def main():
config = configuration.Configuration.load("config.json")
if config.logging_file_name is not None:
logging.basicConfig(level=logging.INFO,
format="%(message)s",
filename=config.logging_file_name,
filemode="w")
else:
logging.basicConfig(level=logging.INFO, format="%(message)s")
vocab = Vocabulary.load_vocabulary("./data/weibo_vocab_word.pkl")
config.common["num_word"] = vocab.truncated_length
frame = inspect.currentframe()
gpu_tracker = MemTracker(frame)
gpu_tracker.track()
model = seq2seqVAD.Seq2SeqVAD(config)
gpu_tracker.track()
with open("./data/weibo_embedding_word.pkl", "rb") as f:
embedding = pickle.load(f)
model.init_weight(embedding)
logging.info(repr(config))
logging.info(repr(model))
batch_size = config.learning["batch_size"]
train_batch_generator = dataIter.BatchIter("data/weibo_train_word.pkl",
batch_size)
valid_batch_generator = dataIter.BatchIter("data/weibo_test_word.pkl",
batch_size)
test_batch_generator = dataIter.BatchIter("data/weibo_test_word.pkl", 10)
train_report_names = [
"ce", "bwd_rnn_kld", "aux_bow", "bwd_ce", "loss", "KLD_weight",
"aux_weight", "bwd_ce_weight"
]
train_report = report.Report(train_report_names)
valid_report_names = ["ce"]
generation_report_names = [
"emb_avg", "emb_ext", "emb_gre", "dist-1", "dist-2", "novel"
]
generation_report = report.Report(generation_report_names)
valid_report = report.Report(valid_report_names)
parent_name = config.learning["parent_name"]
if not exists(parent_name):
os.mkdir(parent_name)
os.mkdir(join(parent_name, "report"))
os.mkdir(join(parent_name, "stochastic_array"))
os.mkdir(join(parent_name, "saved_models"))
os.mkdir(join(parent_name, "generated_dialogues"))
train.train_model(model,
config,
train_batch_generator=train_batch_generator,
valid_batch_generator=valid_batch_generator,
test_batch_generator=test_batch_generator,
vocab=vocab,
embedding=embedding,
report_train=train_report,
report_valid=valid_report,
report_generation=generation_report,
parent_file_name=parent_name)
data_train =Positive_Roi_Dataset('E:/ali_cervical_carcinoma_data',train=True)
data_test =Positive_Roi_Dataset('E:/ali_cervical_carcinoma_data',train=False)
# print('data_test num=', len(data_test), '\nfileds:\n', data_test[0][1])
trainLoader = data.DataLoader(data_train, batch_size=1, shuffle=True, collate_fn=utils.collate_fn)
testLoader = data.DataLoader(data_test, batch_size=1, shuffle=False, collate_fn=utils.collate_fn)
torch.cuda.empty_cache()
#观察GPU
device = torch.device('cuda')
frame = inspect.currentframe() # define a frame to track
gpu_tracker = MemTracker(frame) # define a GPU tracker
gpu_tracker.track() # run function between the code line where uses GPU
model.to(device)
# gpu_tracker.track() # run function between the code line where uses GPU
params = [p for p in model.parameters() if p.requires_grad]
optimizer = torch.optim.SGD(params, lr=0.001,
momentum=0.9, weight_decay=0.0005)
lr_scheduler = torch.optim.lr_scheduler.StepLR(optimizer,
step_size=3,
gamma=0.1)
gpu_tracker.track()
for epoch in range(1, 10):
AlexNet_layers.append(nn.Linear(4096, 4096))
AlexNet_layers.append(nn.ReLU())
AlexNet_layers.append(nn.Dropout(0.5))
AlexNet_layers.append(nn.Linear(4096, 1000))
Stash_List = [3,6,8,10,14,17,20]
for i in AlexNet_layers:
if i != view:
i = i.cuda()
x = inp.cuda()
frame = inspect.currentframe() # define a frame to track
gpu_tracker = MemTracker(frame) # define a GPU tracker
gpu_tracker.track()
# forward()
#import time
for j in Stash_List:
print("Stash first %s Layers:"%(Stash_List.index(j)))
x = inp.cuda()
#begin = time.time()
for i in range(len(AlexNet_layers)):
# y = x
if i < j :
stash = True
else:
stash = False
x = checkpoint(tuple(AlexNet_layers[:i + 1]), stash, x)