search

greenelab / tybalt

Training and evaluating a variational autoencoder for pan-cancer gene expression data
BSD 3-Clause "New" or "Revised" License
162 stars 61 forks source link

Refactoring Tybalt VAE Code #74

Closed gwaybio closed 6 years ago

gwaybio commented 6 years ago

Code is getting unwieldy with the same functions and classes being defined in multiple places. There is also necessarily a lot of customization required to run each model. This issue will include information on how a refactoring should be performed.

Points to consider:

  1. VAE custom classes require specific input but must also communicate to a new tybalt class
  2. tybalt class should have various methods for training, transforming, and evaluating performance metrics b. training should allow multiple architecture and hyper parameter setting and getting a. performance metrics include: avg sample reconstruction, per cancer-type reconstruction, training history visualization, etc.
gwaybio commented 6 years ago

Note that #98 addresses only portions of the performance metrics in 2a.

I added the issue to Budding in Project 1 considering it will be closed by #98 but will require additional modification to fully complete