[Python] Why does my keras model with multiple inputs accept the shape of the training data...

Im currently investigating the effekt of masking attention Scores on MultiHeadAttention Layers in a Transformer model for classification of time series data. I have build a model that accepts a time series and a mask.

def build(params: dict) -> keras.Model:

input_dim = 1
sequence_size = params["sequence_size"]
n_classes = params["n_classes"]
encoder_blocks = params["encoder_blocks"]
n_heads = params["encoder_heads"]
encolder_mlp = params["mlp_dim"]
conv_filters = params["conv_filters"]
encoder_dropout = params["encoder_dropout"]
mlp_dropout = params["mlp_dropout"]
learning_rate = params["learning_rate"]

inputs = keras.Input(shape=(sequence_size, input_dim), name="sequence_input")
mask_input = keras.Input(shape=(sequence_size, sequence_size), name="mask_input")

x = inputs + SinePositionEncoding()(inputs)

for _ in range(encoder_blocks):
x = transformer_encoder(x, head_size=sequence_size, num_heads=n_heads, con_filters=conv_filters, attention_mask=mask_input, dropout=encoder_dropout, seed=SEED)

x = layers.GlobalAveragePooling1D(data_format="channels_last")(x)
x = layers.Dense(encolder_mlp, activation="relu")(x)
x = layers.Dropout(mlp_dropout, seed=SEED)(x)

outputs = layers.Dense(n_classes, activation="softmax")(x)

model = keras.Model(inputs=[inputs, mask_input], outputs=outputs)

model.compile(
loss="categorical_crossentropy",
optimizer=keras.optimizers.Adam(learning_rate=learning_rate),
metrics=["categorical_accuracy", "f1_score"]
)

return model


When i call model.fit([x_train, dummy_mask], y_train) everthing works fine. If i call model.evaluate([x_train, dummy_mask], y_train i get an ValueError:

ValueError: Data cardinality is ambiguous. Make sure all arrays contain the same number of samples.'x' sizes: 60, 577
'y' sizes: 60


x_train shape = (60, 577, 1)

dummy_mask shape = (577, 577, 1)

I don't have a clue why that is, any suggestions?

PS: I am fully aware that calling the .evaluate() method with train data makes no sense. I tried to narrow the error down. At first i guessed something with my test data is worng but when the model could not be evaluated on the same data it was trained on, somethind other must be off.

Continue reading...