def post(self, firmid, clsfid):
clsf_key = ndb.Key( 'Firm', firmid, 'Classification', clsfid )
clsf = clsf_key.get()
if clsf:
self.error(409)
else:
d = json.loads( self.request.body )
clsf = Classification( key=clsf_key, **d )
clsf.put()
def post(self, firmid, clsfid):
clsf_key = ndb.Key('Firm', firmid, 'Classification', clsfid)
clsf = clsf_key.get()
if clsf:
self.error(409)
else:
d = json.loads(self.request.body)
clsf = Classification(key=clsf_key, **d)
clsf.put()
def get(self, firmid):
self.json_content()
firm_key = ndb.Key('Firm', firmid)
query = Classification.query(ancestor=firm_key)
classifications = query.map(
lambda classification: classification.to_dict())
self.w(json.dumps(classifications))
def training():
context_x = Context(inputs_dict.n_words, hidden_size).to(device)
classification_x = Classification().to(device)
context_x, classification_x, plot_losses = trainIters(context_x,
classification_x,
device,
inputs_dict,
target_dict,
pairs,
n_iters,
print_every=50)
return context_x, classification_x, plot_losses
def preview(self):
''' Ajax请求，无论同步或者异步，都不能正确跳转；通过form.submit()实现
article = request.get_json(force=True)
return self.render('article.html', article=article, mode='preview')
'''
classification = Classification.objects(
id=request.form.get('classification')).first()
# TODO 异常处理
article = {
'title': request.form.get('title'),
'content': request.form.get('content'),
'abstract': request.form.get('abstract'),
'classification': classification,
}
return self.render('article.html', article=article, mode='preview')
def main():
#define the dataloader
train_loader = DataLoader(SkeletonFeeder(mode='train', debug=False), batch_size=params['batchsize'], shuffle=True, num_workers=params['numworkers'])
val_loader = DataLoader(SkeletonFeeder(mode='valid', debug=False), batch_size=params['batchsize'], shuffle=False, num_workers=params['numworkers'])
n_class = params['n_class']
cur_time = time.strftime('%Y-%m-%d-%H-%M-%S', time.localtime(time.time()))
#use the model and transfer to gpu
model = Classification(n_class=n_class) #add model cfg
model = model.cuda(params['gpu'][0])#to do--->
model = nn.DataParallel(model, device_ids=params['gpu'])
if params['retrain']:
trained_dict = torch.load(params['retrain'],map_location='cpu')
model_dict = model.state_dict()
trained_dict = {k:v for k,v in trained_dict.items() if k in model_dict}
model_dict.update(trained_dict)
model.load_state_dict(model_dict)
print('load trained model finish')
model_params = filter(lambda p: p.requires_grad, model.parameters())
optimizer = optim.Adam(model_params, lr=params['lr'], weight_decay=params['weight_decay'])
schedule = optim.lr_scheduler.ReduceLROnPlateau(optimizer, 'min', factor=0.333, patience=2, verbose=True)
writer = SummaryWriter()
criterion = nn.CrossEntropyLoss()
min_loss = 1000
print('-------------------start training----------------------')
print('lr:', optimizer.param_groups[0]['lr'])
for i in range(params['epoch']):
train_loss, train_top1, train_top5, batch_time, data_time = train(model, train_loader, optimizer,criterion)
valid_loss, val_top1, val_top5 = valid(model, val_loader,optimizer,criterion)
schedule.step(valid_loss)
f = open(params['log']+'bert_classifylog_'+cur_time+'.txt', 'a')
print('epoch:', str(i + 1) + "/" + str(params['epoch']))
print('data time:%0.3f'%data_time.avg, 'batch time:%0.3f'%batch_time.avg, 'epoch time:%0.3f'%(batch_time.sum))
print('train loss:%0.8f'%train_loss, 'top1:%0.2f'%train_top1, '%', 'top5:%0.2f'%train_top5, '%', 'lr:', optimizer.param_groups[0]['lr'])
print('valid loss:%0.8f'%valid_loss, 'top1:%0.2f'%val_top1, '%', 'top5:%0.2f'%val_top5, '%')
f.write('epoch:'+str(i+1)+"/"+str(params['epoch'])+'\n')
f.write('data time:%0.3f'%data_time.avg+'batch time:%0.3f'%batch_time.avg+'epoch time:%0.3f'%(batch_time.sum)+'\n')
f.write('train loss:%0.8f'%train_loss+'top1:%0.2f'%train_top1+'%'+'top5:%0.2f'%train_top5+'%'+'lr:'+str(optimizer.param_groups[0]['lr'])+'\n')
f.write('valid loss:%0.8f'%valid_loss+'top1:%0.2f'%val_top1+'%'+'top5:%0.2f'%val_top5+'%'+'\n')
f.write('************************************\n')
f.close()
writer.add_scalar('train loss', train_loss, i)
writer.add_scalar('valid loss', valid_loss, i)
writer.add_scalar('train top1', train_top1, i)
writer.add_scalar('valid top1', val_top1, i)
writer.add_scalar('train top5', train_top5, i)
writer.add_scalar('valid top5', val_top5, i)
if valid_loss < min_loss:
torch.save(model.state_dict(), params['save_path']+'bert_classifymodel_'+cur_time+'.pth')
print('saving model successful to --->',params['save_path'])
min_loss = valid_loss
writer.close()
image_row_size = image_size[0] * image_size[1]
image_size = (100, 100)
image_row_size = image_size[0] * image_size[1]
