I tried deploying an LSTM model on the Arduino Nano 33 BLE Sense using the ‘.tflite’ format, but I encountered errors. I discovered that LSTM models are not compatible with impulse. I learned that it might be possible to build an LSTM block using a custom learning block. If anyone has experience with this, could you please guide me through the process? It would be a great help.
Hi @jayram
Not sure where we are with LSTM, @brianmcfadden can you check or comment on that?
Best
Eoin
1 Like
Hi @jayram - what errors did you encounter when deploying your model to device?
Is there a reason you require an LSTM or could you try experimenting with other model architectures that are built into the Edge Impulse platform directly?
If your LSTM model is already trained, you can bring it into the platform using BYOM.
Alternatively, you can use a custom learning block to train the model in the platform. The documentation for custom learning blocks has several examples but, unfortunately, none for training an LSTM. You would have to implement the training code yourself.
One thing to note is that Studio is not set up for training recurrent neural networks, so some functionality may not work as expected.