NVlabs
Why the ProcessAgent use Process while the ThreadTrainer use Thread?
Hello!
Why the ProcessAgent use Process while the ThreadTrainer use Thread? I wonder whether the ProcessAgent.py could use Thread instead of Process
I have the same question. Is it because the efficiency of data transform in Queue between processes?
I might figure out the reason after reimplementing this project. Since the trainers and predictors have the same tensorflow model and parameters, and it's hard/inefficient to share them by Queue in multiprocess class. Although in Python the multithread class cannot use multi-core due to GIL, the trainer and predictors are usually using the same GPU, so the bottleneck happens in GPU computation rather than in the multithread trainers and predictors in the main process.