def pred(dev_iter, net, device=None):
if device is None:
device = list(net.parameters())[0].device
label = []
label_true = []
net = net.to(device)
with torch.no_grad():
net.eval()
for X, Y, z in dev_iter:
# print(torch.argmax(net(X.)), dim=1)
# break
# print(torch.argmax(net(X.to(device)),dim=1))
# print(net(X.to(device)))
# break
label.extend(
torch.argmax(net(X.to(device)), dim=1).cpu().numpy().tolist())
label_true.extend(Y.numpy().tolist())
net.train()
return f1(label, label_true, classifications=2)
load_pretrained_embedding(vocab.itos, glove_vocab))
net.embedding.weight.requires_grad = True # 直接加载预训练好的, 所以不需要更新它
#####
num_epochs = 10
optimizer = torch.optim.Adam(net.parameters())
loss = torch.nn.CrossEntropyLoss() # softmax,交叉熵
train(train_iter, net, loss, optimizer, device, num_epochs)
# 测试
label = []
label_true = []
net = net.to(device)
for X, Y, Z in dev_iter:
# print(torch.argmax(net(X.)), dim=1)
# break
#print(torch.argmax(net(X.to(device)),dim=1))
#print(net(X.to(device)))
#break
label.extend(
torch.argmax(net(X.to(device), Z), dim=1).cpu().numpy().tolist())
label_true.extend(Y.numpy().tolist())
k = 0
for i, j in zip(label, label_true):
if i == j:
k = k + 1
print(k / 1067)
# 评价函数 evaluate model
from eval.score import f1
print('f1_score is :{}'.format(f1(label_true, label, 2)))