Incorporate Keras Network WITH LSTM Layer

I have a compiled and trained networkmodel with an LSTM Layer, as .py or saved as .HDF5 using Keras with Tensorflowbackend. Is there a way to use it (give input, predict output) in Matlab? The importKerasLayer function seems not to work with LSTMs at all.

Answer 1

[Sivylla Paraskevopoulou]

What version are you using? Since R2018b, you can Import LSTM and BiLSTM layers from TensorFlow-Keras. For a list of support Keras layers see the documentation of importKerasNetwork and importKerasLayers functions. Also, note that importKerasNetwork or importKerasLayers support importing a TensorFlow-Keras network only in HDF5 or JSON format.

Answer 2

[Phan Phat]

Dear @Sivylla Paraskevopoulou,

Could you tell more detail about how to import the LSTM model on MATLAB? I used importKerasNetwork; however, seem does not work. Thanks!

[David Willingham]

Hi, I'd recommend using the importTensorFlowNetwork instead as it has more support for the layers that can be imported. Note: you'll need to update your Keras network to TensorFlow 2.0 in order to use this.