In [1]:

    

from theano.sandbox import cuda
import utils; reload(utils)
from utils import *


    

    




WARNING (theano.sandbox.cuda): The cuda backend is deprecated and will be removed in the next release (v0.10).  Please switch to the gpuarray backend. You can get more information about how to switch at this URL:
 https://github.com/Theano/Theano/wiki/Converting-to-the-new-gpu-back-end%28gpuarray%29

Using gpu device 0: Tesla K80 (CNMeM is disabled, cuDNN 5103)
Using Theano backend.

In [2]:

    

vgg = Vgg16()


    

In [3]:

    

vgg.model.pop()
for layer in vgg.model.layers:
    layer.trainable=False
vgg.model.add(Dense(2, activation = 'softmax'))
vgg.compile()


    

In [4]:

    

model = vgg.model


    

In [5]:

    

# load previous weights
model.load_weights('data/dogscats/models/finetune3.h5')


    

In [6]:

    

layers = model.layers
[index for index,layer in enumerate(layers) if type(layer) is Convolution2D]


    

    
Out[6]:





[2, 4, 7, 9, 12, 14, 16, 19, 21, 23, 26, 28, 30]

In [7]:

    

conv_layers = layers[:32]
conv_model = Sequential(conv_layers)


    

In [8]:

    

conv_layers


    

    
Out[8]:





[<keras.layers.core.Lambda at 0x7f696fdcaa50>,
 <keras.layers.convolutional.ZeroPadding2D at 0x7f696fccb190>,
 <keras.layers.convolutional.Convolution2D at 0x7f696fc9c350>,
 <keras.layers.convolutional.ZeroPadding2D at 0x7f696fc5c310>,
 <keras.layers.convolutional.Convolution2D at 0x7f696fbf0a50>,
 <keras.layers.pooling.MaxPooling2D at 0x7f696fccb150>,
 <keras.layers.convolutional.ZeroPadding2D at 0x7f696fbbc310>,
 <keras.layers.convolutional.Convolution2D at 0x7f696fbc7c90>,
 <keras.layers.convolutional.ZeroPadding2D at 0x7f696fbd7cd0>,
 <keras.layers.convolutional.Convolution2D at 0x7f696fb74190>,
 <keras.layers.pooling.MaxPooling2D at 0x7f696fbbc2d0>,
 <keras.layers.convolutional.ZeroPadding2D at 0x7f696fb799d0>,
 <keras.layers.convolutional.Convolution2D at 0x7f696fb94c10>,
 <keras.layers.convolutional.ZeroPadding2D at 0x7f696fb98bd0>,
 <keras.layers.convolutional.Convolution2D at 0x7f696fb36190>,
 <keras.layers.convolutional.ZeroPadding2D at 0x7f696fb401d0>,
 <keras.layers.convolutional.Convolution2D at 0x7f696fb57910>,
 <keras.layers.pooling.MaxPooling2D at 0x7f696fb79990>,
 <keras.layers.convolutional.ZeroPadding2D at 0x7f696fb5cd50>,
 <keras.layers.convolutional.Convolution2D at 0x7f696fafa210>,
 <keras.layers.convolutional.ZeroPadding2D at 0x7f696faff110>,
 <keras.layers.convolutional.Convolution2D at 0x7f696fb16750>,
 <keras.layers.convolutional.ZeroPadding2D at 0x7f696faa1790>,
 <keras.layers.convolutional.Convolution2D at 0x7f696faac050>,
 <keras.layers.pooling.MaxPooling2D at 0x7f696fb5cd10>,
 <keras.layers.convolutional.ZeroPadding2D at 0x7f696fac5250>,
 <keras.layers.convolutional.Convolution2D at 0x7f696fada990>,
 <keras.layers.convolutional.ZeroPadding2D at 0x7f696fa60a50>,
 <keras.layers.convolutional.Convolution2D at 0x7f696fa75a50>,
 <keras.layers.convolutional.ZeroPadding2D at 0x7f696fa7be10>,
 <keras.layers.convolutional.Convolution2D at 0x7f696fa92f90>,
 <keras.layers.pooling.MaxPooling2D at 0x7f696fac5210>]

In [9]:

    

# fully connected layers
fc_layers = layers[32:]


    

In [10]:

    

fc_layers


    

    
Out[10]:





