This repository serves reproducing the experiments from the paper Controlling the Interaction Between Generation and Inference in Semi-Supervised Variational Autoencoders Using Importance Weighting.
The commands for running the set of experiment for each model and each dataset
is in the corresponding .slurm or .sh file
named launch<Model><Dataset> in the launch_scripts folder.
For importance weighted experiments, the memory cost
can be lowered using the --grad_accu parameter which specifies the number
of gradient accumulation steps. The --batch_size parameter must then be lowered
accordingly. Example: --grad_accu 2 --batch_size 32 is equivalent to
--grad_accu 4 --batch_size 16 .