Python实现的下载并显⽰mnist⼿写数据集  1from PIL import Image
2
3try:
quest
5except ImportError:
6raise ImportError('You should use Python 3.x')
7import os.path
8import gzip
9import pickle
10import os
11import numpy as np
12
13 url_base = 'yann.lecun/exdb/mnist/'
14 key_file = {
15'train_img': '',
16'train_label': '',
17'test_img': '',
18'test_label': ''
19 }
20
21 dataset_dir = os.path.dirname(os.path.abspath(__file__))
22 save_file = dataset_dir + "/mnist.pkl"
23
24 train_num = 60000
25 test_num = 10000
26 img_dim = (1, 28, 28)
27 img_size = 784
28
29
30def _download(file_name):
31    file_path = dataset_dir + "/" + file_name
32
33if ists(file_path):
34return
35
36print("Downloading " + file_name + " ... ")
37    quest.urlretrieve(url_base + file_name, file_path)
38print("Done")
39
40
41def download_mnist():
42for v in key_file.values():
43        _download(v)
44
45
46def _load_label(file_name):
47    file_path = dataset_dir + "/" + file_name
48
49print("Converting " + file_name + " to NumPy Array ...")
50    with gzip.open(file_path, 'rb') as f:
51        labels = np.ad(), np.uint8, offset=8)
52print("Done")
53
54return labels
55
56
57def _load_img(file_name):
58    file_path = dataset_dir + "/" + file_name
59
60print("Converting " + file_name + " to NumPy Array ...")
61    with gzip.open(file_path, 'rb') as f:
62        data = np.ad(), np.uint8, offset=16)
63    data = shape(-1, img_size)
64print("Done")
65
66return data
67
68
69def _convert_numpy():
70    dataset = {}
71    dataset['train_img'] = _load_img(key_file['train_img'])
72    dataset['train_label'] = _load_label(key_file['train_label'])
73    dataset['test_img'] = _load_img(key_file['test_img'])
74    dataset['test_label'] = _load_label(key_file['test_label'])
75
76return dataset
77
78
79def init_mnist():
80    download_mnist()
81    dataset = _convert_numpy()
82print("Creating pickle file ...")
83    with open(save_file, 'wb') as f:
84        pickle.dump(dataset, f, -1)
85print("Done!")
86
87
88def _change_one_hot_label(X):
89    T = np.zeros((X.size, 10))
90for idx, row in enumerate(T):
91        row[X[idx]] = 1
92
93return T
94
95
96def load_mnist(normalize=True, flatten=True, one_hot_label=False):
97"""读⼊MNIST数据集
98
99    Parameters
100    ----------
101    normalize : 将图像的像素值正规化为0.0~1.0
102    one_hot_label :
103        one_hot_label为True的情况下，标签作为one-hot数组返回
104        one-hot数组是指[0,0,1,0,0,0,0,0,0,0]这样的数组
105    flatten : 是否将图像展开为⼀维数组
106
107    Returns
108    -------
109    (训练图像, 训练标签), (测试图像, 测试标签)
110"""
111if not ists(save_file):
112        init_mnist()
113
114    with open(save_file, 'rb') as f:
115        dataset = pickle.load(f)
116
117if normalize:
118for key in ('train_img', 'test_img'):
119            dataset[key] = dataset[key].astype(np.float32)
120            dataset[key] /= 255.0
121
122if one_hot_label:
123        dataset['train_label'] = _change_one_hot_label(dataset['train_label'])
124        dataset['test_label'] = _change_one_hot_label(dataset['test_label'])
125
126if not flatten:
127for key in ('train_img', 'test_img'):
128            dataset[key] = dataset[key].reshape(-1, 1, 28, 28)
129
130return (dataset['train_img'], dataset['train_label']), (dataset['test_img'], dataset['test_label']) 131
132def img_show(img):
133    pil_img = Image.fromarray(np.uint8(img))
134    pil_img.show()
135
136
137if__name__ == '__main__':
138    init_mnist()
139    (x_train, t_train), (x_test, t_test) = load_mnist(flatten=True, normalize=False)
140
141    img = x_train[1]
142    label = t_train[1]
143print(label)  # 5
144
145print(img.shape)  # (784,)import pickle
146    img = shape(28, 28)  # 把图像的形状变为原来的尺⼨
147print(img.shape)  # (28, 28)
148
149    img_show(img)

版权声明：本站内容均来自互联网，仅供演示用，请勿用于商业和其他非法用途。如果侵犯了您的权益请与我们联系QQ:729038198，我们将在24小时内删除。