The annoying MNIST - Day 3.

Published: by

The previous work

At the end of day 2, we end up with an error when running the script build and test the model. This is the error

$ python mnist.py
...
mnist.py:29: UserWarning: Update your `Conv2D` call to the Keras 2 API: `Conv2D(32, (3, 3), activation="relu", input_shape=(1, 28, 28...)`
  model.add(Convolution2D(32, 3, 3, activation='relu', input_shape=(1,28,28)))
mnist.py:30: UserWarning: Update your `Conv2D` call to the Keras 2 API: `Conv2D(32, (3, 3), activation="relu")`
  model.add(Convolution2D(32, 3, 3, activation='relu'))
Traceback (most recent call last):
  File "mnist.py", line 35, in <module>
    model.add(Dense(128, activation='relu'))
...

Solution

The error caused by the fact that we use Keras 2.0, while the script is for 1.0 version. Let's switch to 2.0 version by replace the script with:

from __future__ import print_function
import keras
from keras.datasets import mnist
from keras.models import Sequential
from keras.layers import Dense, Dropout, Flatten
from keras.layers import Conv2D, MaxPooling2D
from keras import backend as K

batch_size = 128
num_classes = 10
epochs = 12

# input image dimensions
img_rows, img_cols = 28, 28

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

if K.image_data_format() == 'channels_first':
    x_train = x_train.reshape(x_train.shape[0], 1, img_rows, img_cols)
    x_test = x_test.reshape(x_test.shape[0], 1, img_rows, img_cols)
    input_shape = (1, img_rows, img_cols)
else:
    x_train = x_train.reshape(x_train.shape[0], img_rows, img_cols, 1)
    x_test = x_test.reshape(x_test.shape[0], img_rows, img_cols, 1)
    input_shape = (img_rows, img_cols, 1)

x_train = x_train.astype('float32')
x_test = x_test.astype('float32')
x_train /= 255
x_test /= 255
print('x_train shape:', x_train.shape)
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(Conv2D(32, kernel_size=(3, 3),
                 activation='relu',
                 input_shape=input_shape))
model.add(Conv2D(64, (3, 3), activation='relu'))
model.add(MaxPooling2D(pool_size=(2, 2)))
model.add(Dropout(0.25))
model.add(Flatten())
model.add(Dense(128, activation='relu'))
model.add(Dropout(0.5))
model.add(Dense(num_classes, activation='softmax'))

model.compile(loss=keras.losses.categorical_crossentropy,
              optimizer=keras.optimizers.Adadelta(),
              metrics=['accuracy'])

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])

The only noticeable change in here is, Convolution2D has been changed to Conv2D, and the syntax is also different.

With the new file, running the script generates this result.

Warning: Take time, be patience

$ python mnist.py
Using Theano backend.
x_train shape: (60000, 28, 28, 1)
60000 train samples
10000 test samples
Train on 60000 samples, validate on 10000 samples
Epoch 1/12
60000/60000 [==============================] - 182s 3ms/step - loss: 0.3410 - acc: 0.8954 - val_loss: 0.0859 - val_acc: 0.9731
Epoch 2/12
60000/60000 [==============================] - 198s 3ms/step - loss: 0.1123 - acc: 0.9668 - val_loss: 0.0528 - val_acc: 0.9832
Epoch 3/12
60000/60000 [==============================] - 209s 3ms/step - loss: 0.0869 - acc: 0.9747 - val_loss: 0.0427 - val_acc: 0.9844
Epoch 4/12
60000/60000 [==============================] - 222s 4ms/step - loss: 0.0735 - acc: 0.9785 - val_loss: 0.0369 - val_acc: 0.9866
Epoch 5/12
60000/60000 [==============================] - 194s 3ms/step - loss: 0.0628 - acc: 0.9810 - val_loss: 0.0371 - val_acc: 0.9877
Epoch 6/12
60000/60000 [==============================] - 175s 3ms/step - loss: 0.0570 - acc: 0.9834 - val_loss: 0.0334 - val_acc: 0.9896
Epoch 7/12
60000/60000 [==============================] - 177s 3ms/step - loss: 0.0498 - acc: 0.9845 - val_loss: 0.0327 - val_acc: 0.9887
Epoch 8/12
60000/60000 [==============================] - 194s 3ms/step - loss: 0.0486 - acc: 0.9859 - val_loss: 0.0307 - val_acc: 0.9896
Epoch 9/12
60000/60000 [==============================] - 191s 3ms/step - loss: 0.0445 - acc: 0.9866 - val_loss: 0.0281 - val_acc: 0.9908
Epoch 10/12
60000/60000 [==============================] - 190s 3ms/step - loss: 0.0419 - acc: 0.9877 - val_loss: 0.0302 - val_acc: 0.9898
Epoch 11/12
60000/60000 [==============================] - 185s 3ms/step - loss: 0.0404 - acc: 0.9878 - val_loss: 0.0281 - val_acc: 0.9911
Epoch 12/12
60000/60000 [==============================] - 192s 3ms/step - loss: 0.0371 - acc: 0.9887 - val_loss: 0.0327 - val_acc: 0.9893
Test loss: 0.0326723895625
Test accuracy: 0.9893

So after 12 runs, it reaches the accuracy ~99%

That is it

So this is how the training run, I got the idea behind that by looking at the code. Correct me if I am wrong:

  • mnist.load_data shuffled and split the samples into a training set and a testing set at 6:1
  • We create the model by adding layers provided by Keras, and compile it
  • We keep running with the training set and confirm the accuracy of the test set.

Nothing special got discovered at this moment.

It has been more than 1 hour, and I am out of time for today. Let dig into the model next time