【keras】解决 example 案例中 MNIST 数据集下载不了的问题

前言：

 

keras 源码中下载MNIST的方式是 path = get_file(path, origin='https://s3.amazonaws.com/img-datasets/mnist.npz')，数据源是通过 url = https://s3.amazonaws.com/img-datasets/mnist.npz 进行下载的。访问该 url 地址被墙了，导致 MNIST 相关的案例都卡在数据下载的环节。本文主要提供解决方案，让需要的读者可以跑案例的代码感受一下。

 

本文的贡献主要包括如下：

 

1）提供 mnist_npz 数据集；

2）分析了关于 mnist 几个相关的源代码；

3）提供了一种能够顺利运行 keras 源码中 example 下 mnist 的相关案例；

4）找到了另外几种解决方案，提供了相关的链接。

 

numpy.load(path)

 

numpy.load() 函数起到很重要的作用。它可以读取 .npy .npz 等文件类型，并返回对应的数据类型。

1）如果文件类型是 .pny 则返回一个1维数组。

2）如果文件类型是 .npz 则返回一个类似字典的数据类型，包含 {filename: array} 键值对。如，本例中的键值对如下所示：

 

f = np.load(path)
x_train, y_train = f['x_train'], f['y_train']
x_test, y_test = f['x_test'], f['y_test']
f.close()

 

详情请参考：https://docs.scipy.org/doc/numpy/reference/generated/numpy.load.html

 

 

原始 .\keras\examples\mnist_mlp.py

https://github.com/keras-team/keras/blob/master/examples/mnist_mlp.py

# -*- coding: utf-8 -*-  
'''Trains a simple deep NN on the MNIST dataset.

Gets to 98.40% test accuracy after 20 epochs
(there is *a lot* of margin for parameter tuning).
2 seconds per epoch on a K520 GPU.
'''

from __future__ import print_function

import keras
from keras.datasets import mnist
from keras.models import Sequential
from keras.layers import Dense, Dropout
from keras.optimizers import RMSprop


batch_size = 128
num_classes = 10
epochs = 20

# the data, shuffled and split between train and test sets
(x_train, y_train), (x_test, y_test) = mnist.load_data()

x_train = x_train.reshape(60000, 784)
x_test = x_test.reshape(10000, 784)
x_train = x_train.astype('float32')
x_test = x_test.astype('float32')
x_train /= 255
x_test /= 255
print(x_train.shape[0], 'train samples')
print(x_test.shape[0], 'test samples')

# convert class vectors to binary class matrices
y_train = keras.utils.to_categorical(y_train, num_classes)
y_test = keras.utils.to_categorical(y_test, num_classes)

model = Sequential()
model.add(Dense(512, activation='relu', input_shape=(784,)))
model.add(Dropout(0.2))
model.add(Dense(512, activation='relu'))
model.add(Dropout(0.2))
model.add(Dense(10, activation='softmax'))

model.summary()

###
# 1）categorical_crossentropy(output, target, from_logits=False)：
# 计算输出张量和目标张量的Categorical crossentropy（类别交叉熵），目标张量与输出张量必须shape相同。
# 多分类的对数损失函数，与softmax分类器相对应的。
# 
# 2）RMSprop()
# AdaGrad算法的改进。鉴于神经网络都是非凸条件下的，RMSProp在非凸条件下结果更好，改变梯度累积为指数衰减的移动平均以丢弃遥远的过去历史。
# reference：http://blog.csdn.net/bvl10101111/article/details/72616378
# 
model.compile(loss='categorical_crossentropy',
              optimizer=RMSprop(),
              metrics=['accuracy'])

history = model.fit(x_train, y_train,
                    batch_size=batch_size,
                    epochs=epochs,
                    verbose=1,
                    validation_data=(x_test, y_test))
score = model.evaluate(x_test, y_test, verbose=0)
print('Test loss:', score[0])
print('Test accuracy:', score[1])

 

 

.\keras\keras\datasets\mnist.py - load_data()

 

# -*- coding: utf-8 -*-  
from ..utils.data_utils import get_file
import numpy as np

def load_data(path='mnist.npz'):
    """Loads the MNIST dataset.