[<keras.layers.core.Flatten at 0x7f696fa26510>,
 <keras.layers.core.Dense at 0x7f696f9ba990>,
 <keras.layers.core.Dropout at 0x7f696f9ca590>,
 <keras.layers.core.Dense at 0x7f696f9455d0>,
 <keras.layers.core.Dropout at 0x7f696f949290>,
 <keras.layers.core.Dense at 0x7f696f6bd4d0>]

In [11]:

    

train_batches = get_batches('data/dogscats/train', shuffle=False, batch_size=64)
val_batches = get_batches('data/dogscats/valid', shuffle=False, batch_size=64)


    

    




Found 23000 images belonging to 2 classes.
Found 2000 images belonging to 2 classes.

In [12]:

    

val_classes = val_batches.classes
train_classes = train_batches.classes


    

In [13]:

    

val_classes


    

    
Out[13]:





array([0, 0, 0, ..., 1, 1, 1], dtype=int32)

In [14]:

    

train_classes


    

    
Out[14]:





array([0, 0, 0, ..., 1, 1, 1], dtype=int32)

In [15]:

    

val_labels = onehot(val_classes)
train_labels = onehot(train_classes)


    

In [16]:

    

val_labels


    

    
Out[16]:





array([[ 1.,  0.],
       [ 1.,  0.],
       [ 1.,  0.],
       ..., 
       [ 0.,  1.],
       [ 0.,  1.],
       [ 0.,  1.]])

In [17]:

    

train_labels


    

    
Out[17]:





array([[ 1.,  0.],
       [ 1.,  0.],
       [ 1.,  0.],
       ..., 
       [ 0.,  1.],
       [ 0.,  1.],
       [ 0.,  1.]])

In [18]:

    

train_precalc = conv_model.predict_generator(train_batches, train_batches.nb_sample)


    

In [19]:

    

val_precalc = conv_model.predict_generator(val_batches, val_batches.nb_sample)


    

In [20]:

    

model.summary()


    

    




