Need help with uploading time series data

Help
#1

Question/Issue:
Hello community, I have a dataset that I managed to upload via the csv wizard. I am doing a sweep across a set frequency range to determine the SWR.

I sample each frequency 48 times per second. Start Frequency is 100KHz and end Frequency is 600KHz. I have 6145 raw values for each sample that is measured.

In total I have about 5000 samples containing 6145 raw values each.

When labeling my data, each sample has a different label name, as I get an error when the label names are all the same.

How do I get Edge Impulse to “see” each string of 6145 raw values as one sample instead of 6145 samples x 5000?

I am referring to this topic with a similar case: https://forum.edgeimpulse.com/t/metadata-could-not-be-loaded-after-training-complete/5592

Steps Taken:

  1. Uploaded the data as non time-series data
  2. Labeled data with unique labels
  3. Trained model
  4. Outcome error:
    7146EBEE-920A-4DEA-9702-7D3487E076DD
#2

Hi @shawn.taylor

Have you tried using our CSV Wizard?

When uploading via CSV wizard at the first step you can decide to have each row as a sample and later in the settings adjust the length of sample uploaded

Best

Eoin

#3

Thanks for the reply.

This is exactly what I used and I changed the formatting.

Instead of having the data in rows, I changed it to columns.

I also did not select “constant time series data” as this seems to automatically selects data from each row at the stipulated sampling frequency, which is not what I want.

I need raw values from 1 to 6000 to be “seen” as one sample, as the raw values will draw the waveform - if this makes sense?

It seems like that each raw value is interpreted as exactly one value and the model is being trained on individual raw values.

Any advice will be appreciated.