In [1]:

    

# http://ben.bolte.cc/blog/2016/keras-resnet.html


    

In [2]:

    

from __future__ import absolute_import, division, print_function


    

In [3]:

    

%matplotlib inline 

import numpy as np
import matplotlib.pyplot as plt


    

In [4]:

    

from tensorflow.examples.tutorials.mnist import input_data
mnist = input_data.read_data_sets("../mnist-data/", one_hot=True)


    

    




Extracting ../mnist-data/train-images-idx3-ubyte.gz
Extracting ../mnist-data/train-labels-idx1-ubyte.gz
Extracting ../mnist-data/t10k-images-idx3-ubyte.gz
Extracting ../mnist-data/t10k-labels-idx1-ubyte.gz

In [5]:

    

mnist.train.images.shape


    

    
Out[5]:





(55000, 784)

In [6]:

    

plt.figure(figsize=(15,5))
for i in list(range(10)):
    plt.subplot(1, 10, i+1)
    pixels = mnist.test.images[i]
    pixels = pixels.reshape((28, 28))
    plt.imshow(pixels, cmap='gray_r')
plt.show()


    

In [7]:

    

from keras.models import Model,Sequential
from keras.layers import Input, Dense, Activation
from keras.layers import Dropout, Flatten, Reshape, merge
from keras.layers import Convolution2D


    

    




Using TensorFlow backend.

In [8]:

    

def residual_block(skip=True):
    def f(input):
        conv = Convolution2D(4,3,3,border_mode='same', activation='relu')(input)        
        res  = merge([conv,input], mode='sum')
        return Activation('relu')(res) if skip else conv
    return f


    

In [9]:

    

def resnet(skiplayers=3):
    #select inception module
    _in     = Input(shape=(784,))
    reshape = Reshape((28,28,1))(_in)
    res     = Convolution2D(4,3,3,border_mode='same')(reshape)    

    for i in range(skiplayers):
        res = residual_block()(res)
    
    flat = Flatten()(res)
    flat = Dropout(0.4)(flat)
        
    _out = Dense(10, activation='softmax')(flat)
    model = Model(_in, _out)
    
    return model


    

In [10]:

    

model = resnet(10)


    

In [11]:

    

from keras.utils.layer_utils import print_summary
print_summary(model.layers)


    

    




