Feature Array not matching up with Features generated by Live Classification - Arduino Nano 33 BLE Sense

#1

I have deployed my Edge Impulse to my Arduino Nano 33 BLE Sense (Rev1). As per the Edge Impulse documentation, I have copied the raw features from the Live Classification page, into the features array on Arduino IDE. However I am receiving the following error:

The size of your 'features' array is not correct. Expected 2000 items, but had 4500
Edge Impulse standalone inferencing (Arduino)

My window size is 5000ms, and my window increase is 2500ms. All the data has been taken at 100Hz frequency. I’m using 4 sensors in my spectral analysis - accX, accZ, gyrY, gyrZ. So it makes sense that its expecting 2000 items ((100 * 5 ) * 4). The input axes in the time series data column however have 9 input axes - all 3 accelerometers, all 3 gyros, and all 3 magnetometers. Is that why my live classification is outputting 4500 features?

How can I fix this?

#2

Hello @njofbabylon,

There is maybe a bug with the live classification page for your workflow, I’m checking with our studio team and come back to you.

Best,

Louis

1 Like
#3

@louis Hi Louis any news regarding this?

#4

Hello @njofbabylon,

There is indeed a bug and I’ve opened an internal ticket.
In the meantime, you can go to the Spectral Analysis tab and copy the raw feature for a given sample from there.
You won’t see the raw data for a sample that is in your test set but it will let you test your impulse on a given sample. Not ideal solution but I hope it can unblock you while we’re fixing this.

Best,

Louis