I have a multi-input Keras model combining a text input and a numeric input. Both inputs are passed through Dense layers followed by Embedding layers, then concatenated and pooled using GlobalMaxPooling1D. Despite thorough debugging of layer dimensions at each step, I still encounter an error when feeding data into the model.
Here is the model definition:
def model_builder(hp, vocab_size, max_token_size):
# Hiperparametre tanımları
hp_text_units = hp.Int('text_units', min_value=64, max_value=512, step=32)
hp_numeric_units = hp.Int('numeric_units', min_value=32, max_value=256, step=32)
hp_dropout = hp.Float('dropout', min_value=0.1, max_value=0.5, step=0.1)
hp_learning_rate = hp.Float('learning_rate', min_value=1e-4, max_value=1e-2, sampling='log')
# Girdi tanımlamaları
text_input = tf.keras.Input(shape=(max_token_size,), dtype="int32", name="text_input")
numeric_input = tf.keras.Input(shape=(26,), name="numeric_input")
# Debugging Embedding Layer Output
temp_model = tf.keras.Model(inputs=text_input, outputs=text_input)
print("Text input Shape (Static):", temp_model.output_shape)
temp_model = tf.keras.Model(inputs=numeric_input, outputs=numeric_input)
print("Numeric input Shape (Static):", temp_model.output_shape)
text_dense = tf.keras.layers.Dense(224, activation='relu')(text_input)
# Debugging Text Dense Layer Output
temp_model = tf.keras.Model(inputs=text_input, outputs=text_dense)
print("Text Dense Layer Shape (Static):", temp_model.output_shape)
# Embedding katmanı
embedding_layer = tf.keras.layers.Embedding(
input_dim=224,#vocab_size,
output_dim=224, # Embedding boyutu sabit
#input_length=max_token_size
)(text_dense)
# Debugging Embedding Layer Output
temp_model = tf.keras.Model(inputs=text_input, outputs=embedding_layer)
print("Embedding Layer Shape (Static):", temp_model.output_shape)
# Text Pipeline
text_dense1 = tf.keras.layers.Dense(224, activation='relu')(embedding_layer)
# Debugging Text Dense Layer Output
temp_model = tf.keras.Model(inputs=text_input, outputs=text_dense1)
print("Text Dense Layer 1 Shape (Static):", temp_model.output_shape)
text_batch_normalization = tf.keras.layers.BatchNormalization()(text_dense1)
# Debugging Text Dense Layer Output
temp_model = tf.keras.Model(inputs=text_input, outputs=text_batch_normalization)
print("Text Batch Normalization Shape (Static):", temp_model.output_shape)
text_dropout = tf.keras.layers.Dropout(hp_dropout)(text_batch_normalization)
temp_model = tf.keras.Model(inputs=text_input, outputs=text_dropout)
print("Text Dropout Shape (Static):", temp_model.output_shape)
# Numeric Pipeline
numeric_input_reshaped = tf.keras.layers.Reshape((26,))(numeric_input)
temp_model = tf.keras.Model(inputs=numeric_input, outputs=numeric_input_reshaped)
print("Numeric Reshape Layer Shape (Static):", temp_model.output_shape)
numeric_dense = tf.keras.layers.Dense(224, activation='relu')(numeric_input_reshaped)
# Debugging Numeric Dense Layer Output
temp_model = tf.keras.Model(inputs=numeric_input, outputs=numeric_dense)
print("Numeric Dense Layer Shape (Static):", temp_model.output_shape)
#flattened_numeric_input = tf.keras.layers.Flatten()(numeric_input)
embedding_layer_numeric = tf.keras.layers.Embedding(
input_dim=224,
output_dim=224, # Embedding boyutu sabit
#input_length=max_token_size
)(numeric_dense)
# Debugging Embedding Layer Output
temp_model = tf.keras.Model(inputs=text_input, outputs=embedding_layer_numeric)
print("Embedding Layer Numeric Shape (Static):", temp_model.output_shape)
numeric_dense1 = tf.keras.layers.Dense(224, activation='relu')(embedding_layer_numeric)
# Debugging Numeric Dense Layer Output
temp_model = tf.keras.Model(inputs=numeric_input, outputs=numeric_dense1)
print("Numeric Dense Layer 1 Shape (Static):", temp_model.output_shape)
numeric_batch_normalization = tf.keras.layers.BatchNormalization()(numeric_dense1)
temp_model = tf.keras.Model(inputs=numeric_input, outputs=numeric_batch_normalization)
print("Numeric Batch Normalization Shape (Static):", temp_model.output_shape)
numeric_dropout = tf.keras.layers.Dropout(hp_dropout)(numeric_batch_normalization)
temp_model = tf.keras.Model(inputs=numeric_input, outputs=numeric_dropout)
print("Numeric Dropout Shape (Static):", temp_model.output_shape)
# Birleştirme
combined = tf.keras.layers.Concatenate()([text_dropout, numeric_dropout])
# Debugging Concatenated Layer Output
#temp_model = tf.keras.Model(inputs=[text_input, numeric_input], outputs=combined)
temp_model = tf.keras.Model(inputs=[combined], outputs=combined)
print("Concatenated Layer Shape (Static):", temp_model.output_shape)
# Global Max Pooling ile boyut düşürme
combined_pooled = tf.keras.layers.GlobalMaxPooling1D()(combined)
temp_model = tf.keras.Model(inputs=[combined], outputs=combined_pooled)
print("Concatenated GlobalMaxPooling Shape (Static):", temp_model.output_shape)
# Combined Pipeline
combined_dense1 = tf.keras.layers.Dense(256, activation='relu')(combined_pooled)