mean = [0.485, 0.456, 0.406]
std = [0.229, 0.224, 0.225]
transform = transforms.Compose([
transforms.Resize(image_size),
# transforms.Grayscale(),
transforms.ToTensor(),
transforms.Normalize(mean, std)
])
path = '/home/aims/Documents/Pytorch/pytorch_exercise/data'
net = Classification()
criterion = nn.CrossEntropyLoss()
optimizer = optim.SGD(net.parameters(), lr=0.001, momentum=0.9)
train_data = CatDogDataset(path + "/" + 'train', transform=transform)
test_data = CatDogDataset(path + "/" + 'val', transform=transform)
trainloader = torch.utils.data.DataLoader(test_data,
batch_size=64,
shuffle=True,
num_workers=4)
testloader = torch.utils.data.DataLoader(test_data,
batch_size=64,
shuffle=True,
num_workers=4)
#Training
def get(self, firmid):
self.json_content()
firm_key = ndb.Key('Firm', firmid)
query = Classification.query(ancestor=firm_key)
classifications = query.map(lambda classification: classification.to_dict())
self.w(json.dumps(classifications))
normed[i, 0:data_length_array[i]] = (
data_steps_array[i, 0:data_length_array[i]] -
data_steps_array[i, 0:data_length_array[i]].mean(axis=0)
) / data_steps_array[i, 0:data_length_array[i]].std(axis=0)
# Shuffle data
normed, data_labels_array, data_length_array = shuffle(normed,
data_labels_array,
data_length_array,
random_state=47)
data = tf.placeholder(tf.float32, [None, normed.shape[1], normed.shape[2]])
target = tf.placeholder(tf.float32, [None, num_classes])
length = tf.placeholder(tf.float32, [None])
learning_rate = tf.placeholder(tf.float32, shape=[])
model = Classification(data, target, length, learning_rate, num_RNN,
num_FCN)
# Save only one checkpoint
saver = tf.train.Saver(max_to_keep=1)
all_error = []
best_error = {'epoch': [], 'best_acc': []}
train_index = []
test_index = []
for train_ind, test_ind in KFold(n_split, random_state=47).split(normed):
train_index.append(train_ind)
test_index.append(test_ind)
arg_index = int(sys.argv[1])
sess = tf.Session()
from pytorch_lightning.callbacks import ModelCheckpoint
from dataset import ArgsBase, NSMCDataModule, KorSTSDataModule
from model import SubtaskGPT2, SubtaskGPT2Regression, Classification
parser = argparse.ArgumentParser(description='Train KoGPT2 subtask model')
parser.add_argument('--task', type=str, default=None, help='subtask name')
parser.add_argument('--do_test',
action='store_true',
help='evaluate on test set')
parser.add_argument('--checkpoint_path', type=str, default=None)
if __name__ == '__main__':
parser = ArgsBase.add_model_specific_args(parser)
parser = Classification.add_model_specific_args(parser)
parser = NSMCDataModule.add_model_specific_args(parser)
parser = KorSTSDataModule.add_model_specific_args(parser)
parser = Trainer.add_argparse_args(parser)
args = parser.parse_args()
logging.getLogger().setLevel(logging.INFO)
logging.info(args)
torch.manual_seed(args.seed)
torch.cuda.manual_seed(args.seed)
np.random.seed(args.seed)
if args.task.lower() == 'nsmc':
dm = NSMCDataModule(args.train_data_path,
parse.add_argument("--log", type=int, default=20, help="")
parse.add_argument("--no_cuda", action="store_true", default=False, help="")
args = parse.parse_args()
#####数据处理
normalize = transforms.Compose([
transforms.Resize((64, 64)),
transforms.ToTensor(),
transforms.Normalize([0.5, 0.5, 0.5], [0.5, 0.5, 0.5]),
])
##判断gpu是否可用并生成随机种子
args.cuda = not args.no_cuda and torch.cuda.is_available()
torch.manual_seed(args.seed)
if args.cuda:
torch.cuda.manual_seed(args.seed)
model = Classification(2)
if args.cuda:
model.cuda()
cudnn.benchmark = True
batch_size = args.batch
##构建数据迭代器 训练
dataset = MyData(args.datas, transforms=normalize)
valid = MyData(args.test, transforms=normalize)
dataloader = DataLoader(dataset,
shuffle=True,
batch_size=batch_size,
num_workers=2,
collate_fn=collate_fn)
valid_dataloader = DataLoader(valid,
shuffle=True,
torch.manual_seed(1)
if args.cuda:
torch.cuda.manual_seed(1)
cudnn.benchmark = True
transform = transforms.Compose([
transforms.Resize((64, 64)),
transforms.ToTensor(),
transforms.Normalize((0.5, 0.5, 0.5), (0.5, 0.5, 0.5)),
])
dataset = MyData(args.data, transform)
dataloader = DataLoader(dataset,
batch_size=args.batch,
shuffle=False,
num_workers=4,
collate_fn=collate_fn)
for f in os.listdir(args.models):
model = Classification(2, False)
model.load_state_dict(torch.load(args.models + f)['models'])
model.eval()
if args.cuda:
model.cuda()
cnt = 0.0
for i, (image, label) in enumerate(dataloader):
image = Variable(image.cuda(), requires_grad=False)
output = model(image, None)
output = output.cpu().detach().numpy()
index = np.argmax(output, 1)
cnt += sum(np.array(index, dtype=np.int) == label.detach().numpy())
print(f, "acc:{:.4f}".format(cnt / len(dataset)))
