MNIST是机器学习领域最有名的数据集之一,被应用于从简单的实验到发表的论文研究等各种场合。实际上,在阅读图像识别或机器学习的论文时,MNIST数据集经常作为实验用的数据出现
MNIST数据集是由0到9的数字图像构成的。训练图像有6万张,测试图有1万张,图像数据是28像素×28像素的灰度图像(1通道),各个像素的取值在0到255之间。每个图像数据都相应地标有“7”“2”“1”等标签。鱼书提供了MNIST数据集的下载脚本,我再做了一些细微的调整如下:
mnist.py
# coding: utf-8 try: import urllib.request except ImportErros: raise ImportError('You should use Python 3.x') import os.path import gzip import pickle import os import numpy as np 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 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 load_mnist(normalize=True, flatten=True, one_hot_label=False): """ Load MNIST dataset :param normalize: normalize the pixel value of the image to 0.0~1.0 :param one_hot_label: [0,0,1,0,0,0,0,0,0,0] :param flatten: expand the image into a one-dimensional array :returns: (training image, training label), (test image, test label) """ 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']) 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 _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 _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 _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 if __name__ == '__main__': init_mnist()
注:原代码中的download_mnist()运行报503错误,原因似乎是目前mnist的官方数据集网址http://yann.lecun.com/exdb/mnist/正在维护,因此我直接从https://discuss.pytorch.org/t/mnist-server-down/114433/13里手动下载了四个数据集文件到路径下
Python有pickle这个便利的功能。这个功能可以将程序运行中的对象保存为文件。如果加载保存过的pickle文件,可以立刻复原之前程序运行中的对象。用于读入MNIST数据集的load_mnist()函数内部也使用了pickle功能(在第2次及以后读入时)。利用pickle功能可以高效地完成MNIST数据地准备工作(鱼书原文)
我检索pickle文档和部分资料了解到,pickle是Python对object对象进行序列化和反序列化的工具,与json相似,pickle也可以完成两个不同进程之间的数据同行,也可以序列化后通过web传输后再在远程端反序列化后使用,是一种多程序间通信的一种有效手段。
通过运行下面的代码可以一窥pickle在上述mnist.py的作用
读取mnist数据集:
import sys, os sys.path.append(os.pardir) from dataset.mnist import load_mnist (x_train, t_train), (x_test, t_test) = load_mnist(flatten=True, normalize=False) print(x_train.shape) print(t_train.shape) print(x_test.shape) print(t_test.shape)
PIL模块确认数据:
import sys, os sys.path.append(os.pardir) import numpy as np from dataset.mnist import load_mnist from PIL import Image def img_show(img): pil_img = Image.fromarray(np.uint8(img)) pil_img.show() (x_train, t_train), (x_test, t_test) = load_mnist(flatten=True, normalize=False) img = x_train[0] label = t_train[0] print(label) print(img.shape) img = img.reshape(28, 28) print(img.shape) img_show(img)
这里需要注意的是,flatten=True时读入的图像是以一列(一维)NumPy数组的形式保存的。因此,显示图像时,需要把它变为原来的28像素×28像素的形状。可以通过reshape()方法的参数指定期望的形状,更改NumPy数组的形状。此外,还需要把保存为NumPy数组的图像数据转换为PIL用的数据对象,这个转换处理由Image.fromarray()来完成