mnist.read_data_sets() Is Deprecated

Neural Network Tutorials - Herong's Tutorial Examples

∟mnist.read_data_sets() Is Deprecated

This section provides a tutorial example on how to use tensorflow.keras.datasets.mnist module instead of the tensorflow.examples.tutorials.mnist module to avoid the mnist.read_data_sets() deprecation warning messages.

If you are using TensorFlow 1.14.0, you will notice that the tensorflow.examples.tutorials.mnist marked as deprecated. If you continue to run the mnist_dataset.py script, you will get the following warning messages:

herong$ python3 mnist_dataset.py

WARNING:tensorflow:From mnist_dataset.py:7: read_data_sets 
  (from tensorflow.contrib.learn.python.learn.datasets.mnist) 
  is deprecated and will be removed in a future version.
Instructions for updating:
Please use alternatives such as official/mnist/dataset.py 
  from tensorflow/models.

WARNING:tensorflow:From 
  /.../tensorflow/contrib/learn/python/learn/datasets/mnist.py:260: 
  maybe_download (from tensorflow.contrib.learn.python.learn.datasets.base) 
  is deprecated and will be removed in a future version.
Instructions for updating:
Please write your own downloading logic.

...

The warning message is clear that the tensorflow.examples.tutorials.mnist module is deprecated. But instructions for updating is useless: "Please write your own downloading logic." Why deprecating a module without a good replacement?

Searching the Internet, I found 2 options to deal with the warning messages:

1. Turn off warning messages - Using tf.compat.v1.logging.set_verbosity(tf.compat.v1.logging.ERROR) to stop TensorFlow from printing those warning messages.

Here is the revised version of my MNIST test script, mnist_dataset_no_warning.py:

#- mnist_dataset_no_warning.py
#- Copyright (c) 2019 HerongYang.com. All Rights Reserved.
#
import tensorflow.examples.tutorials.mnist.input_data as mnist

import tensorflow as tf
tf.compat.v1.logging.set_verbosity(tf.compat.v1.logging.ERROR)

# load the MNIST dataset
ds = mnist.read_data_sets("MNIST_data/")

# fetch next 10 samples from the dataset as features and target
xs, ys = ds.train.next_batch(10)

# take the first sample
x = xs[0]
y = ys[0]

# convert features back to a 28x28 grey scale image  
grey = x.reshape(28,28)

# convert grey scale to black and white
import numpy as np
bw = np.rint(grey).astype(int)

# convert grey scale to black and white
ascii = str(bw).replace(' ','').replace('0', ' ')

# show the ascii image and the target label. 
print("The input image in ascii\n"+ascii)
print("The target label: "+str(y))

2. Use tensorflow.keras.datasets.mnist module - The tensorflow.keras.datasets.mnist module offers the load_data() method to download MNIST datasets from the Internet.

Here is the revised version of my MNIST test script, mnist_dataset_keras.py:

#- mnist_dataset_keras.py
#- Copyright (c) 2019 HerongYang.com. All Rights Reserved.
#
import tensorflow.keras.datasets.mnist as mnist
import numpy as np

# load the MNIST dataset from Internet
# as numpy.ndarray of shapes x:(*, 28, 28) and y:(*,)
(x_train, y_train), (x_test, y_test) = mnist.load_data()

# normalize feature values
x_train, x_test = x_train/255.0, x_test/255.0

# fetch next 10 samples from the training set.  
max = x_train.shape[0];
idx = np.random.randint(max, size=10)
xs = x_train.take(idx, 0)
ys = y_train.take(idx, 0)

# take the first sample
x = xs[0]
y = ys[0]

# convert grey scale to black and white
bw = np.rint(x).astype(int)

# convert grey scale to black and white
ascii = str(bw).replace(' ','').replace('0', ' ')

# show the ascii image and the target label. 
print("The input image in ascii\n"+ascii)
print("The target label: "+str(y))

Run the revised version. I see no issues.

herong$ python3 mnist_dataset_keras.py

The input image in ascii
[[                            ]
 [                            ]
 [                            ]
 [                            ]
 [                            ]
 [                            ]
 [                            ]
 [        1111    1 1111      ]
 [        111111111111111     ]
 [       1111111111111111     ]
 [      1111111111111111      ]
 [      1111111    11111      ]
 [     11111       11111      ]
 [     1111       11111       ]
 [     1111      11111        ]
 [     111       11111        ]
 [              11111         ]
 [             11111          ]
 [             11111          ]
 [            11111           ]
 [            11111           ]
 [           111111           ]
 [           111111           ]
 [           111111           ]
 [          11111             ]
 [          11111             ]
 [           11               ]
 [                            ]]

The target label: 7