def train(model, criterion, optimizer, scheduler, epoch, batch_size, use_cuda,
train_ratings, negatives, n_items, n_neg):
model.train()
train_dataset = get_train_instances(train_ratings, negatives, n_items,
n_neg)
train_loader = DataLoader(dataset=train_dataset,
batch_size=batch_size,
num_workers=4,
shuffle=True)
train_steps = (len(train_loader.dataset) // train_loader.batch_size) + 1
running_loss = 0
for data in train_loader:
users = data[:, 0]
items = data[:, 1]
labels = data[:, 2].float()
if use_cuda:
users, items, labels = users.cuda(), items.cuda(), labels.cuda()
optimizer.zero_grad()
preds = model(users, items)
if scheduler:
scheduler.step()
loss = criterion(preds.squeeze(1), labels)
loss.backward()
optimizer.step()
running_loss += loss.item()
return running_loss / train_steps
model.compile(optimizer=optimizers.SGD(lr=configs.learning_rate),
loss='binary_crossentropy')
"""
# --------------Init performance--------------
t1 = time()
(hits, ndcgs) = evaluate_model(model, testRatings, testNegatives, configs.topK)
hr, ndcg = np.array(hits).mean(), np.array(ndcgs).mean()
print('Init: HR = %.4f, NDCG = %.4f [%.1f s]' % (hr, ndcg, time() - t1))
"""
# --------------Train model-----------------
# best_hr, best_ndcg, best_iter = hr, ndcg, -1
for epoch in range(configs.epochs):
t1 = time()
# Generate training instances
user_input, item_input, labels = get_train_instances(
train, configs.num_negatives)
print('Get_train_instances [%.1f s]' % (time() - t1))
t1 = time()
# --------------Fit model----------------
hist = model.fit(
[np.array(user_input), np.array(item_input)],
np.array(labels),
batch_size=configs.batch_size,
epochs=1,
verbose=configs.verbose,
shuffle=True)
t2 = time()
"""
# --------------Evaluation--------------
if epoch % configs.verbose == 0:
(hits, ndcgs) = evaluate_model(model, testRatings, testNegatives, configs.topK)