[PyTorch小試牛刀]實戰三·DNN實現邏輯迴歸對FashionMNIST資料集進行分類

內容還包括了網路模型引數的保存於載入。
資料集
下載地址
程式碼部分

import torch as t
import torchvision as tv
import numpy as np


# 超引數
EPOCH = 10
BATCH_SIZE = 100
DOWNLOAD_MNIST = True   # 下過資料的話, 就可以設定成 False
N_TEST_IMG = 10          # 到時候顯示 5張圖片看效果, 如上圖一



class DNN(t.nn.Module)
 
:
    def __init__(self):
        super(DNN, self).__init__()

        train_data = tv.datasets.FashionMNIST(
        root="./mnist/",
        train=True,
        transform=tv.transforms.ToTensor(),
        download=DOWNLOAD_MNIST
        )

        test_data = tv.datasets.FashionMNIST(
        root="./mnist/"
 
,
        train=False,
        transform=tv.transforms.ToTensor(),
        download=DOWNLOAD_MNIST
        )

        print(test_data)


        # Data Loader for easy mini-batch return in training, the image batch shape will be (50, 1, 28, 28)
        self.train_loader = t.utils.data.DataLoader(
            dataset=
 
train_data, 
            batch_size=BATCH_SIZE,
            shuffle=True)

        self.test_loader = t.utils.data.DataLoader(
            dataset=test_data, 
            batch_size=1000,
            shuffle=True)

        self.dnn = t.nn.Sequential(
            t.nn.Linear(28*28,256),
            t.nn.Dropout(0.5),
            t.nn.ELU(),
            t.nn.Linear(256,64),
            t.nn.Dropout(0.5),
            t.nn.ELU(),
            t.nn.Linear(64,10)
        )

        self.lr = 0.001
        self.loss = t.nn.CrossEntropyLoss()
        self.opt = t.optim.Adam(self.parameters(), lr = self.lr)

    def forward(self,x):
        out = self.dnn(x)
        return(out)

def train():
    model = DNN()
    print(model)
    loss = model.loss
    opt = model.opt
    dataloader = model.train_loader
    testloader = model.test_loader

    

    for e in range(EPOCH):
        step = 0
        for (x, y) in (dataloader):
            model.train()# train model dropout used
            step += 1
            b_x = x.view(-1, 28*28)   # batch x, shape (batch, 28*28)
            b_y = y
            out = model(b_x)
            losses = loss(out,b_y)
            opt.zero_grad()
            losses.backward()
            opt.step()
            if(step%100 == 0):
                print(e,step,losses.data.numpy())
                model.eval() # train model dropout not use
                for (tx,ty) in testloader:
                    t_x = tx.view(-1, 28*28)   # batch x, shape (batch, 28*28)
                    t_y = ty
                    t_out = model(t_x)
                    acc = (np.argmax(t_out.data.numpy(),axis=1) == t_y.data.numpy())
                    print(np.sum(acc)/1000)
                    break#只測試前1000個
            


    t.save(model, './model.pkl')  # 儲存整個網路
    t.save(model.state_dict(), './model_params.pkl')   # 只儲存網路中的引數 (速度快, 佔記憶體少)
    #載入引數的方式
    """net = DNN()
    net.load_state_dict(t.load('./model_params.pkl'))
    net.eval()"""
    #載入整個模型的方式
    net = t.load('./model.pkl')
    net.eval()
    for (tx,ty) in testloader:
        t_x = tx.view(-1, 28*28)   # batch x, shape (batch, 28*28)
        t_y = ty
        t_out = net(t_x)
        acc = (np.argmax(t_out.data.numpy(),axis=1) == t_y.data.numpy())
        print(np.sum(acc)/1000)

if __name__ == "__main__":
    train()

輸出結果

9 500 0.42454192
0.875
9 600 0.4553349
0.888
0.876
0.868
0.868
0.881
0.864
0.87
0.87
0.854
0.871
0.879