search

talmo / leap

LEAP is now deprecated -- check out its successor SLEAP!
https://sleap.ai
Apache License 2.0
206 stars 48 forks source link

Strange new error #15

Closed kbakhurin closed 5 years ago

kbakhurin commented 5 years ago

Hi Talmo,

Suddenly encountering a new problem with LEAP.

during the cluster_samples program, the PCA computation step results in 90% of the variance explained in blank modes, where there are 0 modes. The matlab command prompt reads:

Warning: Columns of X are linearly dependent to within machine precision. Using only the first 0 components to compute TSQUARED.

Do you know what this could be caused by? I tried it with two different videos. I feel like I remember seeing this happen before but I can't remember exactly what where the circumstances.

Thanks!

talmo commented 5 years ago

Hey Konstantin,

After you click Load samples, how many frames does it say loaded?

What is the shape of the dataset(s) in the table?

(Feel free to paste a screenshot here if it's easier)

kbakhurin commented 5 years ago

Hi Talmo,

Here is a screenshot of the cluster sampling window after i load samples and compute PCA.

image

talmo commented 5 years ago

Hey Konstantin,

Very bizarre error. I can't see anything in the code that would be an obvious reason why this is happening.

Try just skipping the PCA Preprocessing and just cluster without it -- it should still work, and if not it'll at least suggest a reason as to what's wrong.

kbakhurin commented 5 years ago

OK I looked at the clusters and they were all identical. The reason was that my .h5 file was messed up and when i was creating it, I accidentally just saved the same frame 2709 times. So PCA had no variance in the video to work with.

Thanks for the troubleshooting tip!