____________________________________________________________________________________________________
Layer (type)                     Output Shape          Param #     Connected to                     
====================================================================================================
lambda_1 (Lambda)                (None, 3, 224, 224)   0           lambda_input_1[0][0]             
____________________________________________________________________________________________________
zeropadding2d_1 (ZeroPadding2D)  (None, 3, 226, 226)   0           lambda_1[0][0]                   
____________________________________________________________________________________________________
convolution2d_1 (Convolution2D)  (None, 64, 224, 224)  1792        zeropadding2d_1[0][0]            
____________________________________________________________________________________________________
zeropadding2d_2 (ZeroPadding2D)  (None, 64, 226, 226)  0           convolution2d_1[0][0]            
____________________________________________________________________________________________________
convolution2d_2 (Convolution2D)  (None, 64, 224, 224)  36928       zeropadding2d_2[0][0]            
____________________________________________________________________________________________________
maxpooling2d_1 (MaxPooling2D)    (None, 64, 112, 112)  0           convolution2d_2[0][0]            
____________________________________________________________________________________________________
zeropadding2d_3 (ZeroPadding2D)  (None, 64, 114, 114)  0           maxpooling2d_1[0][0]             
____________________________________________________________________________________________________
convolution2d_3 (Convolution2D)  (None, 128, 112, 112) 73856       zeropadding2d_3[0][0]            
____________________________________________________________________________________________________
zeropadding2d_4 (ZeroPadding2D)  (None, 128, 114, 114) 0           convolution2d_3[0][0]            
____________________________________________________________________________________________________
convolution2d_4 (Convolution2D)  (None, 128, 112, 112) 147584      zeropadding2d_4[0][0]            
____________________________________________________________________________________________________
maxpooling2d_2 (MaxPooling2D)    (None, 128, 56, 56)   0           convolution2d_4[0][0]            
____________________________________________________________________________________________________
zeropadding2d_5 (ZeroPadding2D)  (None, 128, 58, 58)   0           maxpooling2d_2[0][0]             
____________________________________________________________________________________________________
convolution2d_5 (Convolution2D)  (None, 256, 56, 56)   295168      zeropadding2d_5[0][0]            
____________________________________________________________________________________________________
zeropadding2d_6 (ZeroPadding2D)  (None, 256, 58, 58)   0           convolution2d_5[0][0]            
____________________________________________________________________________________________________
convolution2d_6 (Convolution2D)  (None, 256, 56, 56)   590080      zeropadding2d_6[0][0]            
____________________________________________________________________________________________________
zeropadding2d_7 (ZeroPadding2D)  (None, 256, 58, 58)   0           convolution2d_6[0][0]            
____________________________________________________________________________________________________
convolution2d_7 (Convolution2D)  (None, 256, 56, 56)   590080      zeropadding2d_7[0][0]            
____________________________________________________________________________________________________
maxpooling2d_3 (MaxPooling2D)    (None, 256, 28, 28)   0           convolution2d_7[0][0]            
____________________________________________________________________________________________________
zeropadding2d_8 (ZeroPadding2D)  (None, 256, 30, 30)   0           maxpooling2d_3[0][0]             
____________________________________________________________________________________________________
convolution2d_8 (Convolution2D)  (None, 512, 28, 28)   1180160     zeropadding2d_8[0][0]            
____________________________________________________________________________________________________
zeropadding2d_9 (ZeroPadding2D)  (None, 512, 30, 30)   0           convolution2d_8[0][0]            
____________________________________________________________________________________________________
convolution2d_9 (Convolution2D)  (None, 512, 28, 28)   2359808     zeropadding2d_9[0][0]            
____________________________________________________________________________________________________
zeropadding2d_10 (ZeroPadding2D) (None, 512, 30, 30)   0           convolution2d_9[0][0]            
____________________________________________________________________________________________________
convolution2d_10 (Convolution2D) (None, 512, 28, 28)   2359808     zeropadding2d_10[0][0]           
____________________________________________________________________________________________________
maxpooling2d_4 (MaxPooling2D)    (None, 512, 14, 14)   0           convolution2d_10[0][0]           
____________________________________________________________________________________________________
zeropadding2d_11 (ZeroPadding2D) (None, 512, 16, 16)   0           maxpooling2d_4[0][0]             
____________________________________________________________________________________________________
convolution2d_11 (Convolution2D) (None, 512, 14, 14)   2359808     zeropadding2d_11[0][0]           
____________________________________________________________________________________________________
zeropadding2d_12 (ZeroPadding2D) (None, 512, 16, 16)   0           convolution2d_11[0][0]           
____________________________________________________________________________________________________
convolution2d_12 (Convolution2D) (None, 512, 14, 14)   2359808     zeropadding2d_12[0][0]           
____________________________________________________________________________________________________
zeropadding2d_13 (ZeroPadding2D) (None, 512, 16, 16)   0           convolution2d_12[0][0]           
____________________________________________________________________________________________________
convolution2d_13 (Convolution2D) (None, 512, 14, 14)   2359808     zeropadding2d_13[0][0]           
____________________________________________________________________________________________________
maxpooling2d_5 (MaxPooling2D)    (None, 512, 7, 7)     0           convolution2d_13[0][0]           
____________________________________________________________________________________________________
flatten_1 (Flatten)              (None, 25088)         0           maxpooling2d_5[0][0]             
____________________________________________________________________________________________________
dense_1 (Dense)                  (None, 4096)          102764544   flatten_1[0][0]                  
____________________________________________________________________________________________________
dropout_1 (Dropout)              (None, 4096)          0           dense_1[0][0]                    
____________________________________________________________________________________________________
dense_2 (Dense)                  (None, 4096)          16781312    dropout_1[0][0]                  
____________________________________________________________________________________________________
dropout_2 (Dropout)              (None, 4096)          0           dense_2[0][0]                    
____________________________________________________________________________________________________
dense_4 (Dense)                  (None, 2)             8194        dropout_2[0][0]                  
====================================================================================================
Total params: 134,268,738
Trainable params: 8,194
Non-trainable params: 134,260,544
____________________________________________________________________________________________________

In [21]:

    

conv_layers[-1].output_shape


    

    
Out[21]:





(None, 512, 7, 7)

In [22]:

    

fc_model = Sequential([
    Flatten(input_shape=(512, 7, 7)),
    Dense(4096, activation = 'relu'),
    Dropout(0.),
    Dense(4096, activation = 'relu'),
    Dropout(0.),
    Dense(2, activation = 'softmax')
])


    

In [23]:

    

