在官網下載minist資料集，速度感人。因此特意上傳了一份，供需要的朋友下載：minist資料集
第一次開啟這個資料集，也是一臉懵，不過還好，脫坑程式碼如下：

import numpy as np
import os

# 訓練集
with open('./minist_data/train-images.idx3-ubyte') as f:
    loaded = np.fromfile(file = f, dtype = np.uint8)
    train_data = loaded[16:].reshape((60000, 784))
print(train_data.shape)
 
 # (60000, 784)


with open('./minist_data/train-labels.idx1-ubyte') as f:
    loaded = np.fromfile(file = f, dtype = np.uint8)
    train_labels = loaded[8:]
print(train_labels.shape) # (60000,)


# 測試集
with open('./minist_data/t10k-images.idx3-ubyte') as f:
    loaded = np.fromfile(file=f, dtype=np.uint8)
    test_data =
 
 loaded[16:].reshape((10000, 784))
print(test_data.shape) # (10000, 784)

with open('./minist_data/t10k-labels.idx1-ubyte') as f:
    loaded = np.fromfile(file=f, dtype=np.uint8)
    test_labels = loaded[8:].reshape((10000))
print(test_labels.shape) # (10000,)

可以看到，訓練集有六萬條樣本，784個特徵。測試集一萬條樣本。