    # Arguments
        path: path where to cache the dataset locally
            (relative to ~/.keras/datasets).

    # Returns
        Tuple of Numpy arrays: `(x_train, y_train), (x_test, y_test)`.

    # numpy.load()
    # numpy.load(file, mmap_mode=None, allow_pickle=True, fix_imports=True, encoding='ASCII')
    # 1) Load arrays or pickled objects from .npy, .npz or pickled files
    # 2) 
    # reference: https://docs.scipy.org/doc/numpy/reference/generated/numpy.load.html

    """
    path = get_file(path, origin='https://s3.amazonaws.com/img-datasets/mnist.npz')
    f = np.load(path)
    x_train, y_train = f['x_train'], f['y_train']
    x_test, y_test = f['x_test'], f['y_test']
    f.close()
    return (x_train, y_train), (x_test, y_test)

 

 

下载 mnist.npz 数据集

 

本文使用的 mnist.npz 数据集是通过一个 japan 的服务器下载得到的，在此免费分享给大家。如果下载有问题的话，可以留言哈。

 

下载链接：https://pan.baidu.com/s/1jH6uFFC 密码: dw3d

 

 

改造 mnist_mlp.py

 

方法1：

mnist_mlp.py 源码是使用如下命令获取数据集：

 

# the data, shuffled and split between train and test sets
(x_train, y_train), (x_test, y_test) = mnist.load_data()

 

调用的是 .\keras\keras\datasets\mnist.py 脚本中的 def load_data(path='mnist.npz') 函数，也就是因为网址被墙了导致不能正常运行的原因。本文事先下好了 mnist.npz 数据集，然后改动了一些代码使之正常运行。换句话说，本文使用的是“读取本地数据集”的方法，步骤如下：

1）下载好 mnist_npz 数据集，并将其放于 .\keras\examples 目录下。

2）改动后的 mnist_mlp.py 代码如下：

 

# -*- coding: utf-8 -*-  
'''Trains a simple deep NN on the MNIST dataset.

Gets to 98.40% test accuracy after 20 epochs
(there is *a lot* of margin for parameter tuning).
2 seconds per epoch on a K520 GPU.
'''

from __future__ import print_function

import keras
from keras.datasets import mnist
from keras.models import Sequential
from keras.layers import Dense, Dropout
from keras.optimizers import RMSprop

batch_size = 128
num_classes = 10
epochs = 20

# the data, shuffled and split between train and test sets 
# (x_train, y_train), (x_test, y_test) = mnist.load_data()

import numpy as np
path='./mnist.npz'
f = np.load(path)
x_train, y_train = f['x_train'], f['y_train']
x_test, y_test = f['x_test'], f['y_test']
f.close()

x_train = x_train.reshape(60000, 784).astype('float32')
x_test = x_test.reshape(10000, 784).astype('float32')
x_train /= 255
x_test /= 255
print(x_train.shape[0], 'train samples')
print(x_test.shape[0], 'test samples')

# convert class vectors to binary class matrices
# label为0~9共10个类别，keras要求格式为binary class matrices

y_train = keras.utils.to_categorical(y_train, num_classes)
y_test = keras.utils.to_categorical(y_test, num_classes)

# add by hcq-20171106
# Dense of keras is full-connection.
model = Sequential()
model.add(Dense(512, activation='relu', input_shape=(784,)))
model.add(Dropout(0.2))
model.add(Dense(512, activation='relu'))
model.add(Dropout(0.2))
model.add(Dense(num_classes, activation='softmax'))

model.summary()

model.compile(loss='categorical_crossentropy',
              optimizer=RMSprop(),
              metrics=['accuracy'])

history = model.fit(x_train, y_train,
                    batch_size=batch_size,
                    epochs=epochs,
                    verbose=1,
                    validation_data=(x_test, y_test))
score = model.evaluate(x_test, y_test, verbose=0)
print('Test loss:', score[0])
print('Test accuracy:', score[1])