# copy the weights from the pre-trained model
# since we're removing dropout, we want to half the weights
fc_model.layers[0].set_weights([x/2 for x in model.layers[32].get_weights()])


    

In [24]:

    

fc_model.layers[1].set_weights([x/2 for x in model.layers[33].get_weights()])
fc_model.layers[2].set_weights([x/2 for x in model.layers[34].get_weights()])
fc_model.layers[3].set_weights([x/2 for x in model.layers[35].get_weights()])
fc_model.layers[4].set_weights([x/2 for x in model.layers[36].get_weights()])
fc_model.layers[5].set_weights([x/2 for x in model.layers[37].get_weights()])


    

In [25]:

    

fc_model.compile(optimizer=RMSprop(lr=0.00001,rho=0.7), # since this is a finely tuned model, it needs to be updated very slowly
                loss='categorical_crossentropy',
                metrics=['accuracy'])


    

In [26]:

    

fc_model.fit(train_precalc, train_labels, nb_epoch=8, verbose=2, batch_size=64, validation_data = (val_precalc, val_labels))


    

    




Train on 23000 samples, validate on 2000 samples
Epoch 1/8
31s - loss: 0.2471 - acc: 0.9823 - val_loss: 0.2504 - val_acc: 0.9815
Epoch 2/8
31s - loss: 0.2131 - acc: 0.9852 - val_loss: 0.2648 - val_acc: 0.9815
Epoch 3/8
31s - loss: 0.1610 - acc: 0.9888 - val_loss: 0.2218 - val_acc: 0.9850
Epoch 4/8
31s - loss: 0.1348 - acc: 0.9906 - val_loss: 0.2068 - val_acc: 0.9855
Epoch 5/8
31s - loss: 0.1221 - acc: 0.9916 - val_loss: 0.3051 - val_acc: 0.9800
Epoch 6/8
31s - loss: 0.1036 - acc: 0.9931 - val_loss: 0.2132 - val_acc: 0.9860
Epoch 7/8
31s - loss: 0.0979 - acc: 0.9934 - val_loss: 0.2226 - val_acc: 0.9845
Epoch 8/8
31s - loss: 0.0834 - acc: 0.9946 - val_loss: 0.2074 - val_acc: 0.9855






    
Out[26]:





<keras.callbacks.History at 0x7fa730c4e810>

In [28]:

    

%matplotlib inline
plt.plot(range(1,9), [0.2471, 0.2131, 0.1610, 0.1348, 0.1221, 0.1036, 0.0979, 0.0834],
        range(1,9), [0.2504, 0.2648, 0.2218, 0.2068, 0.3051, 0.2132, 0.2226, 0.2074])


    

    
Out[28]:





[<matplotlib.lines.Line2D at 0x7fa730b10990>,
 <matplotlib.lines.Line2D at 0x7fa730b10a90>]






    

In [29]:

    

# We are certainly in the overfitting zone


    

In [31]:

    

gen = image.ImageDataGenerator(rotation_range = 15, width_shift_range = 0.1, height_shift_range=0.1, zoom_range=0.1, horizontal_flip=True)


    

In [32]:

    

train_batches = get_batches('data/dogscats/train', gen, batch_size=64)
val_batches = get_batches('data/dogscats/valid', gen, batch_size=64)


    

    




Found 23000 images belonging to 2 classes.
Found 2000 images belonging to 2 classes.

In [33]:

    

for layer in conv_model.layers:
    layer.trainable=False


    

In [34]:

    

conv_model.add(Sequential([
    Flatten(input_shape=(512, 7, 7)),
    Dense(4096, activation = 'relu'),
    Dropout(0.),
    Dense(4096, activation = 'relu'),
    Dropout(0.),
    Dense(2, activation = 'softmax')
]))


    

In [39]:

    

conv_model.compile(optimizer=RMSprop(lr=0.00001,rho=0.7),
                loss='categorical_crossentropy',
                metrics=['accuracy'])


    

In [ ]:

    

fit_hist = conv_model.fit_generator(train_batches, samples_per_epoch=train_batches.nb_sample, nb_epoch=2,verbose=2, validation_data=val_batches, nb_val_samples=val_batches.nb_sample)


    

    




Epoch 1/2
646s - loss: 0.1973 - acc: 0.9773 - val_loss: 0.2936 - val_acc: 0.9725
Epoch 2/2
644s - loss: 0.1936 - acc: 0.9790 - val_loss: 0.2559 - val_acc: 0.9715

