0

I am trying to train a single hidden layer Keras sequential multi-output (two outputs) "regression" network on non-image data. I would like each output to have separate activation function assigned to it. This is the form of my model construction: Is this the correct way to set it up for two output neurons each as a vector?

n_epoch = 1000 
hidden_layer1_neurons = 100
learning_rate =0.01
batch_size = 2000 # 10000
activation_1 = 'relu'   # Activation function for Hidden Layer Neurons
activation_2 = 'tanh'   # Activation function for Output Layer Neuron 1
activation_3 = 'sigmoid'  # Activation function for Output Layer Neuron 
2
min_lr = 0.008
lr_reduce_patience = 10
earlystop_patience = 100 
earlystop_min_delta = 0.0001 


model = Sequential()
input_layer = model.add(Dense(units=hidden_layer1_neurons, input_dim= 
  input_dim,
                kernel_initializer = initializers.RandomNormal(mean=0.0, 
  stddev=0.05)))


model.add(Activation(activation_1)) 

model.add(Dense(units=2))
model.add(Activation(activation_2))
model.add(Activation(activation_3))

 model.compile(loss='mean_squared_error', 
   optimizer=Adam(lr=learning_rate), metrics=['mae'])

And this is my training cell:

    hist = model.fit_generator(generator=training_generator,validation_data = 
       validation_generator, epochs = n_epoch,
              use_multiprocessing=False, callbacks = [reduce, earlystop,checkpointer, 
        csv_logger])
bluered_earth
  • 149
  • 5
  • 1
    No, multi-output models can only be made with the Functional API. – Dr. Snoopy Nov 25 '22 at 00:25
  • [This](https://stackoverflow.com/questions/66845924/multi-input-multi-output-model-with-keras-functional-api/66849164#66849164) might be helpful for your case. – Innat Nov 25 '22 at 12:23

0 Answers0