 

 

运行效果如下所示：

 

60000 train samples
10000 test samples
_________________________________________________________________
Layer (type)                 Output Shape              Param #   
=================================================================
dense_1 (Dense)              (None, 512)               401920    
_________________________________________________________________
dropout_1 (Dropout)          (None, 512)               0         
_________________________________________________________________
dense_2 (Dense)              (None, 512)               262656    
_________________________________________________________________
dropout_2 (Dropout)          (None, 512)               0         
_________________________________________________________________
dense_3 (Dense)              (None, 10)                5130      
=================================================================
Total params: 669,706
Trainable params: 669,706
Non-trainable params: 0
_________________________________________________________________
Train on 60000 samples, validate on 10000 samples
Epoch 1/20
2017-11-09 23:06:16.881800: I tensorflow/core/common_runtime/gpu/gpu_device.cc:977] Creating TensorFlow device (/gpu:0) -> (device: 0, name: GeForce GTX 1080 Ti, pci bus id: 0000:01:00.0)

... ...

60000/60000 [==============================] - 1s 23us/step - loss: 0.0387 - acc: 0.9888 - val_loss: 0.0706 - val_acc: 0.9814
Epoch 8/20
60000/60000 [==============================] - 1s 23us/step - loss: 0.0341 - acc: 0.9899 - val_loss: 0.0789 - val_acc: 0.9827
Epoch 9/20
60000/60000 [==============================] - 1s 23us/step - loss: 0.0304 - acc: 0.9911 - val_loss: 0.0851 - val_acc: 0.9833
Epoch 10/20
60000/60000 [==============================] - 1s 23us/step - loss: 0.0290 - acc: 0.9918 - val_loss: 0.0867 - val_acc: 0.9818
Epoch 11/20
60000/60000 [==============================] - 1s 23us/step - loss: 0.0264 - acc: 0.9924 - val_loss: 0.0881 - val_acc: 0.9833
Epoch 12/20
60000/60000 [==============================] - 1s 23us/step - loss: 0.0261 - acc: 0.9928 - val_loss: 0.1095 - val_acc: 0.9801
Epoch 13/20
60000/60000 [==============================] - 1s 23us/step - loss: 0.0246 - acc: 0.9931 - val_loss: 0.1012 - val_acc: 0.9830
Epoch 14/20
60000/60000 [==============================] - 1s 23us/step - loss: 0.0233 - acc: 0.9935 - val_loss: 0.1116 - val_acc: 0.9812
Epoch 15/20
60000/60000 [==============================] - 1s 23us/step - loss: 0.0223 - acc: 0.9942 - val_loss: 0.1016 - val_acc: 0.9832
Epoch 16/20
60000/60000 [==============================] - 1s 23us/step - loss: 0.0214 - acc: 0.9943 - val_loss: 0.1053 - val_acc: 0.9832
Epoch 17/20
60000/60000 [==============================] - 1s 23us/step - loss: 0.0178 - acc: 0.9950 - val_loss: 0.1095 - val_acc: 0.9838
Epoch 18/20
60000/60000 [==============================] - 1s 23us/step - loss: 0.0212 - acc: 0.9949 - val_loss: 0.1158 - val_acc: 0.9822
Epoch 19/20
60000/60000 [==============================] - 1s 23us/step - loss: 0.0197 - acc: 0.9951 - val_loss: 0.1112 - val_acc: 0.9831
Epoch 20/20
60000/60000 [==============================] - 1s 23us/step - loss: 0.0203 - acc: 0.9951 - val_loss: 0.1097 - val_acc: 0.9833
Test loss: 0.109655842465
Test accuracy: 0.9833

 

 

 

方法2：参考该【博文】

 

 

Reference：

 

keras 中文文档：http://keras-cn.readthedocs.io/en/latest/

阅读源码遇到的一些TF、keras函数及问题：http://blog.csdn.net/jsliuqun/article/details/64444302

python读取mnist数据集：https://blog.mythsman.com/2016/01/25/1/