Hello, I have a case when the model starts to produce the same class for some time, what could it be? (on chip) thanks Norik.
I am not sure I understood your question, could you give some more context please?
yes, of course, I train the model, I test it and then I throw it on my stm32 chip, at first the model works well, then it starts to see only one wrong class, I can’t understand why it happens and why it works first and gives the correct classification, then they start to show only one class. Thnaks
Have you implemented any custom post-processing (long moving average or something similar)?
What kind of sensor are you using? Is it possible that it’s getting saturated over time? It can also happen if it heats too much.
On the Edge Impulse SDK, I have never experienced that kind of behavior. Is it possible that your buffer is not freed and get stuck with the same values (you should get an out of memory error I guess if that was the case)? How have you implemented your sampling strategy?
i use accelerometers a few after reading the sensor i free up memory i use kalman filter maybe it’s because of this thanks and I’ll check I wanted to make sure that this is not an SDK