ur simple test running live classification on accelerometer on a ardu board seems to be to quick in deciding what it detected… how do we make it wait a spesific time before getting to the conclution?
Question/Issue:
Project ID:
Context/Use case:
ur simple test running live classification on accelerometer on a ardu board seems to be to quick in deciding what it detected… how do we make it wait a spesific time before getting to the conclution?
Question/Issue:
Project ID:
Context/Use case:
Adjust the baud rates according to your requirement.
I dont think that is the issue… its more like: it decides what event its detected before the event is over… I need it to wait and analyse more.- since the beginning of the events seem to be very similar?
it concludes to fast…
You can adjust the window size of your impulse if your events are longer (window size should basically be the duration of the pattern you want to detect) and then retrain your model.
Aurelien
Think I found it, its in the arduinocode
// Determine the next tick (and then sleep later)
uint64_t next_tick = micros() + (EI_CLASSIFIER_INTERVAL_MS * 5000);// changed from 1 sec to 5 sec to make more acurate?
i guess it decided in 1000 msec. now i will try 5000to give it more time to evaluate the input
@Vard I don’t think that is correct. With your mods your locally running inference is now making predictions on 5 second samples and your model is trained on 1 seconds samples. I think it is better to change your Impulse in the Studio Create Impulse
page.
hmm, so is it the length of samples it has been trained for that sets the “attention span”.
Do you mean settings:" Time series data-window size/Window increase"?
or somewhere else in the output code?
think i found the bug, one of the sets had 1 sec lenght, while others were 4 sec, this made it “find” the short one more than the others