dataset = Dataset(config.filename)
config.item_count = dataset.item_count
config.user_count = dataset.user_count
tf.logging.info("\n\n%s\n\n" % config)
model = PairwiseGMF(config)
sv = tf.train.Supervisor(logdir=None, save_model_secs=0, save_summaries_secs=0)
sess = sv.prepare_or_wait_for_session(
config=tf.ConfigProto(gpu_options=tf.GPUOptions(
per_process_gpu_memory_fraction=0.1,
allow_growth=True)))
for i in range(FLAGS.iters):
if sv.should_stop():
break
progress = tqdm(enumerate(dataset.get_data(FLAGS.batch_size, False, FLAGS.neg)),
dynamic_ncols=True, total=(dataset.train_size * FLAGS.neg) // FLAGS.batch_size)
loss = []
for k, example in progress:
feed = {
model.input_users: example[:, 0],
model.input_items: example[:, 1],
model.input_items_negative: example[:, 2],
}
batch_loss, _ = sess.run([model.loss, model.train], feed)
loss.append(batch_loss)
progress.set_description(u"[{}] Loss: {:,.4f} » » » » ".format(i, batch_loss))
print("Epoch {}: Avg Loss/Batch {:<20,.6f}".format(i, np.mean(loss)))
user_embed, item_embed, v = sess.run([model.user_memory.embeddings, model.item_memory.embeddings, model.v.w])
dataset = Dataset(config.filename)
config.item_count = dataset.item_count
config.user_count = dataset.user_count
tf.logging.info("\n\n%s\n\n" % config)
model = PairwiseGMF(config)
sv = tf.train.Supervisor(logdir=None, save_model_secs=0, save_summaries_secs=0)
sess = sv.prepare_or_wait_for_session(config=tf.ConfigProto(
gpu_options=tf.GPUOptions(per_process_gpu_memory_fraction=0.1,
allow_growth=True)))
for i in range(FLAGS.iters):
if sv.should_stop():
break
progress = tqdm(enumerate(
dataset.get_data(FLAGS.batch_size, False, FLAGS.neg)),
dynamic_ncols=True,
total=(dataset.train_size * FLAGS.neg) // FLAGS.batch_size)
loss = []
for k, example in progress:
feed = {
model.input_users: example[:, 0],
model.input_items: example[:, 1],
model.input_items_negative: example[:, 2],
}
batch_loss, _ = sess.run([model.loss, model.train], feed)
loss.append(batch_loss)
progress.set_description(u"[{}] Loss: {:,.4f} » » » » ".format(
i, batch_loss))
print("Epoch {}: Avg Loss/Batch {:<20,.6f}".format(i, np.mean(loss)))
config.user_count = dataset.user_count
tf.logging.info("\n\n%s\n\n" % config)
model = PairwiseGMF(config)
sv = tf.train.Supervisor(logdir=None,
save_model_secs=0,
save_summaries_secs=0)
sess = sv.prepare_or_wait_for_session(
config=tf.ConfigProto(gpu_options=tf.GPUOptions(
per_process_gpu_memory_fraction=0.1,
allow_growth=True)))
for i in range(FLAGS.iters):
if sv.should_stop():
break
progress = tqdm(enumerate(dataset.get_data(FLAGS.batch_size, False, FLAGS.neg)),
dynamic_ncols=True, total=(dataset.train_size * FLAGS.neg) // FLAGS.batch_size)
loss = []
for k, example in progress:
feed = {
model.input_users: example[:, 0],
model.input_items: example[:, 1],
model.input_items_negative: example[:, 2],
model.input_positive_items_popularity: settings.Settings.normalized_popularity[example[:, 1]], # Added by LC
model.input_negative_items_popularity: settings.Settings.normalized_popularity[example[:, 2]] # Added by LC
}
if settings.Settings.loss_type == 2:
batch_loss, _, parameter_k = sess.run([model.loss, model.train, model.k], feed)
else:
parameter_k = 0
batch_loss, _ = sess.run([model.loss, model.train], feed)
sess.graph._unsafe_unfinalize()
tf.logging.info('Loading Pretrained Embeddings.... from %s' %
FLAGS.pretrain)
sess.run([
model.user_memory.embeddings.assign(pretrain['user'] * 0.5),
model.item_memory.embeddings.assign(pretrain['item'] * 0.5)
])
# Train Loop
for i in range(FLAGS.iters):
if sv.should_stop():
break
# 对应参数含义def get_data(self, batch_size, neighborhood, neg_count):
# tqdm是一个强大的终端进度条工具
progress = tqdm(enumerate(
dataset.get_data(FLAGS.batch_size, True, FLAGS.neg)),
dynamic_ncols=True,
total=(dataset.train_size * FLAGS.neg) // FLAGS.batch_size)
loss = []
for k, example in progress:
ratings, pos_neighborhoods, pos_neighborhood_length, \
neg_neighborhoods, neg_neighborhood_length = example
feed = {
# 所有行的第0列
model.input_users: ratings[:, 0],
# 所有行的第1列
model.input_items: ratings[:, 1],
model.input_items_negative: ratings[:, 2],
model.input_neighborhoods: pos_neighborhoods,
model.input_neighborhood_lengths: pos_neighborhood_length,
model.input_neighborhoods_negative: neg_neighborhoods,
sess.run([
model.user_memory.embeddings.assign(pretrain['user'] * 0.5),
model.item_memory.embeddings.assign(pretrain['item'] * 0.5)
])
else:
sess.graph._unsafe_unfinalize()
# Train Loop
results = []
for i in range(FLAGS.iters):
if sv.should_stop():
break
progress = tqdm(enumerate(
dataset.get_data(FLAGS.batch_size,
True,
FLAGS.neg,
use_popularity=use_popularity)),
dynamic_ncols=True,
total=(dataset.train_size * FLAGS.neg) // FLAGS.batch_size)
loss = []
for k, example in progress:
ratings, pos_neighborhoods, pos_neighborhood_length, neg_neighborhoods, neg_neighborhood_length = example
feed = {
model.input_users:
ratings[:, 0],
model.input_items:
ratings[:, 1],
model.input_positive_items_popularity:
settings.Settings.normalized_popularity[ratings[:,
1]], # Added by LC
model.item_memory.embeddings.assign(pretrain['item'] * 0.5)
])
print('Embeddings loaded!')
except:
print('Embeddings not found!')
sess.graph._unsafe_unfinalize()
else:
sess.graph._unsafe_unfinalize()
# Train Loop
results = []
for i in range(FLAGS.iters):
if sv.should_stop():
break
progress = tqdm(enumerate(dataset.get_data(FLAGS.batch_size, True, FLAGS.neg, use_popularity=use_popularity)),
dynamic_ncols=True, total=(dataset.train_size * FLAGS.neg) // FLAGS.batch_size)
loss = []
for k, example in progress:
ratings, pos_neighborhoods, pos_neighborhood_length, neg_neighborhoods, neg_neighborhood_length = example
# print('len(ratings[:, 0]):',len(ratings[:, 0]))
# print('len(ratings[:, 1]):', len(ratings[:, 1]))
# print('len(pos_neighborhoods):', len(pos_neighborhoods))
# print('len(neg_neighborhoods):', len(neg_neighborhoods))
feed = {
model.input_users: ratings[:, 0],
model.input_items: ratings[:, 1],
model.input_positive_items_popularity: settings.Settings.normalized_popularity[ratings[:, 1]], # Added by LC