파이토치

import torch

import torchvision.datasets as dsets

import torchvision.transforms as transforms

import torch.nn.init

device = 'cuda' if torch.cuda.is_available() else 'cpu'

# 랜덤 시드 고정

torch.manual_seed(777)

# GPU 사용 가능일 경우 랜덤 시드 고정

if device == 'cuda':

    torch.cuda.manual_seed_all(777)

#텐서에는 없음

learning_rate = 0.001 #옵티마이저에서 자동으로 제공됨, 변경가능함 

training_epochs = 15 

batch_size = 100 #옵티마이저에서 자동으로 32로 세팅됨. 변경은 가능함. 

# 토치 transform

mnist_train = dsets.MNIST(root='MNIST_data/', # 다운로드 경로 지정

                          train=True, # True를 지정하면 훈련 데이터로 다운로드

                          transform=transforms.ToTensor(), # 텐서로 변환

                          download=True)

mnist_test = dsets.MNIST(root='MNIST_data/', # 다운로드 경로 지정

                         train=False, # False를 지정하면 테스트 데이터로 다운로드

                         transform=transforms.ToTensor(), # 텐서로 변환

                         download=True)

data_loader = torch.utils.data.DataLoader(dataset=mnist_train,

                                          batch_size=batch_size,

                                          shuffle=True,

                                          drop_last=True)

first_batch=data_loader.__iter__().__next__()

x,y = first_batch[0],first_batch[1] 

 

cnt=0

plt.figure(figsize=(30,10)) 

for row in range(1,6):

  for col in range(1,21):

    plt.subplot(5,20,cnt+1)

    plt.imshow(x[cnt][0],cmap='gray')

    plt.title(y[cnt].numpy())

    cnt+=1