def train(epochs=10, batchSize=1024, lr=0.001, dim=128, eva_per_epochs=1):
#读取数据
train, test, allItems = dp.getTrainAndTestSeqs(fp.Ml_latest_small.SEQS)
#初始化模型
net = RNN_ALS_rec(max(allItems) + 1, dim)
#定义损失函数
criterion = torch.nn.BCELoss()
#初始化优化器
optimizer = torch.optim.AdamW(net.parameters(), lr=lr, weight_decay=5e-3)
#开始训练
for e in range(epochs):
all_lose = 0
for seq in DataLoader(
train,
batch_size=batchSize,
shuffle=True,
):
x = torch.LongTensor(seq[:, :-2].detach().numpy())
item = torch.LongTensor(seq[:, -2].detach().numpy())
y = torch.FloatTensor(seq[:, -1].detach().numpy())
optimizer.zero_grad()
logits = net(x, item)
loss = criterion(logits, y)
all_lose += loss
loss.backward()
optimizer.step()
print('epoch {},avg_loss={:.4f}'.format(
e, all_lose / (len(train) // batchSize)))
#评估模型
if e % eva_per_epochs == 0:
p, r, acc = doEva(net, train)
print('train:p:{:.4f}, r:{:.4f}, acc:{:.4f}'.format(p, r, acc))
p, r, acc = doEva(net, test)
print('test:p:{:.4f}, r:{:.4f}, acc:{:.4f}'.format(p, r, acc))
def train(epochs=10, batchSize=1024, lr=0.001, dim=128, eva_per_epochs=1):
#读取数据
train, test, allItems = dp.getTrainAndTestSeqs(fp.Ml_latest_small.SEQS)
all_seq_lens = len(train[0]) - 1
#初始化模型
net = Transformer4Rec(max(allItems) + 1, all_seq_lens, dim)
#初始化优化器
optimizer = torch.optim.AdamW(net.parameters(), lr=lr, weight_decay=5e-3)
#开始训练
for e in range(epochs):
all_lose = 0
for seq in tqdm(DataLoader(train, batch_size=batchSize, shuffle=True)):
x = torch.LongTensor(seq[:, :-2].detach().numpy())
history_seqs = torch.LongTensor(seq[:, 1:-1].detach().numpy())
target_item = torch.LongTensor(seq[:, -2].detach().numpy())
target_label = torch.FloatTensor(seq[:, -1].detach().numpy())
optimizer.zero_grad()
loss = net(x, history_seqs, target_item, target_label)
all_lose += loss
loss.backward()
optimizer.step()
print('epoch {},avg_loss={:.4f}'.format(
e, all_lose / (len(train) // batchSize)))
#评估模型
if e % eva_per_epochs == 0:
p, r, acc = doEva(net, train)
print('train:p:{:.4f}, r:{:.4f}, acc:{:.4f}'.format(p, r, acc))
p, r, acc = doEva(net, test)
print('test:p:{:.4f}, r:{:.4f}, acc:{:.4f}'.format(p, r, acc))