____________________________________________________________________________________________________
Layer (type)                     Output Shape          Param #     Connected to                     
====================================================================================================
input_1 (InputLayer)             (None, 784)           0                                            
____________________________________________________________________________________________________
reshape_1 (Reshape)              (None, 28, 28, 1)     0           input_1[0][0]                    
____________________________________________________________________________________________________
convolution2d_1 (Convolution2D)  (None, 28, 28, 4)     40          reshape_1[0][0]                  
____________________________________________________________________________________________________
convolution2d_2 (Convolution2D)  (None, 28, 28, 4)     148         convolution2d_1[0][0]            
____________________________________________________________________________________________________
merge_1 (Merge)                  (None, 28, 28, 4)     0           convolution2d_2[0][0]            
                                                                   convolution2d_1[0][0]            
____________________________________________________________________________________________________
activation_1 (Activation)        (None, 28, 28, 4)     0           merge_1[0][0]                    
____________________________________________________________________________________________________
convolution2d_3 (Convolution2D)  (None, 28, 28, 4)     148         activation_1[0][0]               
____________________________________________________________________________________________________
merge_2 (Merge)                  (None, 28, 28, 4)     0           convolution2d_3[0][0]            
                                                                   activation_1[0][0]               
____________________________________________________________________________________________________
activation_2 (Activation)        (None, 28, 28, 4)     0           merge_2[0][0]                    
____________________________________________________________________________________________________
convolution2d_4 (Convolution2D)  (None, 28, 28, 4)     148         activation_2[0][0]               
____________________________________________________________________________________________________
merge_3 (Merge)                  (None, 28, 28, 4)     0           convolution2d_4[0][0]            
                                                                   activation_2[0][0]               
____________________________________________________________________________________________________
activation_3 (Activation)        (None, 28, 28, 4)     0           merge_3[0][0]                    
____________________________________________________________________________________________________
convolution2d_5 (Convolution2D)  (None, 28, 28, 4)     148         activation_3[0][0]               
____________________________________________________________________________________________________
merge_4 (Merge)                  (None, 28, 28, 4)     0           convolution2d_5[0][0]            
                                                                   activation_3[0][0]               
____________________________________________________________________________________________________
activation_4 (Activation)        (None, 28, 28, 4)     0           merge_4[0][0]                    
____________________________________________________________________________________________________
convolution2d_6 (Convolution2D)  (None, 28, 28, 4)     148         activation_4[0][0]               
____________________________________________________________________________________________________
merge_5 (Merge)                  (None, 28, 28, 4)     0           convolution2d_6[0][0]            
                                                                   activation_4[0][0]               
____________________________________________________________________________________________________
activation_5 (Activation)        (None, 28, 28, 4)     0           merge_5[0][0]                    
____________________________________________________________________________________________________
convolution2d_7 (Convolution2D)  (None, 28, 28, 4)     148         activation_5[0][0]               
____________________________________________________________________________________________________
merge_6 (Merge)                  (None, 28, 28, 4)     0           convolution2d_7[0][0]            
                                                                   activation_5[0][0]               
____________________________________________________________________________________________________
activation_6 (Activation)        (None, 28, 28, 4)     0           merge_6[0][0]                    
____________________________________________________________________________________________________
convolution2d_8 (Convolution2D)  (None, 28, 28, 4)     148         activation_6[0][0]               
____________________________________________________________________________________________________
merge_7 (Merge)                  (None, 28, 28, 4)     0           convolution2d_8[0][0]            
                                                                   activation_6[0][0]               
____________________________________________________________________________________________________
activation_7 (Activation)        (None, 28, 28, 4)     0           merge_7[0][0]                    
____________________________________________________________________________________________________
convolution2d_9 (Convolution2D)  (None, 28, 28, 4)     148         activation_7[0][0]               
____________________________________________________________________________________________________
merge_8 (Merge)                  (None, 28, 28, 4)     0           convolution2d_9[0][0]            
                                                                   activation_7[0][0]               
____________________________________________________________________________________________________
activation_8 (Activation)        (None, 28, 28, 4)     0           merge_8[0][0]                    
____________________________________________________________________________________________________
convolution2d_10 (Convolution2D) (None, 28, 28, 4)     148         activation_8[0][0]               
____________________________________________________________________________________________________
merge_9 (Merge)                  (None, 28, 28, 4)     0           convolution2d_10[0][0]           
                                                                   activation_8[0][0]               
____________________________________________________________________________________________________
activation_9 (Activation)        (None, 28, 28, 4)     0           merge_9[0][0]                    
____________________________________________________________________________________________________
convolution2d_11 (Convolution2D) (None, 28, 28, 4)     148         activation_9[0][0]               
____________________________________________________________________________________________________
merge_10 (Merge)                 (None, 28, 28, 4)     0           convolution2d_11[0][0]           
                                                                   activation_9[0][0]               
____________________________________________________________________________________________________
activation_10 (Activation)       (None, 28, 28, 4)     0           merge_10[0][0]                   
____________________________________________________________________________________________________
flatten_1 (Flatten)              (None, 3136)          0           activation_10[0][0]              
____________________________________________________________________________________________________
dropout_1 (Dropout)              (None, 3136)          0           flatten_1[0][0]                  
____________________________________________________________________________________________________
dense_1 (Dense)                  (None, 10)            31370       dropout_1[0][0]                  
====================================================================================================
Total params: 32,890
Trainable params: 32,890
Non-trainable params: 0
____________________________________________________________________________________________________

In [ ]:

    

from keras.optimizers import Adam
model.compile(loss='categorical_crossentropy', optimizer=Adam(lr=0.01), metrics=["accuracy"])

model.fit(mnist.train.images, mnist.train.labels,
          batch_size=128, nb_epoch=10, verbose=1, 
          validation_data=(mnist.test.images, mnist.test.labels))


    

    




Train on 55000 samples, validate on 10000 samples
Epoch 1/10
55000/55000 [==============================] - 71s - loss: 0.5071 - acc: 0.8644 - val_loss: 0.1356 - val_acc: 0.9574
Epoch 2/10
55000/55000 [==============================] - 76s - loss: 0.1575 - acc: 0.9515 - val_loss: 0.1074 - val_acc: 0.9674
Epoch 3/10
51968/55000 [===========================>..] - ETA: 4s - loss: 0.1329 - acc: 0.9587 

In [ ]: