Add replicate runs - Githubissues

matsengrp / vampire

🧛 Deep generative models for TCR sequences 🧛

Apache License 2.0

16 stars 4 forks source link

Closed matsen closed 5 years ago

matsen commented 5 years ago

And try to figure out if validation loss (or Pvae) predicts eventual out-of-sample performance.

matsen commented 5 years ago

Note that https://github.com/matsengrp/vampire/commit/25f9d3e45a23204a95cc307bba72432a91398ae6 does replicates explicitly using SCons. After staring at Tensorboard I'm convinced that val_loss after pre-training is a good indicator of full validation loss, so I'm going to use that.

matsen commented 5 years ago

This sure seems to be working, and makes pretty plots besides:

This shows the warmups for 10 different random starts for three values of beta.