15.데이터로더

# 1. 손글씨 인식 모델 만들기

import torch

import torch.nn as nn

import torch.optim as optim

import torchvision.transforms as transforms

import matplotlib.pyplot as plt

from sklearn.datasets import load_digits

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

print(device)

결과 : 

cuda

------------------------------------------------

digits = load_digits()

x_data = digits['data']

y_data = digits['target']

print(x_data.shape)

print(y_data.shape)

결과 :

(1797, 64)
(1797,)

------------------------------------------------

 import matplotlib.pyplot as plt

fig, axes = plt.subplots(nrows=2, ncols=5, figsize=(14,8))

for i, ax in enumerate(axes.flatten()):

  ax.imshow(x_data[i].reshape((8,8)), cmap='gray')

  ax.set_title(y_data[i])

  ax.axis('off')

 

결과 :

------------------------------------------------

x_data = torch.FloatTensor(x_data)

y_data = torch.LongTensor(y_data)

 

print(x_data.shape)

print(y_data.shape)

결과 :

torch.Size([1797, 64])
torch.Size([1797])

------------------------------------------------

y_one_hot = nn.functional.one_hot(y_data, num_classes=10).float()

y_one_hot[:10]

결과 :

tensor([[1., 0., 0., 0., 0., 0., 0., 0., 0., 0.],
        [0., 1., 0., 0., 0., 0., 0., 0., 0., 0.],
        [0., 0., 1., 0., 0., 0., 0., 0., 0., 0.],
        [0., 0., 0., 1., 0., 0., 0., 0., 0., 0.],
        [0., 0., 0., 0., 1., 0., 0., 0., 0., 0.],
        [0., 0., 0., 0., 0., 1., 0., 0., 0., 0.],
        [0., 0., 0., 0., 0., 0., 1., 0., 0., 0.],
        [0., 0., 0., 0., 0., 0., 0., 1., 0., 0.],
        [0., 0., 0., 0., 0., 0., 0., 0., 1., 0.],
        [0., 0., 0., 0., 0., 0., 0., 0., 0., 1.]])

------------------------------------------------

from sklearn.model_selection import train_test_split

x_train, x_test, y_train, y_test = train_test_split(x_data, y_one_hot, test_size=0.2, random_state=10)

print(x_train.shape, y_train.shape)

print(x_test.shape, y_test.shape)

결과 :

torch.Size([1437, 64]) torch.Size([1437, 10])
torch.Size([360, 64]) torch.Size([360, 10])

------------------------------------------------

# 2. 데이터 로더

* 데이터의 양이 많을 때 배치 단위로 학습하는 방법

 

loader = torch.utils.data.DataLoader(

    dataset=list(zip(x_train, y_train)),

    batch_size=64,

    shuffle=True

)

imgs, labels = next(iter(loader))

fig, axes = plt.subplots(nrows=8, ncols=8, figsize=(14,14))

for ax, img, label in zip(axes.flatten(), imgs, labels):

  ax.imshow(img.reshape((8,8)), cmap='gray')

  ax.set_title(str(torch.argmax(label)))

  ax.axis('off')

결과 :

------------------------------------------------

model = nn.Sequential(

    nn.Linear(64, 10)

 

)

 

optimizer = optim.Adam(model.parameters(), lr=0.01)

 

epochs = 50

 

for epoch in range(epochs + 1):

  sum_losses = 0

  sum_accs =0

 

  for x_batch, y_batch in loader:

    y_pred = model(x_batch)

    loss = nn.CrossEntropyLoss()(y_pred, y_batch)

 

    optimizer.zero_grad()

    loss.backward()

    optimizer.step()

 

    sum_losses = sum_losses + loss

    # 배치 단위 정확도 저장

    y_prob = nn.Softmax(1)(y_pred)

    y_pred_index = torch.argmax(y_prob, axis=1)

    y_batch_index = torch.argmax(y_batch, axis=1)

    acc = (y_batch_index == y_pred_index).float().sum() / len(y_batch) * 100

 

    sum_accs = sum_accs + acc

 

  avg_loss = sum_losses / len(loader)

  avg_acc = sum_accs / len(loader)

 

  print(f'Epoch {epoch:4d}/{epochs} Loss: {avg_loss:.6f} Accuracy: {avg_acc:.2f}%')

결과 :

