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

Separate Diagnostic Notebook #57

Closed gwaybio closed 6 years ago

gwaybio commented 7 years ago

tybalt_vae.ipynb currently trains a VAE on pancan data and outputs the latent layer along with compiled Keras models.

the notebook also does a few sanity checks to ensure the model is actually learning something. This includes ensuring there are no zeroed out features and a quick check on reconstruction fidelity.

A more thorough diagnostic notebook would help ensure robust training. I will use this issue to discuss what should be included in the future nb.

  1. Cancer-type specific reconstruction - possibly point to needing to update batches in training
gwaybio commented 7 years ago

Closing this issue will also require removing the quick sanity checks in the current implementation of tybalt_vae.ipynb

gwaybio commented 6 years ago

more specific issues required - see #87 for current discussion about training diagnostics