DataFountain-FashionMNIST：CNN基础 准确率0.85

import pandas as pdimport numpy as np# 数据读取train_df = pd.read_csv('fashion-mnist_train.csv')
test_df = pd.read_csv('fashion-mnist_test_data.csv')

train_df.head()

%pylab inline# 绘制10*10的图片plt.figure(figsize=(10, 10))for idx in range(100):    # 图片尺寸从784 -》 28 * 28
    xy = train_df.iloc[idx].values[1:].reshape(28,28)
    plt.subplot(10, 10, idx+1)    # 展示图片
    plt.imshow(xy, cmap='gray')
    plt.xticks([]); plt.yticks([])

import paddle
paddle.__version__

from paddle.io import DataLoader, Datasetfrom PIL import Image# 自定义数据集完成数据读取class MyDataset(Dataset):
    def __init__(self, img, label):
        super(MyDataset, self).__init__()
        self.img = img
        self.label = label    
    def __getitem__(self, index):
        img = self.img[index]        # 对数据集进行归一化
        return img/255, int(self.label[index])    def __len__(self):
        return len(self.label)# 使用5.9w进行训练train_dataset = MyDataset(
    train_df.iloc[:-1000, 1:].values.reshape(59000, 28, 28).astype(np.float32), 
    paddle.to_tensor(train_df.label.iloc[:-1000].values.astype(np.float32))
)
train_loader = DataLoader(train_dataset, batch_size=300, shuffle=True)# 最后1k作为验证集val_dataset = MyDataset(
    train_df.iloc[-1000:, 1:].values.reshape(1000, 28, 28).astype(np.float32), 
    paddle.to_tensor(train_df.label.iloc[-1000:].values.astype(np.float32))
)
val_loader = DataLoader(val_dataset, batch_size=300, shuffle=False)# 最终测试集test_dataset = MyDataset(
    test_df.iloc[:, 1:].values.reshape(10000, 28, 28).astype(np.float32),
    paddle.to_tensor(np.zeros((test_df.shape[0])))
)
test_loader = DataLoader(test_dataset, batch_size=64, shuffle=False)

# 构建全连接模型，先需要从多维转变为二维model = paddle.nn.Sequential(
    paddle.nn.Flatten(),
    paddle.nn.Linear(28*28,128),
    paddle.nn.LeakyReLU(),
    paddle.nn.Linear(128, 10)
)

paddle.summary(model, (64, 28, 28))

# 定义优化器，损失函数optimizer = paddle.optimizer.Adam(parameters=model.parameters(), learning_rate=0.0001)
criterion = paddle.nn.CrossEntropyLoss()# 每个epoch迭代训练#     每个batch -> 正向传播 -> 计算损失 -> 更新参数for epoch in range(0, 5):
    Train_Loss, Val_Loss = [], []
    Train_ACC, Val_ACC = [], []    # 训练部分
    model.train()    for i, (x, y) in enumerate(train_loader):
        pred = model(x)
        loss = criterion(pred, y)
        Train_Loss.append(loss.item())
        loss.backward()
        optimizer.step()
        optimizer.clear_grad()
        Train_ACC.append((pred.numpy().argmax(1) == y.numpy()).mean())        
        if i % 100 == 0:            print(f'{i}/{len(train_loader)}\t Loss {np.mean(Train_Loss):3.5f} {np.mean(Train_ACC):3.5f}')    # 验证部分
    model.eval()    for i, (x, y) in enumerate(val_loader):
        pred = model(x)
        loss = criterion(pred, y)
        Val_Loss.append(loss.item())
        Val_ACC.append((pred.numpy().argmax(1) == y.numpy()).mean())    
    if epoch % 1 == 0:        print(f'\nEpoch: {epoch}')        print(f'Loss {np.mean(Train_Loss):3.5f}/{np.mean(Val_Loss):3.5f}')        print(f'ACC {np.mean(Train_ACC):3.5f}/{np.mean(Val_ACC):3.5f}')

# 卷积模型model = paddle.nn.Sequential(
    paddle.nn.Conv2D(1, 10, (5, 5)),
    paddle.nn.ReLU(),
    paddle.nn.Conv2D(10, 20, (5, 5)),
    paddle.nn.ReLU(),
    paddle.nn.MaxPool2D((2, 2)),

    paddle.nn.Flatten(),
    paddle.nn.Linear(2000, 10),
)

paddle.summary(model, (64, 1, 28, 28))

# 优化器与损失函数optimizer = paddle.optimizer.Adam(parameters=model.parameters(), learning_rate=0.0001)
criterion = paddle.nn.CrossEntropyLoss()# 训练部分for epoch in range(0, 5):
    Train_Loss, Val_Loss = [], []
    Train_ACC, Val_ACC = [], []

    model.train()    for i, (x, y) in enumerate(train_loader):
        pred = model(x.reshape((-1, 1, 28, 28)))
        loss = criterion(pred, y)
        Train_Loss.append(loss.item())
        loss.backward()
        optimizer.step()
        optimizer.clear_grad()
        Train_ACC.append((pred.numpy().argmax(1) == y.numpy()).mean())        
        if i % 100 == 0:            print(f'{i}/{len(train_loader)}\t Loss {np.mean(Train_Loss):3.5f} {np.mean(Train_ACC):3.5f}')

    model.eval()    for i, (x, y) in enumerate(val_loader):
        pred = model(x.reshape((-1, 1, 28, 28)))
        loss = criterion(pred, y)
        Val_Loss.append(loss.item())
        Val_ACC.append((pred.numpy().argmax(1) == y.numpy()).mean())    
    if epoch % 1 == 0:        print(f'\nEpoch: {epoch}')        print(f'Loss {np.mean(Train_Loss):3.5f}/{np.mean(Val_Loss):3.5f}')        print(f'ACC {np.mean(Train_ACC):3.5f}/{np.mean(Val_ACC):3.5f}')