Epoch    0/50 Loss: 1.973190 Accuracy: 54.77%
Epoch    1/50 Loss: 0.283309 Accuracy: 90.74%
Epoch    2/50 Loss: 0.173487 Accuracy: 94.17%
Epoch    3/50 Loss: 0.136195 Accuracy: 95.20%
Epoch    4/50 Loss: 0.109064 Accuracy: 96.98%
Epoch    5/50 Loss: 0.096432 Accuracy: 97.08%
Epoch    6/50 Loss: 0.080393 Accuracy: 97.96%
Epoch    7/50 Loss: 0.067421 Accuracy: 98.51%
Epoch    8/50 Loss: 0.071652 Accuracy: 98.15%
Epoch    9/50 Loss: 0.060040 Accuracy: 98.55%
Epoch   10/50 Loss: 0.061195 Accuracy: 98.56%
Epoch   11/50 Loss: 0.053115 Accuracy: 98.70%
Epoch   12/50 Loss: 0.055170 Accuracy: 98.76%
Epoch   13/50 Loss: 0.063474 Accuracy: 98.34%
Epoch   14/50 Loss: 0.078649 Accuracy: 97.81%
Epoch   15/50 Loss: 0.049750 Accuracy: 98.42%
Epoch   16/50 Loss: 0.035162 Accuracy: 99.17%
Epoch   17/50 Loss: 0.034921 Accuracy: 99.12%
Epoch   18/50 Loss: 0.037563 Accuracy: 99.25%
Epoch   19/50 Loss: 0.026590 Accuracy: 99.73%
Epoch   20/50 Loss: 0.025800 Accuracy: 99.59%
Epoch   21/50 Loss: 0.031414 Accuracy: 99.18%
Epoch   22/50 Loss: 0.026794 Accuracy: 99.58%
Epoch   23/50 Loss: 0.021841 Accuracy: 99.80%
Epoch   24/50 Loss: 0.020110 Accuracy: 99.80%
Epoch   25/50 Loss: 0.019018 Accuracy: 99.80%
Epoch   26/50 Loss: 0.020113 Accuracy: 99.73%
Epoch   27/50 Loss: 0.016201 Accuracy: 99.93%
Epoch   28/50 Loss: 0.018653 Accuracy: 99.73%
Epoch   29/50 Loss: 0.018847 Accuracy: 99.80%
Epoch   30/50 Loss: 0.019495 Accuracy: 99.73%
Epoch   31/50 Loss: 0.016786 Accuracy: 99.80%
Epoch   32/50 Loss: 0.015192 Accuracy: 99.71%
Epoch   33/50 Loss: 0.014195 Accuracy: 99.86%
Epoch   34/50 Loss: 0.011766 Accuracy: 99.93%
Epoch   35/50 Loss: 0.014165 Accuracy: 99.80%
Epoch   36/50 Loss: 0.011472 Accuracy: 99.86%
Epoch   37/50 Loss: 0.011846 Accuracy: 99.93%
Epoch   38/50 Loss: 0.012318 Accuracy: 100.00%
Epoch   39/50 Loss: 0.011651 Accuracy: 99.93%
Epoch   40/50 Loss: 0.013640 Accuracy: 99.66%
Epoch   41/50 Loss: 0.014027 Accuracy: 99.80%
Epoch   42/50 Loss: 0.009366 Accuracy: 100.00%
Epoch   43/50 Loss: 0.009039 Accuracy: 100.00%
Epoch   44/50 Loss: 0.009845 Accuracy: 99.93%
Epoch   45/50 Loss: 0.010043 Accuracy: 99.93%
Epoch   46/50 Loss: 0.010249 Accuracy: 99.93%
Epoch   47/50 Loss: 0.009615 Accuracy: 100.00%
Epoch   48/50 Loss: 0.008398 Accuracy: 100.00%
Epoch   49/50 Loss: 0.009546 Accuracy: 99.93%
Epoch   50/50 Loss: 0.011114 Accuracy: 99.93%

------------------------------------------------

y_pred = model(x_test)

y_pred[0]

 

결과 :

tensor([ 3.8967, -2.9256, -5.0389, -0.7665, -1.3685, 15.9555, -0.4424,  1.6149,
         1.7313, -0.0853], grad_fn=<SelectBackward0>)

------------------------------------------------

 

y_prob = nn.Softmax(1)(y_pred)

y_prob[0]

결과 :

tensor([5.7936e-06, 6.3101e-09, 7.6254e-10, 5.4668e-08, 2.9941e-08, 9.9999e-01,
        7.5594e-08, 5.9151e-07, 6.6452e-07, 1.0804e-07],
       grad_fn=<SelectBackward0>)

------------------------------------------------

for i in range(10):

  print(f'숫자 {i}일 확률: {y_prob[0][i]:.2f}')

 

결과 :

숫자 0일 확률: 0.00
숫자 1일 확률: 0.00
숫자 2일 확률: 0.00
숫자 3일 확률: 0.00
숫자 4일 확률: 0.00
숫자 5일 확률: 1.00
숫자 6일 확률: 0.00
숫자 7일 확률: 0.00
숫자 8일 확률: 0.00
숫자 9일 확률: 0.00

------------------------------------------------

# 배치 단위 정확도 저장

y_pred_index = torch.argmax(y_prob, axis=1)

y_test_index = torch.argmax(y_test, axis=1)

accuracy = (y_test_index == y_pred_index).float().sum() / len(y_test) * 100

print(f'테스트 정확도는 {accuracy:.2f}%입니다')

 

결과 :

테스트 정확도는 96.39%입니다