Using Expert Mode to train a model and deploy it

Help
#1

Hi,
Currently, I download edge impulse Arduino Library and use continuous inference to classify sounds. However, sometimes after deployment of the device (Arduino Nano BLE Sense 33) it is required to update the model/ learning block. I am doing so using expert mode so that I can train the model locally without requiring Edge Impulse UI. Once I have trained the model I want to change the original downloaded library just so that the model is updated (Not the processing block) so that I can add more targets. Can someone suggest how would I be able to accomplish this. When I save the model in expert mode its in .pb file. How do I use it with Arduino Nano BLE Sense 33 to update the current library.

Thank you.

#2

Hi @umaid,

There is no easy way to do this. The downloaded firmware source code (e.g. C++ or Arduino library) combines both the model and processing block code, and they are tightly connected so that changing one means you have to change a lot of other code in the library.

Your best bet would be to use the Expert Mode in the Edge Impulse Studio to design your model. So long as it can be converted to a TensorFlow Lite for Microcontroller model (i.e. it contains OpCodes found in TFLM), you should be able to download the firmware library from Edge Impulse and have it compile.