In [18]:

    

bn_model = Sequential([
    Flatten(input_shape = (512, 7, 7)),
    Dense(4096, activation = 'relu'),
    BatchNormalization(),
    Dropout(0.6),
    Dense(4096, activation = 'relu'),
    BatchNormalization(),
    Dropout(0.6),
    Dense(2, activation = 'softmax')
])


    

In [19]:

    

# load weights for the model from the dense layers of the Vgg16BN model
from vgg16bn import Vgg16BN


    

In [20]:

    

vgg16_bn = Vgg16BN()


    

    




Downloading data from http://files.fast.ai/models/vgg16_bn.h5

In [24]:

    

vgg16_bn.model.summary()


    

    




____________________________________________________________________________________________________
Layer (type)                     Output Shape          Param #     Connected to                     
====================================================================================================
lambda_2 (Lambda)                (None, 3, 224, 224)   0           lambda_input_2[0][0]             
____________________________________________________________________________________________________
zeropadding2d_14 (ZeroPadding2D) (None, 3, 226, 226)   0           lambda_2[0][0]                   
____________________________________________________________________________________________________
convolution2d_14 (Convolution2D) (None, 64, 224, 224)  1792        zeropadding2d_14[0][0]           
____________________________________________________________________________________________________
zeropadding2d_15 (ZeroPadding2D) (None, 64, 226, 226)  0           convolution2d_14[0][0]           
____________________________________________________________________________________________________
convolution2d_15 (Convolution2D) (None, 64, 224, 224)  36928       zeropadding2d_15[0][0]           
____________________________________________________________________________________________________
maxpooling2d_6 (MaxPooling2D)    (None, 64, 112, 112)  0           convolution2d_15[0][0]           
____________________________________________________________________________________________________
zeropadding2d_16 (ZeroPadding2D) (None, 64, 114, 114)  0           maxpooling2d_6[0][0]             
____________________________________________________________________________________________________
convolution2d_16 (Convolution2D) (None, 128, 112, 112) 73856       zeropadding2d_16[0][0]           
____________________________________________________________________________________________________
zeropadding2d_17 (ZeroPadding2D) (None, 128, 114, 114) 0           convolution2d_16[0][0]           
____________________________________________________________________________________________________
convolution2d_17 (Convolution2D) (None, 128, 112, 112) 147584      zeropadding2d_17[0][0]           
____________________________________________________________________________________________________
maxpooling2d_7 (MaxPooling2D)    (None, 128, 56, 56)   0           convolution2d_17[0][0]           
____________________________________________________________________________________________________
zeropadding2d_18 (ZeroPadding2D) (None, 128, 58, 58)   0           maxpooling2d_7[0][0]             
____________________________________________________________________________________________________
convolution2d_18 (Convolution2D) (None, 256, 56, 56)   295168      zeropadding2d_18[0][0]           
____________________________________________________________________________________________________
zeropadding2d_19 (ZeroPadding2D) (None, 256, 58, 58)   0           convolution2d_18[0][0]           
____________________________________________________________________________________________________
convolution2d_19 (Convolution2D) (None, 256, 56, 56)   590080      zeropadding2d_19[0][0]           
____________________________________________________________________________________________________
zeropadding2d_20 (ZeroPadding2D) (None, 256, 58, 58)   0           convolution2d_19[0][0]           
____________________________________________________________________________________________________
convolution2d_20 (Convolution2D) (None, 256, 56, 56)   590080      zeropadding2d_20[0][0]           
____________________________________________________________________________________________________
maxpooling2d_8 (MaxPooling2D)    (None, 256, 28, 28)   0           convolution2d_20[0][0]           
____________________________________________________________________________________________________
zeropadding2d_21 (ZeroPadding2D) (None, 256, 30, 30)   0           maxpooling2d_8[0][0]             
____________________________________________________________________________________________________
convolution2d_21 (Convolution2D) (None, 512, 28, 28)   1180160     zeropadding2d_21[0][0]           
____________________________________________________________________________________________________
zeropadding2d_22 (ZeroPadding2D) (None, 512, 30, 30)   0           convolution2d_21[0][0]           
____________________________________________________________________________________________________
convolution2d_22 (Convolution2D) (None, 512, 28, 28)   2359808     zeropadding2d_22[0][0]           
____________________________________________________________________________________________________
zeropadding2d_23 (ZeroPadding2D) (None, 512, 30, 30)   0           convolution2d_22[0][0]           
____________________________________________________________________________________________________
convolution2d_23 (Convolution2D) (None, 512, 28, 28)   2359808     zeropadding2d_23[0][0]           
____________________________________________________________________________________________________
maxpooling2d_9 (MaxPooling2D)    (None, 512, 14, 14)   0           convolution2d_23[0][0]           
____________________________________________________________________________________________________
zeropadding2d_24 (ZeroPadding2D) (None, 512, 16, 16)   0           maxpooling2d_9[0][0]             
____________________________________________________________________________________________________
convolution2d_24 (Convolution2D) (None, 512, 14, 14)   2359808     zeropadding2d_24[0][0]           
____________________________________________________________________________________________________
zeropadding2d_25 (ZeroPadding2D) (None, 512, 16, 16)   0           convolution2d_24[0][0]           
____________________________________________________________________________________________________
convolution2d_25 (Convolution2D) (None, 512, 14, 14)   2359808     zeropadding2d_25[0][0]           
____________________________________________________________________________________________________
zeropadding2d_26 (ZeroPadding2D) (None, 512, 16, 16)   0           convolution2d_25[0][0]           
____________________________________________________________________________________________________
convolution2d_26 (Convolution2D) (None, 512, 14, 14)   2359808     zeropadding2d_26[0][0]           
____________________________________________________________________________________________________
maxpooling2d_10 (MaxPooling2D)   (None, 512, 7, 7)     0           convolution2d_26[0][0]           
____________________________________________________________________________________________________
flatten_3 (Flatten)              (None, 25088)         0           maxpooling2d_10[0][0]            
____________________________________________________________________________________________________
dense_8 (Dense)                  (None, 4096)          102764544   flatten_3[0][0]                  
____________________________________________________________________________________________________
batchnormalization_3 (BatchNorma (None, 4096)          16384       dense_8[0][0]                    
____________________________________________________________________________________________________
dropout_5 (Dropout)              (None, 4096)          0           batchnormalization_3[0][0]       
____________________________________________________________________________________________________
dense_9 (Dense)                  (None, 4096)          16781312    dropout_5[0][0]                  
____________________________________________________________________________________________________
batchnormalization_4 (BatchNorma (None, 4096)          16384       dense_9[0][0]                    
____________________________________________________________________________________________________
dropout_6 (Dropout)              (None, 4096)          0           batchnormalization_4[0][0]       
____________________________________________________________________________________________________
dense_10 (Dense)                 (None, 1000)          4097000     dropout_6[0][0]                  
====================================================================================================
Total params: 138,390,312
Trainable params: 138,373,928
Non-trainable params: 16,384
____________________________________________________________________________________________________

