下载结果

运行代码

# coding: utf-8

try:

   import urllib.request

except ImportError:

   raise ImportError('You should use Python 3.x')

import os.path

import gzip

import pickle

import os

import numpy as np

url_base = 'http://yann.lecun.com/exdb/mnist/'

key_file = {

   'train_img':'train-images-idx3-ubyte.gz',

   'train_label':'train-labels-idx1-ubyte.gz',

   'test_img':'t10k-images-idx3-ubyte.gz',

   'test_label':'t10k-labels-idx1-ubyte.gz'

}

dataset_dir = os.path.dirname(os.path.abspath(__file__))

save_file = dataset_dir + "/mnist.pkl"

train_num = 60000

test_num = 10000

img_dim = (1, 28, 28)

img_size = 784

def _download(file_name):

   file_path = dataset_dir + "/" + file_name

   

   if os.path.exists(file_path):

       return

   print("Downloading " + file_name + " ... ")

   urllib.request.urlretrieve(url_base + file_name, file_path)

   print("Done")

   

def download_mnist():

   for v in key_file.values():

      _download(v)

       

def _load_label(file_name):

   file_path = dataset_dir + "/" + file_name

   

   print("Converting " + file_name + " to NumPy Array ...")

   with gzip.open(file_path, 'rb') as f:

           labels = np.frombuffer(f.read(), np.uint8, offset=8)

   print("Done")

   

   return labels

def _load_img(file_name):

   file_path = dataset_dir + "/" + file_name

   

   print("Converting " + file_name + " to NumPy Array ...")    

   with gzip.open(file_path, 'rb') as f:

           data = np.frombuffer(f.read(), np.uint8, offset=16)

   data = data.reshape(-1, img_size)

   print("Done")

   

   return data

   

def _convert_numpy():

   dataset = {}

   dataset['train_img'] =  _load_img(key_file['train_img'])

   dataset['train_label'] = _load_label(key_file['train_label'])    

   dataset['test_img'] = _load_img(key_file['test_img'])

   dataset['test_label'] = _load_label(key_file['test_label'])

   

   return dataset

def init_mnist():

   download_mnist()

   dataset = _convert_numpy()

   print("Creating pickle file ...")

   with open(save_file, 'wb') as f:

       pickle.dump(dataset, f, -1)

   print("Done!")

def _change_one_hot_label(X):

   T = np.zeros((X.size, 10))

   for idx, row in enumerate(T):

       row[X[idx]] = 1

       

   return T

   

def load_mnist(normalize=True, flatten=True, one_hot_label=False):

   """读入MNIST数据集

   

   Parameters

   ----------

   normalize : 将图像的像素值正规化为0.0~1.0

   one_hot_label :

       one_hot_label为True的情况下，标签作为one-hot数组返回

       one-hot数组是指[0,0,1,0,0,0,0,0,0,0]这样的数组

   flatten : 是否将图像展开为一维数组

   

   Returns

   -------

   (训练图像, 训练标签), (测试图像, 测试标签)

   """

   if not os.path.exists(save_file):

       init_mnist()

       

   with open(save_file, 'rb') as f:

       dataset = pickle.load(f)

   

   if normalize:

       for key in ('train_img', 'test_img'):

           dataset[key] = dataset[key].astype(np.float32)

           dataset[key] /= 255.0

           

   if one_hot_label:

       dataset['train_label'] = _change_one_hot_label(dataset['train_label'])

       dataset['test_label'] = _change_one_hot_label(dataset['test_label'])

   

   if not flatten:

        for key in ('train_img', 'test_img'):

           dataset[key] = dataset[key].reshape(-1, 1, 28, 28)

   return (dataset['train_img'], dataset['train_label']), (dataset['test_img'], dataset['test_label'])

if __name__ == '__main__':

   init_mnist()