def __init__(self, small, batch_size):
BaseRecommender.__init__(self, small, batch_size)
self.user_num = self.reader.get_user_num()
self.prod_num = self.reader.get_prod_num()
self.batch_num_train = self.reader.get_batch_num_train()
self.batch_num_test = self.reader.get_batch_num_test()
self.epoch_num = 1 #50
print "Number of Users : %d\tNumber of Prods : %d" \
% (self.user_num, self.prod_num)
print "Number of train batches : %d\tNumber of test batches : %d" \
% (self.batch_num_train, self.batch_num_test)
def __init__(self, small, batch_size): BaseRecommender.__init__(self, small, batch_size) self.user_num = self.reader.get_user_num() self.prod_num = self.reader.get_prod_num() self.batch_num_train = self.reader.get_batch_num_train() self.batch_num_test = self.reader.get_batch_num_test() self.epoch_num = 1 #50 print "Number of Users : %d\tNumber of Prods : %d" \ % (self.user_num, self.prod_num) print "Number of train batches : %d\tNumber of test batches : %d" \ % (self.batch_num_train, self.batch_num_test)
def __init__(self, small, batch_size):
BaseRecommender.__init__(self, small, batch_size)
self.user_num = self.reader.get_user_num()
self.prod_num = self.reader.get_prod_num()
self.batch_num_train = self.reader.get_batch_num_train()
self.batch_num_test = self.reader.get_batch_num_test()
print("Number of Users : %d\tNumber of Prods : %d" %
(self.user_num, self.prod_num))
print
"Number of train batches : {:d}\tNumber of test batches : {:d}" \
.format(self.batch_num_train, self.batch_num_test)
self.prods_profile = None
self.users_profile = None
def __init__(self, small, batch_size):
BaseRecommender.__init__(self, small, batch_size)
self.user_num = self.reader.get_user_num()
self.prod_num = self.reader.get_prod_num()
self.batch_num_train = self.reader.get_batch_num_train()
self.batch_num_test = self.reader.get_batch_num_test()
print "Number of Users : %d\tNumber of Prods : %d" \
% (self.user_num, self.prod_num)
print "Number of train batches : %d\tNumber of test batches : %d" \
% (self.batch_num_train, self.batch_num_test)
self.overall_avg_score = None
self.prod_avg = None
self.similarity_matrix = None
def __init__(self, small, batch_size, factor_num, wd = 0.001):
BaseRecommender.__init__(self, small, batch_size)
self.user_num = self.reader.get_user_num()
self.prod_num = self.reader.get_prod_num()
self.batch_num_train = self.reader.get_batch_num_train()
self.batch_num_test = self.reader.get_batch_num_test()
self.factor_num = factor_num
self.user_dic = {}
self.epoch_num = 20
self.wd = 0
print "Number of Users : %d\tNumber of Prods : %d" \
% (self.user_num, self.prod_num)
print "Number of train batches : %d\tNumber of test batches : %d" \
% (self.batch_num_train, self.batch_num_test)
def __init__(self, small, batch_size, factor_num):
BaseRecommender.__init__(self, small, batch_size)
self.user_num = self.reader.get_user_num()
self.prod_num = self.reader.get_prod_num()
self.batch_num_train = self.reader.get_batch_num_train()
self.batch_num_test = self.reader.get_batch_num_test()
self.factor_num = factor_num
self.user_dic = {}
self.epoch_num = 3 #50
print("Number of Users : %d\tNumber of Prods : %d"
% (self.user_num, self.prod_num))
print("Number of train batches : %d\tNumber of test batches : %d"
% (self.batch_num_train, self.batch_num_test))
def build(self):
("\n"
" build recommender system\n"
" ")
# Variables
self.Wp = tf.Variable(tf.random_uniform([self.prod_num, self.factor_num], -1.0, 1.0), name='Wp')
self.Wu = tf.Variable(tf.random_uniform([self.user_num, self.factor_num], -1.0, 1.0), name='Wu')
self.bp = tf.Variable(tf.zeros([self.prod_num]), name='bp')
self.bu = tf.Variable(tf.zeros([self.user_num]), name='bu')
self.b = tf.Variable(0.0, name='b')
# Placeholders
self.u_ind = tf.placeholder('int32', shape=[self.batch_size])
self.y_rate = tf.placeholder('float32', shape=[self.batch_size, self.prod_num])
self.y_mask = tf.placeholder('float32', shape=[self.batch_size, self.prod_num])
# models
for i in range(self.batch_size):
ind = tf.gather(self.u_ind, i)
currU = tf.gather(self.Wu, ind)
currBu = tf.reshape(tf.gather(self.bu, ind), [1, 1])
if i==0:
U = currU
Bu = currBu
else:
U = tf.concat(0, [U, currU])
Bu = tf.concat(0, [Bu, currBu])
# U : [batch_size, factor_num]
# Wp : [prod_num, factor_num]
# Bu : [batch_size]
Umat = tf.reshape(U, [self.batch_size, self.factor_num])
Pmat = tf.transpose(self.Wp)
biasU = tf.reshape(tf.tile(tf.reshape(Bu, [self.batch_size]), [self.prod_num]), [self.batch_size, self.prod_num])
biasP = tf.transpose(tf.reshape(tf.tile(self.bp, [self.batch_size]), [self.prod_num, self.batch_size]))
b_expand = tf.expand_dims(self.b, 0)
bias = tf.reshape(tf.tile(b_expand, [self.batch_size*self.prod_num]), [self.batch_size, self.prod_num])
self.pred_y_rate = tf.matmul(Umat, Pmat) + biasU + biasP + bias
loss_m = tf.sqrt(tf.squared_difference(self.y_rate, self.pred_y_rate))
self.loss = tf.reduce_sum(loss_m * self.y_mask)/self.batch_size
optimizer = tf.train.AdamOptimizer(0.01)
train_op = optimizer.minimize(self.loss)
#training
self.sess = tf.Session()
self.sess.run(tf.initialize_all_variables())
for epoch in range(self.epoch_num):
t = time.time()
tot_loss = 0.0
for i in range(self.batch_num_train):
_, loss = self.sess.run([train_op, self.loss], feed_dict = self.get_feed_dict(self.reader.get_next_train()))
tot_loss += loss
print (loss)
avg_loss = tot_loss / self.batch_num_train
print ("Epoch %d\tLoss\t%.2f\tTime %dmin" \
% (epoch, avg_loss, (time.time()-t)/60))
print ("Recommender is built!")