快捷導(dǎo)航

python MNIST手寫識(shí)別數(shù)據(jù)調(diào)用API的方法

更新時(shí)間：2018年08月08日 10:54:35 作者：caichao08

這篇文章主要介紹了python MNIST手寫識(shí)別數(shù)據(jù)調(diào)用API的方法,小編覺(jué)得挺不錯(cuò)的，現(xiàn)在分享給大家，也給大家做個(gè)參考。一起跟隨小編過(guò)來(lái)看看吧

MNIST數(shù)據(jù)集比較小，一般入門機(jī)器學(xué)習(xí)都會(huì)采用這個(gè)數(shù)據(jù)集來(lái)訓(xùn)練

下載地址：yann.lecun.com/exdb/mnist/

有4個(gè)有用的文件：
train-images-idx3-ubyte: training set images
train-labels-idx1-ubyte: training set labels
t10k-images-idx3-ubyte: test set images
t10k-labels-idx1-ubyte: test set labels

The training set contains 60000 examples, and the test set 10000 examples. 數(shù)據(jù)集存儲(chǔ)是用binary file存儲(chǔ)的，黑白圖片。

下面給出load數(shù)據(jù)集的代碼：

import os
import struct
import numpy as np
import matplotlib.pyplot as plt

def load_mnist():
  '''
  Load mnist data
  http://yann.lecun.com/exdb/mnist/

  60000 training examples
  10000 test sets

  Arguments:
    kind: 'train' or 'test', string charater input with a default value 'train'

  Return:
    xxx_images: n*m array, n is the sample count, m is the feature number which is 28*28
    xxx_labels: class labels for each image, (0-9)
  '''

  root_path = '/home/cc/deep_learning/data_sets/mnist'

  train_labels_path = os.path.join(root_path, 'train-labels.idx1-ubyte')
  train_images_path = os.path.join(root_path, 'train-images.idx3-ubyte')

  test_labels_path = os.path.join(root_path, 't10k-labels.idx1-ubyte')
  test_images_path = os.path.join(root_path, 't10k-images.idx3-ubyte')

  with open(train_labels_path, 'rb') as lpath:
    # '>' denotes bigedian
    # 'I' denotes unsigned char
    magic, n = struct.unpack('>II', lpath.read(8))
    #loaded = np.fromfile(lpath, dtype = np.uint8)
    train_labels = np.fromfile(lpath, dtype = np.uint8).astype(np.float)

  with open(train_images_path, 'rb') as ipath:
    magic, num, rows, cols = struct.unpack('>IIII', ipath.read(16))
    loaded = np.fromfile(train_images_path, dtype = np.uint8)
    # images start from the 16th bytes
    train_images = loaded[16:].reshape(len(train_labels), 784).astype(np.float)

  with open(test_labels_path, 'rb') as lpath:
    # '>' denotes bigedian
    # 'I' denotes unsigned char
    magic, n = struct.unpack('>II', lpath.read(8))
    #loaded = np.fromfile(lpath, dtype = np.uint8)
    test_labels = np.fromfile(lpath, dtype = np.uint8).astype(np.float)

  with open(test_images_path, 'rb') as ipath:
    magic, num, rows, cols = struct.unpack('>IIII', ipath.read(16))
    loaded = np.fromfile(test_images_path, dtype = np.uint8)
    # images start from the 16th bytes
    test_images = loaded[16:].reshape(len(test_labels), 784)  

  return train_images, train_labels, test_images, test_labels

再看看圖片集是什么樣的：

def test_mnist_data():
  '''
  Just to check the data

  Argument:
    none

  Return:
    none
  '''
  train_images, train_labels, test_images, test_labels = load_mnist()
  fig, ax = plt.subplots(nrows = 2, ncols = 5, sharex = True, sharey = True)
  ax =ax.flatten()
  for i in range(10):
    img = train_images[i][:].reshape(28, 28)
    ax[i].imshow(img, cmap = 'Greys', interpolation = 'nearest')
    print('corresponding labels = %d' %train_labels[i])

if __name__ == '__main__':
  test_mnist_data()

跑出的結(jié)果如下：