search

Bishop-Laboratory / autoPLIER-analysis

Analysis for the autoPLIER project
2 stars 1 forks source link

Need to figure out why precision for PLIER is so much higher than autoPLIER in current tests #38

Closed millerh1 closed 2 years ago

millerh1 commented 2 years ago

The precision for autoPLIER appears to be much lower than PLIER on the CELLO dataset.

  1. Is the reconstruction error correlated with the classification error?
  2. Is autoPLIER mis-classifying the same labels as PLIER?
  3. If you remove the label with the low number of samples, does it improve the performance?
  4. Can you add the standard deviation to the plots?
millerh1 commented 2 years ago

We also decided that the Z matrix from PLIER is probably not very sparse... We should analyze the sparsity of Z and target that sparsity with autoPLIER. That may improve autoPLIER performance compared to PLIER.

Julian-Borbeck commented 2 years ago

PLIER is not true NMF, the embedding contains negative values.