How to import Keras layers for regression?

Hi all. I am playing around with importing Keras layers for an LSTM problem but can't seem to get even a basic fully connected single layer network to work. Even though my Keras model just has a basic input layer, Matlab reads it as an "ImageInputLayer". This is for a simple sequence-to-sequence regression problem. I just want to feed in a 2D matrix with multiple features and a series of timesteps but it expects a 3D image tensor. Is there something wrong with the Keras model or do I need to preprocess my data differently? Thanks in advance!

[Friedrich Seiffarth]

Did you find a solution for your problem ? Because I am running into the same problem.

[Divya Gaddipati]

Could you mention the error you are getting?

Answer 1

[Sivylla Paraskevopoulou]

Since R2020b, Deep Learning Toolbox provides the featureInputLayer layer, and since R2021a you can import the TensorFlow-Keras layer Input as a featureInputLayer. For a complete list, see TensorFlow-Keras Layers Supported for Conversion into Built-In MATLAB Layers.

The importTensorFlowNetwork function tries to append an output layer to the imported network by interpreting the loss function of the TensorFlow model. If your model doesn't specify a loss function, specify the OutputLayerType name-value argument of importTensorFlowNetwork as "regression".