# Debugging Combined Dense Layer 1 Output
#temp_model = tf.keras.Model(inputs=[text_input, numeric_input], outputs=combined_dense1)
temp_model = tf.keras.Model(inputs=[combined], outputs=combined_dense1)
print("Combined Dense Layer 1 Shape (Static):", temp_model.output_shape)
combined_batch_normalization = tf.keras.layers.BatchNormalization()(combined_dense1)
#temp_model = tf.keras.Model(inputs=[text_input, numeric_input], outputs=combined_batch_normalization)
temp_model = tf.keras.Model(inputs=[combined], outputs=combined_batch_normalization)
print("Combined Batch Normalization Shape (Static):", temp_model.output_shape)
combined_dropout = tf.keras.layers.Dropout(hp_dropout)(combined_batch_normalization)
#temp_model = tf.keras.Model(inputs=[text_input, numeric_input], outputs=combined_dropout)
temp_model = tf.keras.Model(inputs=[combined], outputs=combined_dropout)
print("Combined Dropout Shape (Static):", temp_model.output_shape)
combined_dense2 = tf.keras.layers.Dense(128, activation='relu')(combined_dropout)
# Debugging Combined Dense Layer 2 Output
#temp_model = tf.keras.Model(inputs=[text_input, numeric_input], outputs=combined_dense2)
temp_model = tf.keras.Model(inputs=[combined], outputs=combined_dense2)
print("Combined Dense Layer 2 Shape (Static):", temp_model.output_shape)
combined_batch_normalization1 = tf.keras.layers.BatchNormalization()(combined_dense2)
#temp_model = tf.keras.Model(inputs=[text_input, numeric_input], outputs=combined_batch_normalization1)
temp_model = tf.keras.Model(inputs=[combined], outputs=combined_batch_normalization1)
print("Combined Batch Normalization 1 Shape (Static):", temp_model.output_shape)
combined_dropout1= tf.keras.layers.Dropout(hp_dropout)(combined_batch_normalization1)
#temp_model = tf.keras.Model(inputs=[text_input, numeric_input], outputs=combined_dropout1)
temp_model = tf.keras.Model(inputs=[combined], outputs=combined_dropout1)
print("Combined Dropout 1 Shape (Static):", temp_model.output_shape)
combined_dense3 = tf.keras.layers.Dense(64, activation='relu')(combined_dropout1)
# Debugging Combined Dense Layer 3 Output
#temp_model = tf.keras.Model(inputs=[text_input, numeric_input], outputs=combined_dense3)
temp_model = tf.keras.Model(inputs=[combined], outputs=combined_dense3)
print("Combined Dense Layer 3 Shape (Static):", temp_model.output_shape)
combined_dropout2 = tf.keras.layers.Dropout(hp_dropout)(combined_dense3)
#temp_model = tf.keras.Model(inputs=[text_input, numeric_input], outputs=combined_dropout2)
temp_model = tf.keras.Model(inputs=[combined], outputs=combined_dropout2)
print("Combined Dropout2 Shape (Static):", temp_model.output_shape)
# Çıkış Katmanı
output = tf.keras.layers.Dense(1, activation='sigmoid')(combined_dropout2)
# Debugging Output Layer Shape
#temp_model = tf.keras.Model(inputs=[text_input, numeric_input], outputs=output)
temp_model = tf.keras.Model(inputs=[combined], outputs=output)
print("Output Layer Shape (Static):", temp_model.output_shape)
# Model Derleme
model = tf.keras.Model(inputs=[text_input, numeric_input], outputs=output)
model.compile(
optimizer=tf.keras.optimizers.Adam(learning_rate=hp_learning_rate),
loss='binary_crossentropy',
metrics=['accuracy']
)
return model
Error Logs:
Despite verifying dimensions throughout the model using debug print statements, I get the following error when the model starts training:
ValueError: Exception encountered when calling Functional.call(). Input 0 of layer "dense" is incompatible with the layer: expected axis -1 of input shape to have value 1149, but received input with shape (None, 26)
Debug Logs: Here are the layer dimensions printed during model building:
Text input Shape (Static): (None, 1149) Text Dense Layer Shape (Static): (None, 224) Embedding Layer Shape (Static): (None, 224, 224) Text Dense Layer 1 Shape (Static): (None, 224, 224) Text Batch Normalization Shape (Static): (None, 224, 224) Text Dropout Shape (Static): (None, 224, 224) Numeric input Shape (Static): (None, 26) Numeric Reshape Layer Shape (Static): (None, 26) Numeric Dense Layer Shape (Static): (None, 224) Embedding Layer Numeric Shape (Static): (None, 224, 224) Numeric Dense Layer 1 Shape (Static): (None, 224, 224) Numeric Batch Normalization Shape (Static): (None, 224, 224) Numeric Dropout Shape (Static): (None, 224, 224) Concatenated Layer Shape (Static): (None, 224, 448) Concatenated GlobalMaxPooling Shape (Static): (None, 448) Combined Dense Layer 1 Shape (Static): (None, 256) Combined Batch Normalization Shape (Static): (None, 256) Combined Dropout Shape (Static): (None, 256) Combined Dense Layer 2 Shape (Static): (None, 128) Combined Batch Normalization 1 Shape (Static): (None, 128) Combined Dropout 1 Shape (Static): (None, 128) Combined Dense Layer 3 Shape (Static): (None, 64) Combined Dropout2 Shape (Static): (None, 64) Output Layer Shape (Static): (None, 1)
