это моя модель в керасе.у меня есть 3 единицы отсева (модель, модель_h1, модель_h2).в определенной модели «model_h1» и «model_h2» не связаны с «моделью», но почему в графе, созданном TensorBoard, есть входная ссылка от «model» до «model_h1» и «model_h2»?
e = Embedding(11619, 200, weights=[wv], input_length=sentMax,trainable=False)
## Original Sentence
input1=Input(shape=(np.shape(W_train)[1],),name='text')
input2=Input(shape=(157,1),name='d1')
input3=Input(shape=(157,1),name='d2')
x1=e(input1)
x2=Dense(1,activation='linear')(input2)
x3=Dense(1,activation='linear')(input3)
inputs=concatenate([x1,x2,x3],axis=-1,name="concat")
t=Bidirectional(LSTM(100,return_sequences=True),merge_mode='concat')(inputs)
model=Dropout(0.5, noise_shape=None, seed=None)(t)
pool = Attention()(model)
## Help1 Sentence
input1_h1=Input(shape=(np.shape(W_h1_train)[1],),name='text_h1')
input2_h1=Input(shape=(157,1),name='d1_h1')
input3_h1=Input(shape=(157,1),name='d2_h1')
x1_h1=e(input1_h1)
x2_h1=Dense(1,activation='linear')(input2_h1)
x3_h1=Dense(1,activation='linear')(input3_h1)
inputs_h1=concatenate([x1_h1,x2_h1,x3_h1],axis=-1,name="concat_h1")
t_h1=Bidirectional(LSTM(100,return_sequences=True),merge_mode='concat')(inputs_h1)
model_h1=Dropout(0.5, noise_shape=None, seed=None)(t_h1)
pool_h1 = Attention()(model_h1)
## Help2 Sentence
input1_h2=Input(shape=(np.shape(W_h2_train)[1],),name='text_h2')
input2_h2=Input(shape=(157,1),name='d1_h2')
input3_h2=Input(shape=(157,1),name='d2_h2')
x1_h2=e(input1_h2)
x2_h2=Dense(1,activation='linear')(input2_h2)
x3_h2=Dense(1,activation='linear')(input3_h2)
inputs_h2=concatenate([x1_h2,x2_h2,x3_h2],axis=-1,name="concat_h2")
t_h2=Bidirectional(LSTM(100,return_sequences=True),merge_mode='concat')(inputs_h2)
model_h2=Dropout(0.5, noise_shape=None, seed=None)(t_h2)
pool_h2 = Attention()(model_h2)
combined=concatenate([pool,pool_h1,pool_h2],axis=1)
X=Dense(300,activation='relu',kernel_regularizer=regularizers.l2(0.0001))(combined)
X=Dense(100,activation='relu',kernel_regularizer=regularizers.l2(0.0001))(X)
main_output = Dense(5,activation='softmax',kernel_regularizer=regularizers.l2(0.0001))(X)
output_model = Model(inputs=[input1,input2,input3,input1_h1,input2_h1,input3_h1,input1_h2,input2_h2,input3_h2], outputs=main_output)
График сделан тензорной доской