CNN

# sample train: 60000 test: 10000, 28 x 28  (nb_samples, 28, 28)

The train dataset ， X_train == image data ， y_train == label number， equal as X&y_test

print(‘sample train : ‘,  len(X_train))      # 60000

print(‘test : ‘,  len(X_test))           # 10000

np_utils.to_categorical to run One-hot encoding ， let 0~9 Classification use One-hot encoding to trans 10 binary features ， ex: 4 = 0000100000 , 0 = 1000000000

This slideshow requires JavaScript.

Train on 40199 samples，validate on 19801 samples

40 minutes of Epoch 10/10 ， RESULT :    dense_1= 128 neuron   ， dense_2= 10 neuron

Hidden Layer dense_1 =  784 x 128 + 128  100480

Outout Layer dense_2 = 1290 = 128 x 10 + 10

Total = 100480 +1290 = 101770

This slideshow requires JavaScript.

ACC / LOSS

This slideshow requires JavaScript.

EarlyStopping(monitor='val_loss', patience=3,

This slideshow requires JavaScript.

binary image  —->   Convolutional  —–> ReLU

This slideshow requires JavaScript.

Pooling   —–>   Max-Pooling

This slideshow requires JavaScript.

Flatten  —-> Full connected                factor :  Memory 、CPU、GPU