def training(device, w2v_model, train_iter, dev_iter, test_iter, batch_size,
num_epoch, lr, weight_decay, ckp, max_acc):
embedding_matrix = w2v_model.wv.vectors
input_size, hidden_size = embedding_matrix.shape[
0], embedding_matrix.shape[1]
loss_func = torch.nn.CrossEntropyLoss()
net = model.ESIM(input_size, hidden_size, 4, embedding_matrix).to(device)
#net.load_state_dict(torch.load(ckp, map_location='cpu'))
optimizer = torch.optim.Adam(net.parameters(),
lr=lr,
weight_decay=weight_decay)
for epoch in range(num_epoch):
net.train()
train_l, n = 0.0, 0
start = datetime.datetime.now()
out_epoch, label_epoch = [], []
for batch in train_iter:
seq1 = batch.sentence1
seq2 = batch.sentence2
label = batch.label
mask1 = (seq1 == 1)
mask2 = (seq2 == 1)
out = net(seq1.to(device), seq2.to(device), mask1.to(device),
mask2.to(device))
loss = loss_func(out, label.squeeze(-1))
optimizer.zero_grad()
loss.backward()
optimizer.step()
prediction = out.argmax(dim=1).data.cpu().numpy().tolist()
label = label.view(1, -1).squeeze().data.cpu().numpy().tolist()
out_epoch.extend(prediction)
label_epoch.extend(label)
train_l += loss.item()
n += 1
train_acc = accuracy_score(label_epoch, out_epoch)
dev_loss, dev_acc, max_acc = dev_evaluate(device, net, dev_iter,
max_acc, ckp)
test_loss, test_acc = test_evaluate(device, net, test_iter)
end = datetime.datetime.now()
print(
'epoch %d, train_acc %f, dev_acc %f, test_acc %f, max_acc %f, time %s'
% (epoch + 1, train_acc, dev_acc, test_acc, max_acc, end - start))
# 定义超参数
parser = argparse.ArgumentParser()
#本地
#parser.add_argument('--data_path', type=str, default="/Users/ren/Desktop/nlp相关/实验1/aclImdb/")#文件路径
parser.add_argument('--data_path', type=str, default="data/")#文件路径
parser.add_argument('--embed_size', type=int, default=300)#embeding层宽度
parser.add_argument('--hidden_size', type=int, default=128)
parser.add_argument('--seq_len', type=int, default=20)#文件长度，需要截断和填充
parser.add_argument('--batch_size', type=int, default=64)#批次
parser.add_argument('--bidirectional', type=bool, default=True)#是否开启双向
parser.add_argument('--classification_num', type=int, default=4)#分类个数
parser.add_argument('--lr', type=float, default=1e-3)#学习率
parser.add_argument('--dropout', type=float, default=0.5)#丢弃率
parser.add_argument('--num_epochs', type=int, default=100)#训练论数
parser.add_argument('--vocab_size', type=int, default=0)#vocab大小
parser.add_argument('--if_vail', type=bool, default=True)
parser.add_argument('--word2vec_path', type=str, default="/Users/ren/Desktop/nlp相关/glove_to_word2vec.txt")#预训练词向量路径
#parser.add_argument('--word2vec_path', type=str, default="/data/renhongjie/zouye1_new/data/glove_to_word2vec.txt")#预训练词向量路径
parser.add_argument('--save_path', type=str, default="best3.pth")#保存路径
parser.add_argument('--weight_decay', type=float, default=1e-4)#权重衰减
args = parser.parse_args()
if args.if_vail:
train_iter, test_iter,vail_iter,weight = utils.data(args)
else:
train_iter, test_iter, weight = utils.data(args)
net=model.ESIM(args,weight=weight)
if args.if_vail:
train.train(args, device, net,train_iter, test_iter,vail_iter)
else:
train.train(args,device,train_iter,test_iter,None)
args.vocab_size = len(text_field.vocab)
args.target_size = len(label_field.vocab)
args.weight_matrix = text_field.vocab.vectors
print(label_field.vocab.itos)
#print(label_field.vocab.itos)
print("\nParameters:")
for attr, value in sorted(args.__dict__.items()):
print("\t{}={}".format(attr.upper(), value))
train_iter = data.Iterator(dataset=train_data,
batch_size=args.batch_size,
shuffle=True)
valid_iter = data.Iterator(dataset=valid_data,
batch_size=args.batch_size,
shuffle=False)
test_iter = data.Iterator(dataset=test_data,
batch_size=args.batch_size,
shuffle=False)
if args.train is True:
print("Start training...")
esim = model.ESIM(args)
if args.cuda:
esim = esim.cuda()
train.train(train_iter, valid_iter, esim, args)
else:
print("\nStart predicting...")
esim = torch.load("./model/v6.0/model_{}.pkl".format(5)).cuda()
train.predict(test_iter, esim, args)
print("Finished.")
for idx, word in enumerate(w2v_model.wv.index2word)} # id -> word
feature_pad = 0
label2id = {'neutral': 0, 'entailment': 1, 'contradiction': 2, '-': 3}
id2label = {idx: word for idx, word in enumerate(label2id)}
test_feature1 = [[
word2id[word] if word in word2id else feature_pad for word in line
] for line in test_feature1_line]
test_feature2 = [[
word2id[word] if word in word2id else feature_pad for word in line
] for line in test_feature2_line]
embedding_matrix = w2v_model.wv.vectors
input_size, hidden_size = embedding_matrix.shape[0], embedding_matrix.shape[1]
loss_func = torch.nn.CrossEntropyLoss()
net = model.ESIM(input_size, hidden_size, 4, embedding_matrix).to(device)
sentence1_field = Field(sequential=True,
use_vocab=False,
batch_first=True,
fix_length=50,
pad_token=feature_pad)
sentence2_field = Field(sequential=True,
use_vocab=False,
batch_first=True,
fix_length=50,
pad_token=feature_pad)
fields = [('sentence1', sentence1_field), ('sentence2', sentence2_field)]
#获得测试集的Iterator
test_examples = []
for index in range(len(test_feature_line)):