Extending data for training

[irjambi]

Hi all,

I'm fairly new to ML and certainly very new to AutoKeras.
I had a question though:
I'm running a simple model search on MNIST and get some decent results.
However, I would now like to extend the inputs with some additional data. E.g. for MNISTs 28x28 values for each image, I would like to add another 1120 values that have additional information on the images encoded in a certain way.
Is this at all possible? Ideally I would add it in the form it is.
I read AutoKeras likes images, so could I potentially just make the 1120 values into another pseudo-image e.g. 32x35, and somehow add it to the original data?
Atm I'm trying to train the model on the original data, save the best model and then try to add the extra data via concatinating the model output and the extra branch (just another dense layer). Something like this (rtrain has shape (60000, 1120):

extra= Input(shape=(rtrain.shape[1],), name="extra_data")
extra_branch = layers.Dense(64, activation="relu")(extra)

merged = layers.concatenate([image_model.output, extra_branch])
output = layers.Dense(10, activation="softmax")(merged)

multi_input_model = Model(inputs=[image_model.input, reservoir], outputs=output)

but I get

ValueError: A Concatenate layer requires inputs with matching shapes except for the concatenation axis. Received: input_shape=[[(None, 10)], (None, 64)]

What do I need to do to have them match?

And more importantly: is there a better way to add additional data to my images but still use AutoKeras (or even without autokeras, open to any suggestions)?

Thanks and apologies if some bits are not quite clear :), happy to answer any follow ups!
Jambi