I implemented the following changes with the motivation to improve BVAEs initial performance:
Model 1:
- Using .mean() on the loss term to make every node/trees contribution of same significance
Model 2:
- Using target indexes for F.Cross_entropy instead of providing the targets as is
I run the JOB light benchmark for these two variants of BVAEs and one for Native and will have to create plots to evaluate the approaches once the benchmark concludes.
I implemented the following changes with the motivation to improve BVAEs initial performance:
Model 1:
Model 2:
I run the JOB light benchmark for these two variants of BVAEs and one for Native and will have to create plots to evaluate the approaches once the benchmark concludes.