Fashion-MNIST时尚物品分类

 

In [1]:

import torch
from torch import nn
from torch.utils.data import DataLoader
from torchvision import datasets
from torchvision.transforms import ToTensor
import matplotlib.pyplot as plt

 

In [2]:

train_data = datasets.FashionMNIST(root='data', train=True, download=True, transform=ToTensor())
test_data = datasets.FashionMNIST(root='data', train=False, download=True, transform=ToTensor())

 

In [3]:

batch_size = 64
train_dl = DataLoader(train_data, batch_size=batch_size, shuffle=True)
test_dl = DataLoader(test_data, batch_size=batch_size, shuffle=True)

 

In [4]:

images, labels = next(iter(train_data))
display(images.shape)
labels

 

 

torch.Size([1, 28, 28])

Out[4]:

9

 

In [5]:

plt.imshow(images[0].squeeze())

 

Out[5]:

<matplotlib.image.AxesImage at 0x1e2673644f0>

 

 

In [6]:

class Model(nn.Module):
    def __init__(self):
        super().__init__()
        self.flatten = nn.Flatten()
        self.linear_selu_stack = nn.Sequential(
            nn.Linear(28*28, 512),
            nn.SELU(),
            nn.Linear(512, 512),
            nn.SELU(),
            nn.Linear(512, 10)
        )

    def forward(self, x):
        x = self.flatten(x)
        return self.linear_selu_stack(x)

 

In [7]:

model = Model()
loss_fn = nn.CrossEntropyLoss()
optimizer = torch.optim.Adam(model.parameters(), lr=0.001)

 

In [8]:

def train(dataloader, model, loss_fn, optimizer):
    size = len(dataloader.dataset)
    model.train()
    for batch, (X, y) in enumerate(dataloader):
        pred = model(X)
        loss = loss_fn(pred, y)
        optimizer.zero_grad()
        loss.backward()
        optimizer.step()

        if batch%100 == 0:
            loss, current = loss.item(), batch*len(X)

 

In [9]:

def test(dataloader, model, loss_fn):
    size = len(dataloader.dataset)
    num_batch = len(dataloader)
    model.eval()
    test_loss, correct = 0, 0
    with torch.no_grad():
        for X, y in dataloader:
            pred = model(X)
            test_loss += loss_fn(pred, y).item()
            correct += (pred.argmax(1) == y).type(torch.float).sum().item()
    test_loss /= num_batch
    correct /= size
    return correct, test_loss

 

In [10]:

epochs = 10
for t in range(epochs):
    train(train_dl, model, loss_fn, optimizer)
    correct, test_loss = test(test_dl, model, loss_fn)
    print('Epoch:', t+1, '  Accuracy:', round(correct*100, 2), '  Avg Loss:', round(test_loss, 5))

 

 

Epoch: 1   Accuracy: 82.49   Avg Loss: 0.46003
Epoch: 2   Accuracy: 84.79   Avg Loss: 0.42927
Epoch: 3   Accuracy: 84.72   Avg Loss: 0.41863
Epoch: 4   Accuracy: 86.33   Avg Loss: 0.38519
Epoch: 5   Accuracy: 84.8   Avg Loss: 0.42816
Epoch: 6   Accuracy: 86.44   Avg Loss: 0.3858
Epoch: 7   Accuracy: 86.78   Avg Loss: 0.37337
Epoch: 8   Accuracy: 87.14   Avg Loss: 0.35727
Epoch: 9   Accuracy: 87.54   Avg Loss: 0.38788
Epoch: 10   Accuracy: 87.98   Avg Loss: 0.36436

 

In [12]:

# 如上可以看出这是个很简单的问题，不用卷积神经网络也有很好的准确率
classes = ['T-shirt/top', 'Trouser', 'Pullover', 'Dress', 'Coat', 'Sandal', 'Shirt', 'Sneaker', 'Bag', 'Ankle boot']
model.eval()
x, y = test_data[0][0], test_data[0][1]
with torch.no_grad():
    pred = model(x)
    predicted, actual = classes[pred[0].argmax(0)], classes[y]
    print(f'Predicted: "{predicted}", Actual: "{actual}"')

 

 

Predicted: "Ankle boot", Actual: "Ankle boot"