MNIST数据库格式的解析和生成

　　如上一篇博客所写MNIST是ML界的’hello world’，为了将自己的图像转化为类似MNIST数据文件类型的格式，先对它的文件进行了解析．先给出我的程序所提取的训练样本的前十张图像及对应的label，截图如下：

　　该数据格式是bytestream，无论是训练样本还是测试样本，其图像数据文件均在开头有一个2051的标志，之后便是图像的个数／行值／列值，紧接着按行读取所有的图像，且图像数据间无间隔；label数据文件均在开头有一个2049的标志，然后是图像的个数，以及每个图像的标志（如0,1）依次列出，以bytestream形式排列的文件在进行压缩，便是我们下载到的数据文件．

　　以下为文章开始给出的结果的源码：

#coding=utf-8from __future__ import absolute_importfrom __future__ import divisionfrom __future__ import print_functionimport matplotlib.pyplot as pltimport gzipimport osimport Imageimport tensorflow.python.platformimport numpyfrom six.moves import urllibfrom six.moves import xrange  # pylint: disable=redefined-builtinimport tensorflow as tfdef _read32(bytestream):  dt = numpy.dtype(numpy.uint32).newbyteorder('>')  return numpy.frombuffer(bytestream.read(4), dtype=dt)[0]def extract_images(filename,nth):  with gzip.open(filename) as bytestream:    magic = _read32(bytestream)    if magic != 2051:      raise ValueError(          'Invalid magic number %d in MNIST image file: %s' %          (magic, filename))    num_images = _read32(bytestream)    #print(num_images)    rows = _read32(bytestream)    cols = _read32(bytestream)    #print(rows)#28    #print(cols)#28    for i in range(nth-1):       bytestream.read(rows * cols)    buf = bytestream.read(rows * cols )    data = numpy.frombuffer(buf, dtype=numpy.uint8)#按行读取，图片间无间隔    data = numpy.reshape(data, (rows, cols))    return datadef extract_labels(filename, one_hot=False):  with gzip.open(filename) as bytestream:    magic = _read32(bytestream)    if magic != 2049:      raise ValueError(          'Invalid magic number %d in MNIST label file: %s' %          (magic, filename))    num_items = _read32(bytestream)    print(num_items)    buf = bytestream.read(10)#num_items    labels = numpy.frombuffer(buf, dtype=numpy.uint8)    if one_hot:      return dense_to_one_hot(labels)    return labelsif __name__=='__main__':    plt.figure(1)     for nth in range(1,11):    data = extract_images('train-images-idx3-ubyte.gz',nth)    new_im = Image.fromarray(data)                plt.subplot(2,5,nth)    plt.imshow(new_im, cmap ='gray')    plt.title(nth)    train_labels = extract_labels('train-labels-idx1-ubyte.gz', one_hot=False)    print(train_labels)    plt.show()