Keras concatenate Sequential and Dense models

Question

I have the following models:

Dense Model:

def dense_model(num_features):
  inputs = Input(shape=(num_features,), dtype=tf.float32)
  layers = Dense(32, activation='relu')(inputs)
  model = Model(inputs=inputs, outputs=layers)
  
  return model

LSTM Model:

def lstm_model(num_features):

    inputs = Input(shape=[None, num_features], dtype=tf.float32, ragged=True)
    layers = LSTM(16, activation='tanh')(
        inputs.to_tensor(), mask=tf.sequence_mask(inputs.row_lengths()))

    layers = BatchNormalization()(layers)
    layers = Dense(16,activation='relu')(layers)
    layers = Dense(1, activation='sigmoid')(layers)
    
    model = Model(inputs=inputs, outputs=layers)
    model.compile(loss='mse', optimizer='adam', metrics=['mse'])
        
    return model

---

I'm trying to concatenate the two networks like so:

def concatenate_model(num_features):
    model_1 = dense_model(10)
    inputs = Input(shape=[None, num_features], dtype=tf.float32, ragged=True)
    layers = LSTM(16, activation='tanh')(
        inputs.to_tensor(), mask=tf.sequence_mask(inputs.row_lengths()))

    layers = BatchNormalization()(layers)
    layers = Dense(16,activation='relu')(concatenate([layers,model_1]))
    layers = Dense(1, activation='sigmoid')(layers)
    
    model = Model(inputs=inputs, outputs=layers)
    model.compile(loss='mse', optimizer='adam', metrics=['mse'])
        
    return model

to get the following result:

But I get an error:

ValueError: as_list() is not defined on an unknown TensorShape.

Here is a Colab that demonstrate my issue. What am I missing?

Answer 1

Try this:

import tensorflow as tf

def dense_model(num_features):
  inputs = tf.keras.layers.Input(shape=(num_features,), dtype=tf.float32)
  layers = tf.keras.layers.Dense(32, activation='relu')(inputs)
  model = tf.keras.Model(inputs=inputs, outputs=layers)
  
  return model

def lstm_model(num_features):

    model_1 = dense_model(10)
    inputs = tf.keras.layers.Input(shape=[None, num_features], dtype=tf.float32, ragged=True)
    layers = tf.keras.layers.LSTM(16, activation='tanh')(
        inputs.to_tensor(), mask=tf.sequence_mask(inputs.row_lengths()))

    layers = tf.keras.layers.BatchNormalization()(layers)
    concatenated = tf.keras.layers.Concatenate(axis=-1)([layers, model_1.output])
    layers = tf.keras.layers.Dense(16,activation='relu')(concatenated)
    layers = tf.keras.layers.Dense(1, activation='sigmoid')(layers)
    
    model = tf.keras.Model(inputs=[inputs, model_1.input], outputs=layers)
    model.compile(loss='mse', optimizer='adam', metrics=['mse'])
        
    return model

model = lstm_model(20)
dot_img_file = '/model.png'
tf.keras.utils.plot_model(model, to_file=dot_img_file, show_shapes=True)