In [27]:

    

bn_model.layers[0]


    

    
Out[27]:





<keras.layers.core.Flatten at 0x7f696f4f9410>

In [28]:

    

vgg16_bn.model.layers[32]


    

    
Out[28]:





<keras.layers.core.Flatten at 0x7f696f11f910>

In [29]:

    

bn_model.layers[0].set_weights(vgg16_bn.model.layers[32].get_weights())
bn_model.layers[1].set_weights(vgg16_bn.model.layers[33].get_weights())
bn_model.layers[2].set_weights(vgg16_bn.model.layers[34].get_weights())
bn_model.layers[3].set_weights(vgg16_bn.model.layers[35].get_weights())
bn_model.layers[4].set_weights(vgg16_bn.model.layers[36].get_weights())
bn_model.layers[5].set_weights(vgg16_bn.model.layers[37].get_weights())
bn_model.layers[6].set_weights(vgg16_bn.model.layers[38].get_weights())


    

In [30]:

    

# rescaling because the dropout percentages are different for the loaded weights
# only changing for dense layers
bn_model.layers[1].set_weights([x*(1-0.5)/(1-0.6) for x in bn_model.layers[1].get_weights()])
bn_model.layers[4].set_weights([x*(1-0.5)/(1-0.6) for x in bn_model.layers[4].get_weights()])


    

In [31]:

    

bn_model.compile(Adam(), 'categorical_crossentropy', metrics=['accuracy'])


    

In [ ]: