手写数字MNIST数据集

import struct
import numpy as np
from torch.utils.data import Dataset, DataLoader

def softmax(x):
    ex = np.exp(x)
    sum_ex = np.sum(ex, axis=-1, keepdims = True)
    return ex/sum_ex

def load_labels(file):  # 加载数据
    with open(file, "rb") as f:
        data = f.read()
    return np.asanyarray(bytearray(data[8:]), dtype=np.int32)

def load_images(file):  # 加载数据
    with open(file, "rb") as f:
        data = f.read()
    magic_number, num_items, rows, cols = struct.unpack(">iiii", data[:16])
    return np.asanyarray(bytearray(data[16:]), dtype=np.uint8).reshape(num_items, -1)

class Model():
    def __init__(self):
        self.w = np.ones(shape = (784, 10))

    def forward(self, x):
        predict = x @ self.w
        p = softmax(predict)
        return p

class MyDataset(Dataset):
    def __init__(self, data, label):
        self.data = data
        self.label = label

    def __len__(self):
        return len(self.label)

    def __getitem__(self, item):
        image = self.data[item] / 255
        label = self.label[item]

        one_hot_label = [0 for i in range(10)]
        one_hot_label[label] = 1

        return image, np.array(one_hot_label)







if __name__ == "__main__":
    train_images = load_images("mnist_data/MNIST/raw/train-images-idx3-ubyte")
    train_label = load_labels("mnist_data/MNIST/raw/train-labels-idx1-ubyte")

    test_images = load_images("mnist_data/MNIST/raw/t10k-images-idx3-ubyte")
    test_label = load_labels("mnist_data/MNIST/raw/t10k-labels-idx1-ubyte")

    batch_size = 20
    epoch = 30
    lr = 0.0005

    train_datasets = MyDataset(train_images,train_label)
    train_dataloader = DataLoader(train_datasets,batch_size,shuffle=True)

    test_datasets = MyDataset(test_images, test_label)
    test_dataloader = DataLoader(test_datasets, batch_size, shuffle=False)


    model = Model()

    for e in range(epoch):
        for train_data, train_label in train_dataloader:
            train_data = np.array(train_data)
            train_label = np.array(train_label)

            predict = model.forward(train_data)

            loss = -np.sum(train_label * np.log(predict))/batch_size

            G = predict - train_label

            delt_w = train_data.T @ G
            model.w -= lr*delt_w

        right_num = 0
        for text_data, text_label in test_dataloader:  # 10 * 784
                text_data = np.array(text_data)
                text_label = np.array(text_label)

                predict = model.forward(text_data)

                predict = np.argmax(predict,axis=-1)
                text_label = np.argmax(text_label,axis=-1)

                right_num += np.sum(text_label==predict)

        acc = right_num / len(test_datasets)
        print("acc:", acc)

一定要确保整套任务中，各变量类型统一为numpy类型或tensor类型