f_table = r_f_table
self.max_epoch = itr
self.max_embedded_x = plot1
self.max_embedded2_x = plot2
self.max_scaling = scaling
print("Testing //", "Epoch %d //" % itr,
" Total cost = {:.2f}".format(numerator),
" RMSE = {:.5f}".format(RMSE),
"Elapsed time : %d sec" % (time.time() - start_time))
print("=" * 100)
def test_model(
self,
itr): # calculate the cost and rmse of testing set in each epoch
start_time = time.time()
_, Encoder, Decoder = self.sess.run(
[
self.cost, self.U_Encoder[:, :self.U_hidden_neuron],
self.Decoder
],
feed_dict={
self.input_R_U: self.train_R,
self.input_R_U_index: self.U_OH_mat,
self.is_training_ph: 0,
self.anneal_ph: 0.2,
self.row_idx: np.reshape(range(self.num_rows),
(self.num_rows, 1))
})
if itr % self.display_step == 0:
pre_numerator = np.multiply((Decoder - self.vali_R), self.vali_R)
numerator = np.sum(np.square(pre_numerator))
denominator = self.num_test_ratings
RMSE = np.sqrt(numerator / float(denominator))
if itr % 1 == 0:
if self.base == 'i':
[precision, recall, f_score, NDCG, r_f_table
] = utility.test_model_all(Decoder.T, self.vali_R.T,
self.train_R.T)
else:
[precision, recall, f_score, NDCG, r_f_table
] = utility.test_model_all(Decoder, self.vali_R,
self.train_R)
if self.max_f1_avg < f_score[2]:
self.max_f1_avg = f_score[2]
self.max_r_f_table = r_f_table
self.max_epoch = itr
self.max_embedded_x = Encoder
print("Testing //", "Epoch %d //" % itr,
" Total cost = {:.2f}".format(numerator),
" RMSE = {:.5f}".format(RMSE),
"Elapsed time : %d sec" % (time.time() - start_time))
print("=" * 100)
def test_model(
self,
itr): # calculate the cost and rmse of testing set in each epoch
start_time = time.time()
_, Decoder = self.sess.run(
[self.cost, self.Decoder],
feed_dict={
self.input_R_U: self.train_R,
self.input_R_I: self.train_R,
self.input_OH_I: self.I_OH_mat,
self.input_P_cor: [[0, 0]],
self.input_N_cor: [[0, 0]],
self.row_idx: np.reshape(xrange(self.num_rows),
(self.num_rows, 1)),
self.col_idx: np.reshape(xrange(self.num_cols),
(self.num_cols, 1))
})
if itr % self.display_step == 0:
pre_numerator = np.multiply((Decoder - self.vali_R), self.vali_R)
numerator = np.sum(np.square(pre_numerator))
denominator = self.num_test_ratings
RMSE = np.sqrt(numerator / float(denominator))
if itr % 1 == 0:
if self.base == 'i':
[precision, recall, f_score,
NDCG] = utility.test_model_all(Decoder.T, self.vali_R.T,
self.train_R.T,
self.price_R.T)
else:
[precision, recall, f_score,
NDCG] = utility.test_model_all(Decoder, self.vali_R,
self.train_R, self.price_R)
print("Testing //", "Epoch %d //" % itr,
" Total cost = {:.2f}".format(numerator),
" RMSE = {:.5f}".format(RMSE),
"Elapsed time : %d sec" % (time.time() - start_time))
print("=" * 100)
def test_model(self, itr): # calculate the cost and rmse of testing set in each epoch
if itr % self.display_step == 0:
start_time = time.time() * 1000.0
P, Q = self.sess.run([self.P, self.Q])
Rec = np.matmul(P, Q.T)
[precision, recall, f_score, NDCG] = utility.test_model_all(Rec, self.vali_df, self.train_df)
utility.ranking_analysis(Rec, self.vali_df, self.train_df, self.key_genre, self.item_genre_list,
self.user_genre_count)
print((
"Testing //", "Epoch %d //" % itr,
"Testing time : %d ms" % (time.time() * 1000.0 - start_time)))
print(("=" * 200))
def make_records(self): # record all the results' details into files
P, Q = self.sess.run([self.P, self.Q])
Rec = np.matmul(P, Q.T)
[precision, recall, f_score, NDCG] = utility.test_model_all(Rec, self.vali_df, self.train_df)
return precision, recall, f_score, NDCG, Rec
