In [1]:

    

%matplotlib inline
import numpy as np

from keras.models import model_from_json
from skimage.io import imread, imshow
from skimage.transform import resize


    

    




Using TensorFlow backend.






    




Couldn't import dot_parser, loading of dot files will not be possible.

In [2]:

    

json_file = open('model.json','r')
model_json = json_file.read()
json_file.close()


    

In [3]:

    

model = model_from_json(model_json)
model.load_weights("model.h5")
print("Loaded Model from disk")


    

    




Loaded Model from disk

In [4]:

    

model.summary()


    

    




_________________________________________________________________
Layer (type)                 Output Shape              Param #   
=================================================================
conv2d_3 (Conv2D)            (None, 26, 26, 32)        320       
_________________________________________________________________
conv2d_4 (Conv2D)            (None, 24, 24, 65)        18785     
_________________________________________________________________
max_pooling2d_2 (MaxPooling2 (None, 12, 12, 65)        0         
_________________________________________________________________
dropout_3 (Dropout)          (None, 12, 12, 65)        0         
_________________________________________________________________
flatten_2 (Flatten)          (None, 9360)              0         
_________________________________________________________________
dense_3 (Dense)              (None, 128)               1198208   
_________________________________________________________________
dropout_4 (Dropout)          (None, 128)               0         
_________________________________________________________________
dense_4 (Dense)              (None, 10)                1290      
=================================================================
Total params: 1,218,603
Trainable params: 1,218,603
Non-trainable params: 0
_________________________________________________________________

In [5]:

    

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


    

In [6]:

    

x = imread('test.png')
x = np.invert(x)
x = resize(x,(28,28))
imshow(x[:,:,1])


    

    
Out[6]:





<matplotlib.image.AxesImage at 0x7f1b011cdcd0>






    

In [7]:

    

x = x[:,:,1].reshape(1,28,28,1)


    

In [8]:

    

out = model.predict(x)
print(out)

print(np.argmax(out, axis=1))


    

    




[[  1.41521161e-06   1.43153528e-07   1.86247972e-03   2.02288142e-09
    2.44262683e-05   9.36024368e-01   1.17116736e-03   1.01040439e-06
    5.99916838e-02   9.23346379e-04]]
[5]

In [ ]: