Clarification needed on calculating the number of training iterations for pretrain ac in trainer.py

[yixiaoshenghua]

Hello,

I have a question regarding the code in the trainer.py file, specifically lines 251-252 in the # pretrain ac section:

budget = config['pretrain_budget'] * (1 - config['pretrain_obs_p'] + config['pretrain_dyn_p'])

I would like to clarify whether the intention of this line is to calculate the training iterations for pretraining the action-conditioned (ac) model as a proportion subtracted from 1. In other words, is the calculation meant to determine the training budget as a percentage of the total pretrain budget?

Like this:

budget = config['pretrain_budget'] * (1 - config['pretrain_obs_p'] - config['pretrain_dyn_p'])

Thank you for your attention and clarification.