Mismatching windows from classifier. Expected: 24248, but got: 12124

Help
, ,
#1

Question/Issue:
Mismatching windows from classifier. Expected: 24248, but got: 12124

Project ID:
708193

Context/Use case:
Running Model testing after finishing training model got this issue.

Expected Outcome:
Expect testing outcome

Actual Outcome:
Provide test outcomes

Reproducibility:

  • [x ] Always
  • [ ] Sometimes
  • [ ] Rarely

Environment:

  • Platform: Edge Impulse Studio
  • Build Environment Details: NA
  • OS Version: Window 11
  • Edge Impulse Version (Firmware): NA
  • Edge Impulse CLI Version: NA
  • Project Version: Impulse #window 2s, stride 1s cnn
  • Custom Blocks / Impulse Configuration: NA
    Logs/Attachments:
    Error:
    image
    image948×593 42.1 KB

Additional Information:
[Any other information that might be relevant]

#2

My Impusle design:

image
image1920×959 84.5 KB

#3

HI @quangp

This is common to happen when you change some settings between training

  • Your model was trained with input windows of 24,248 values, but the test sample you’re using has only 12,124 values.
  • This usually happens if:
    • The window size or stride changed after training.
    • You’re testing with data collected under a different sampling frequency or duration.
    • There’s a mismatch between the DSP block output and model input.

Retrain all blocks again with the new settings.

Best

Eoin

#4

Hi @Eoin, in my case, I collected data with a duration of 4 seconds. Then I proceeded to create a 2s window, 1s stride as below.

image
image1920×700 66.3 KB

Should I recollect data with only 2 seconds?