X_test = np.array(X_test)
y_test = np.array(y_test)
mask_test = np.array(mask_test)
y_test = to_categorical(y_test, 2)
# 对数据进行随机打乱
permutation = np.random.permutation(X_test.shape[0])
X_test = X_test[permutation, :, :, :]
y_test = y_test[permutation, :]
mask_test = mask_test[permutation, :, :]
testData_sum = len(X_test) # 训练样本总数
print("--------------加载测试数据完成------------")
use_cuda = torch.cuda.is_available() # 环境中是否有GPU
net = CNN_LSTM.CNN_LSTM(cfg.ts)
if use_cuda and cfg.gpu:
net = net.cuda()
net = nn.DataParallel(net)
if cfg.re:
print('------------------------------')
print('==> 加载checkpoint ')
if not os.path.exists(cfg.ckpt):
raise AssertionError['找不到路径']
checkpoint = torch.load(cfg.ckpt)
net.load_state_dict(checkpoint['net'])
best_test_acc = checkpoint['best_test_acc']
print('best_test_acc is %.4f%%' % best_test_acc)
best_test_acc_epoch = checkpoint['best_test_acc_epoch']
def main(config):
device = torch.device('cuda' if torch.cuda.is_available() else 'cpu')
train_dataset = CNN_LSTM_Triplet_Dataset(config)
print(len(train_dataset))
train_sampler = RandomSampler(train_dataset)
train_loader = DataLoader(train_dataset,
batch_size=config["model"]["batch_size"],
sampler=train_sampler)
model = CNN_LSTM(config).double().to(device=device)
# if True:
# model.load_state_dict(torch.load("./best_model_eval/model_acc_83.3276.pth"))
# optimizer = torch.optim.Adam(lr=1e-5, betas=(0.9, 0.98), eps=1e-9)
param_optimizer = list(model.named_parameters())
no_decay = ['bias', 'gamma', 'beta']
optimizer_grouped_parameters = [{
'params':
[p for n, p in param_optimizer if not any(nd in n for nd in no_decay)],
'weight_decay_rate':
0.01
}, {
'params':
[p for n, p in param_optimizer if any(nd in n for nd in no_decay)],
'weight_decay_rate':
0.0
}]
optimizer = torch.optim.Adam(lr=1e-2,
betas=(0.9, 0.98),
eps=1e-9,
params=optimizer_grouped_parameters)
loss_fn = nn.TripletMarginLoss()
acc_dev_previous = 0
loss_dev_previous = 1000
loss_train = 1000
for epoch in range(1, config["model"]["epoch"]):
print(f"Training epoch {str(epoch)}")
loss_train = train(train_loader, model, optimizer, loss_fn, loss_train,
epoch, device)
print(f"Evaluate model.............")
acc_dev, loss_dev = evaluate(dev_loader, model, loss_fn, epoch,
len(dev_dataset), device)
print(f"Accuracy score: {acc_dev:.4f} at epoch {epoch}")
print(f"Loss Dev: {loss_dev:.4f} at epoch {epoch}")
print("=" * 15, f"END EPOCH {epoch}", "=" * 15)
if acc_dev > acc_dev_previous:
acc_dev_previous = acc_dev
torch.save(model.state_dict(),
f"./best_model_eval/model_acc_{round(acc_dev, 4)}.pth")
if loss_dev < loss_dev_previous:
loss_dev_previous = loss_dev
torch.save(
model.state_dict(),
f"./best_model_eval/model_loss_{round(int(loss_dev), 4)}.pth")
testRoot, dl, 160, ts, lb)
X_test = np.array(X_test)
y_test = np.array(y_test)
mask_test = np.array(mask_test)
y_test = to_categorical(y_test, 22)
# 对数据进行随机打乱
permutation = np.random.permutation(X_test.shape[0])
X_test = X_test[permutation, :, :, :]
y_test = y_test[permutation, :]
mask_test = mask_test[permutation, :, :]
testData_sum = len(X_test) # 训练样本总数
print("--------------加载测试数据完成------------")
inputs = torch.Tensor(X_test)
target = torch.Tensor(y_test)
mask = torch.Tensor(mask_test)
net = CNN_LSTM.CNN_LSTM(ts)
checkpoint = torch.load("./checkpoints/res1_52/CNN-LSTM-82.t7")
net.load_state_dict(checkpoint['net'])
x2_atten, atten2, x3_atten, atten3, x1, x2, x3, x4, x = net(inputs, mask)
print(x2_atten.size())
# Y = tsne(X, 2, 50, 20.0)
# pylab.scatter(Y[:, 0], Y[:, 1], 20, labels)
# pylab.show()
train_iter, dev_iter, test_iter = load_dataset(text_field, label_field, args, device=-1, repeat=False, shuffle=True)#数据载入
args.VOCAB_SIZE = len(text_field.vocab)#文章中的表达词汇数目
args.OUTPUT_SIZE = args.HIDDEN_DIM
args.BEST_MODEL = "best_" + str(args.EMBEDDING_DIM) + ".pt"
args.KERNEL_SIZES = [int(k) for k in args.KERNEL_SIZES.split(',')]
#创建vocab字典
index2class = label_field.vocab.itos[1:]
index2class = list(map(int, index2class))
print('Parameters:')
for attr, value in sorted(args.__dict__.items()):
if attr in {'vectors'}:
continue
print('\t{}={}'.format(attr.upper(), value))
model = CNN_LSTM(index2class, args)
model_name = train_model(model, train_iter, dev_iter, args)
print('\n'+model_name)
if args.DEVICE == 'cuda':
model.cuda()
model.load_state_dict(torch.load(model_name, map_location='cpu'))
test_acc = test_model(model,test_iter,args, index2class)
print(f'Test Acc: {test_acc:.2f}%')