greenelab / multi-plier

An unsupervised transfer learning approach for rare disease transcriptomics
BSD 3-Clause "New" or "Revised" License
44 stars 10 forks source link

Add short experiment re: L2 parameter #63

Closed jaclyn-taroni closed 5 years ago

jaclyn-taroni commented 5 years ago

What happens when setting L2 = 0 during training, all else being equal? Explored in this new notebook 39-L2_penalty. In practice, the loadings (Z) are no longer sparse and pathway coverage tanks.

Here's the notebook for easy viewing: 39-L2_penalty.nb.html.zip