Implementing Multitask Classification in Keras

Question

I am Trying to implement Multitask Learning in Keras with share constitutional Layers and task specific FC layers. Following is my model:

input1=Input(shape=(8,8,128))
conv1=Conv2D(128,kernel_size=3, strides=1,
                 padding='SAME', use_bias=True, 
                 activation='relu',kernel_initializer='glorot_normal')

conv2=Conv2D(128,kernel_size=3, strides=1,
                  padding='SAME', use_bias=True, 
                  activation='relu',input_shape=[8,8,128],kernel_initializer='glorot_normal')
maxpool1=MaxPooling2D(pool_size=2,strides=2)

conv3=Conv2D(64,kernel_size=3, strides=1,
                 padding='SAME', use_bias=True, 
                 activation='relu',kernel_initializer='glorot_normal')
conv4=Conv2D(64,kernel_size=3, strides=1,
                  padding='SAME', use_bias=True, 
                  activation='relu',input_shape=[8,8,128],kernel_initializer='glorot_normal')

maxpool2=MaxPooling2D(pool_size=2,strides=2)
model=conv1(input1)
model=conv2(model)
model=maxpool1(model)
model=conv3(model)
model=conv4(model)
model=maxpool2(model)
model=Flatten()(model)

exten1=Dense(1024, activation='relu')(model)
exten1=Dropout(0.2)(exten1)
exten1=Dense(512, activation='relu')(exten1)
exten1=Dropout(0.2)(exten1)
exten1=Dense(256, activation='relu')(exten1)
exten1=Dense(2, activation='sigmoid',name='output1')(exten1)

exten2=Dense(1024, activation='relu')(model)
exten2=Dropout(0.2)(exten2)
exten2=Dense(512, activation='relu')(exten2)
exten2=Dropout(0.2)(exten2)
exten2=Dense(256, activation='relu')(exten2)
exten2=Dense(2, activation='sigmoid',name='output2')(exten2)


model = Model(inputs=[input1], outputs=[exten1, exten2])


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

Model summary is:

Layer (type)                    Output Shape         Param #     Connected to                     
==================================================================================================
input_4 (InputLayer)            (None, 8, 8, 128)    0                                            
__________________________________________________________________________________________________
conv2d_24 (Conv2D)              (None, 8, 8, 128)    147584      input_4[0][0]                    
__________________________________________________________________________________________________
conv2d_25 (Conv2D)              (None, 8, 8, 128)    147584      conv2d_24[0][0]                  
__________________________________________________________________________________________________
max_pooling2d_12 (MaxPooling2D) (None, 4, 4, 128)    0           conv2d_25[0][0]                  
__________________________________________________________________________________________________
conv2d_26 (Conv2D)              (None, 4, 4, 64)     73792       max_pooling2d_12[0][0]           
__________________________________________________________________________________________________
conv2d_27 (Conv2D)              (None, 4, 4, 64)     36928       conv2d_26[0][0]                  
__________________________________________________________________________________________________
max_pooling2d_13 (MaxPooling2D) (None, 2, 2, 64)     0           conv2d_27[0][0]                  
__________________________________________________________________________________________________
flatten_2 (Flatten)             (None, 256)          0           max_pooling2d_13[0][0]           
__________________________________________________________________________________________________
dense_14 (Dense)                (None, 1024)         263168      flatten_2[0][0]                  
__________________________________________________________________________________________________
dense_17 (Dense)                (None, 1024)         263168      flatten_2[0][0]                  
__________________________________________________________________________________________________
dropout_8 (Dropout)             (None, 1024)         0           dense_14[0][0]                   
__________________________________________________________________________________________________
dropout_10 (Dropout)            (None, 1024)         0           dense_17[0][0]                   
__________________________________________________________________________________________________
dense_15 (Dense)                (None, 512)          524800      dropout_8[0][0]                  
__________________________________________________________________________________________________
dense_18 (Dense)                (None, 512)          524800      dropout_10[0][0]                 
__________________________________________________________________________________________________
dropout_9 (Dropout)             (None, 512)          0           dense_15[0][0]                   
__________________________________________________________________________________________________
dropout_11 (Dropout)            (None, 512)          0           dense_18[0][0]                   
__________________________________________________________________________________________________
dense_16 (Dense)                (None, 256)          131328      dropout_9[0][0]                  
__________________________________________________________________________________________________
dense_19 (Dense)                (None, 256)          131328      dropout_11[0][0]                 
__________________________________________________________________________________________________
output1 (Dense)                 (None, 2)            514         dense_16[0][0]                   
__________________________________________________________________________________________________
output2 (Dense)                 (None, 2)            514         dense_19[0][0]                   
==================================================================================================
Total params: 2,245,508
Trainable params: 2,245,508
Non-trainable params: 0

This is where i am facing Problems

model.fit(X_train,{"output1":label1, "output2":label2},batch_size=10,verbose=1 ,callbacks=[checkpoint],validation_data=(X_test,y_test),epochs=40)

the error i get is

ValueError: Error when checking model target: the list of Numpy arrays that you are passing to your model is not the size the model expected. Expected to see 2 array(s), but instead got the following list of 1 arrays

While both label1 and label2 are of same shape (2077,2). Is there something i am missing?

Answer 1

You need to pass an array object. You don't need to map the label array with the output node name ( like TensorFlow low-level APIs ). Pass the list containing the label arrays directly.

model.fit(X_train, [ label1, label2 ] ,batch_size=10,verbose=1 ,callbacks=[checkpoint],validation_data=(X_test,y_test),epochs=40)