def main_attention():
args = parse_args()
print_args(args)
set_seed(args.seed)
# load data
train_data, val_data, test_data, vocab = loader.load_dataset(args)
# initialize model
model = {}
model["G"], model["D"] = get_embedding(vocab, args)
model["clf"] = get_classifier(model["G"].ebd_dim, args)
best_path = '../bin/tmp-runs/16116280768954578/18'
model['G'].load_state_dict(torch.load(best_path + '.G'))
# model['D'].load_state_dict(torch.load(best_path + '.D'))
# model['clf'].load_state_dict(torch.load(best_path + '.clf'))
# if args.pretrain is not None:
# model["ebd"] = load_model_state_dict(model["G"], args.pretrain)
file_path = r'../data/attention_data.json'
Print_Attention(file_path, vocab, model, args)
def main():
args = parse_args()
print_args(args)
set_seed(args.seed)
# load data
train_data, val_data, test_data, class_names, vocab = loader.load_dataset(
args)
args.id2word = vocab.itos
# initialize model
model = {}
model["G"] = get_embedding(vocab, args) # model["G"]里面 是 词向量平均 + FC
criterion = ContrastiveLoss()
# model["G2"] = get_embedding_M2(vocab, args)
# model["clf"] = get_classifier(model["G"].hidden_size * 2, args)
if args.mode == "train":
# train model on train_data, early stopping based on val_data
optG = train(train_data, val_data, model, class_names, criterion,
args) # 使用孪生网络，来进行maml的方法，只改变FC
# val_acc, val_std, _ = test(val_data, model, args,
# args.val_episodes)
test_acc, test_std = test(test_data, class_names, optG, model, criterion,
args, args.test_epochs, True)
# path_drawn = args.path_drawn_data
# with open(path_drawn, 'w') as f_w:
# json.dump(drawn_data, f_w)
# print("store drawn data finished.")
# file_path = r'../data/attention_data.json'
# Print_Attention(file_path, vocab, model, args)
if args.result_path:
directory = args.result_path[:args.result_path.rfind("/")]
if not os.path.exists(directory):
os.mkdirs(directory)
result = {
"test_acc": test_acc,
"test_std": test_std,
# "val_acc": val_acc,
# "val_std": val_std
}
for attr, value in sorted(args.__dict__.items()):
result[attr] = value
with open(args.result_path, "wb") as f:
pickle.dump(result, f, pickle.HIGHEST_PROTOCOL)
def main():
# make_print_to_file(path='/results')
args = parse_args()
print_args(args)
set_seed(args.seed)
# load data
train_data, val_data, test_data, vocab = loader.load_dataset(args)
args.id2word = vocab.itos
# initialize model
model = {}
model["G"], model["D"] = get_embedding(vocab, args)
model["clf"] = get_classifier(model["G"].ebd_dim, args)
if args.mode == "train":
# train model on train_data, early stopping based on val_data
train(train_data, val_data, model, args)
# val_acc, val_std, _ = test(val_data, model, args,
# args.val_episodes)
test_acc, test_std, drawn_data = test(test_data, model, args,
args.test_episodes)
# path_drawn = args.path_drawn_data
# with open(path_drawn, 'w') as f_w:
# json.dump(drawn_data, f_w)
# print("store drawn data finished.")
# file_path = r'../data/attention_data.json'
# Print_Attention(file_path, vocab, model, args)
if args.result_path:
directory = args.result_path[:args.result_path.rfind("/")]
if not os.path.exists(directory):
os.mkdirs(directory)
result = {
"test_acc": test_acc,
"test_std": test_std,
# "val_acc": val_acc,
# "val_std": val_std
}
for attr, value in sorted(args.__dict__.items()):
result[attr] = value
with open(args.result_path, "wb") as f:
pickle.dump(result, f, pickle.HIGHEST_PROTOCOL)
def main():
args = parse_args()
# 可以打印到本地！存储下来
if args.path != "":
path = args.path
sys.stdout = open(path, "w")
print("test sys.stdout")
print_args(args)
set_seed(args.seed)
# load data
train_data, val_data, test_data, class_names, vocab = loader.load_dataset(
args)
args.id2word = vocab.itos
# initialize model
model = {}
model["G"] = get_embedding(vocab, args)
print(
"-------------------------------------param----------------------------------------------"
)
sum = 0
for name, param in model["G"].named_parameters():
num = 1
for size in param.shape:
num *= size
sum += num
print("{:30s} : {}".format(name, param.shape))
print("total param num {}".format(sum))
print(
"-------------------------------------param----------------------------------------------"
)
criterion = ContrastiveLoss()
# model["G2"] = get_embedding_M2(vocab, args)
# model["clf"] = get_classifier(model["G"].hidden_size * 2, args)
if args.mode == "train":
# train model on train_data, early stopping based on val_data
optG = train(train_data, val_data, test_data, model, class_names,
criterion, args)
# val_acc, val_std, _ = test(val_data, model, args,
# args.val_episodes)
test_acc, test_std = test(test_data, class_names, optG, model, criterion,
args, args.test_epochs, False)
print(
("[TEST] {}, {:s} {:s}{:>7.4f} ± {:>6.4f}, ").format(
datetime.datetime.now(),
colored("test ", "cyan"),
colored("acc:", "blue"),
test_acc,
test_std,
# colored("train stats", "cyan"),
# colored("G_grad:", "blue"), np.mean(np.array(grad['G'])),
# colored("clf_grad:", "blue"), np.mean(np.array(grad['clf'])),
),
flush=True)
# path_drawn = args.path_drawn_data
# with open(path_drawn, 'w') as f_w:
# json.dump(drawn_data, f_w)
# print("store drawn data finished.")
# file_path = r'../data/attention_data.json'
# Print_Attention(file_path, vocab, model, args)
if args.result_path:
directory = args.result_path[:args.result_path.rfind("/")]
if not os.path.exists(directory):
os.mkdirs(directory)
result = {
"test_acc": test_acc,
"test_std": test_std,
# "val_acc": val_acc,
# "val_std": val_std
}
for attr, value in sorted(args.__dict__.items()):
result[attr] = value
with open(args.result_path, "wb") as f:
pickle.dump(result, f, pickle.HIGHEST_PROTOCOL)