zh_hidden = 1800
zh_dims = 712
input_dropout_p = 0.5
dropout_p = 0.5
enc_layers = 2
dec_layers = 2
en_max_len = 50
zh_max_len = 60
beam_size = 5
net = Seq2Seq(en_dims=en_dims,
zh_dims=zh_dims,
input_dropout_p=input_dropout_p,
dropout_p=dropout_p,
en_hidden=en_hidden,
zh_hidden=zh_hidden,
enc_layers=enc_layers,
dec_layers=dec_layers,
beam_size=beam_size,
en_max_len=en_max_len,
zh_max_len=zh_max_len)
pre_trained = torch.load(
'/data/xuwenshen/ai_challenge/code/fix_lens/models/' + args.model_path)
net.load_state_dict(pre_trained)
print(net)
net.eval()
test(test_loader=test_loader,
transform=transform,
net=net,
zh_voc_path = '/data/xuwenshen/ai_challenge/data/train/train/zh_voc.json'
transform = Transform(en_voc_path=en_voc_path, zh_voc_path=zh_voc_path)
en_dims = 800
en_voc = 50004
zh_dims = 800
zh_voc = 4004
en_hidden = 800
zh_hidden = 1000
atten_vec_size = 1200
net = Seq2Seq(en_dims=en_dims,
en_voc=en_voc,
zh_dims=zh_dims,
zh_voc=zh_voc,
dropout=1,
en_hidden=en_hidden,
zh_hidden=zh_hidden,
atten_vec_size=atten_vec_size,
entext_len=60)
pre_trained = torch.load(
'/data/xuwenshen/ai_challenge/code/bengio/models/ssprob-0.666313-loss-5.194733-score-0.339406-steps-41200-model.pkl'
)
net.load_state_dict(pre_trained)
print(net)
test(test_loader=test_loader,
transform=transform,
net=net,
batch_size=batch_size)
en_dims = 712
en_voc = 50004
zh_dims = 712
zh_voc = 4004
dropout = 0.5
en_hidden = 800
zh_hidden = 712
channels = 128
kernel_size = 3
net = Seq2Seq(en_dims=en_dims,
en_voc=en_voc,
zh_dims=zh_dims,
zh_voc=zh_voc,
dropout=dropout,
en_hidden=en_hidden,
zh_hidden=zh_hidden,
channels=channels,
kernel_size=kernel_size,
entext_len=60)
#pre_trained = torch.load('./models/ssprob-1.000000-loss-6.934007-steps-134044 model.pkl')
#net.load_state_dict(pre_trained)
print(net)
epoch = 10000
lr = 0.001
hyperparameters = {
'learning rate': lr,
'batch size': batch_size,
en_dims = 256
zh_dims = 256
dropout = 0.5
en_hidden = 256
zh_hidden = 400
atten_vec_size = 712
channels = 1024
kernel_size = 1
net = Seq2Seq(en_dims=en_dims,
zh_dims=zh_dims,
dropout=dropout,
en_hidden=en_hidden,
zh_hidden=zh_hidden,
atten_vec_size=atten_vec_size,
channels=channels,
kernel_size=kernel_size,
entext_len=60)
pre_trained = torch.load('./models/ssprob-0.777249-loss-5.034320-score-0.366200-steps-100400-model.pkl')
net.load_state_dict(pre_trained)
print (net)
epoch = 10000
lr = 0.001
hyperparameters = {
'learning rate': lr,
'batch size': batch_size,