Parameter Sweep over a Representative Set of Latent Space Dimensionality

greenelab / tybalt

Training and evaluating a variational autoencoder for pan-cancer gene expression data

BSD 3-Clause "New" or "Revised" License

162 stars 62 forks source link

Parameter Sweep over a Representative Set of Latent Space Dimensionality #116

Closed gwaybio closed 6 years ago

gwaybio commented 6 years ago

This pull request is a precursor to the larger plan of sweeping over many more latent space combinations. The purpose of this pull request is to test hyperparameter stability while varying the latent space dimensionality in adage and tybalt models.

When we perform the larger sweep, we want to be sure that the results are not impacted by poor hyperparameter combinations.

jaclyn-taroni commented 6 years ago

I have no comments about the code @gwaygenomics, but let's see what @danich1 has to say about it. I am assuming that this smaller set hyperparameter choices (and how they differ for each of the models) is based on some prior experience. You might consider recording that here for posterity.

gwaybio commented 6 years ago

Thanks for comments @danich1 - I have addressed your comments (and added some changes of my own) to the PR. Ready for review again!