notes.txt

Experiments

net_VAE - normalized
recon_loss - normalized
omics_mode - unnormalized

ABC_inter vs ABC_inter2 = 2 saves model for visualization of z
ABC_inter_MSE - MSE on unnormalized data (because recon_loss tuning was on normalized data)

Chosen A dataset (the one in main data folder): unnormalized
Chosen B dataset (the one in main data folder): B_meanimp
Chosen C dataset (the one in main data folder): unnormalized

C dataset didnt have values of GBM, so they were mean imputed for around 61 patients that had both A and B data

Plots
1. tsne
2. ABC_inter - recon_loss comparison between A, B and C
3. net_VAE / recon_loss / latent_space_dim / omics_mode: test accuracy comparison btw hyperparamter values
4. ABC_inter - KL loss : test and train : change in phases

Tables
1. omics mode
2. net_VAE
3. recon_loss
4. latent_space_dim

Final report modifications
1. Normalize C dataset too, and test ABC_inter_MSE
2. future work section on baseline paper
3. zero shot learning

final Experiments
all normalized
1. k_embed