owner,
usernum,
itemnum,
batch_size=args.batch_size,
n_workers=1)
sess = tf.Session(config=tf.ConfigProto(allow_soft_placement=True))
model = Model(usernum, itemnum, args)
sess.run(tf.initialize_all_variables())
best_valid_auc = 0.5
best_iter = 0
for i in range(args.maximum_epochs):
for _ in range(oneiteration / args.batch_size):
batch = sampler.next_train_batch()
batch_u, batch_i, batch_j, batch_oi, batch_oj = batch
_, train_loss, train_auc = sess.run(
(model.gds, model.loss, model.auc), {
model.batch_u: batch_u,
model.batch_i: batch_i,
model.batch_j: batch_j,
model.batch_oi: batch_oi,
model.batch_oj: batch_oj
})
print train_loss, train_auc
if i % 10 == 0:
f.write('#iter %d: loss %f, train auc %f \n' %
(i, train_loss, train_auc))
_valid_auc = 0.0
class LossHistory(keras.callbacks.Callback):
def on_train_begin(self, logs={}):
self.losses = 0.0
def on_batch_end(self, batch, logs={}):
self.losses += logs.get('loss')
for i in range(args.maximum_epochs):
history = LossHistory()
loss = 0.0
for _ in range(int(oneiteration / args.batch_size)):
batch_u, batch_i, batch_j, batch_oi, batch_oj = sampler.next_train_batch(
)
if (args.alg == 'posneg'):
labels = [0] * len(batch_u)
margin = np.full(len(batch_u), 0.2)
hist = model.fit([
np.array(batch_u),
np.array(batch_i),
np.array(batch_j),
np.array(batch_oi),
np.array(batch_oj), margin
],
labels,
batch_size=args.batch_size,
verbose=0,