FAQs

Please also refer to https://github.com/xuzijian629/combopt-zero/issues/2

When I run ./t_sample.sh, it outputs [generator] best model not found

This is intended. Data generators load the current best model when started. At first several minutes (configured in {problem}/config.sh. Default: 1 mins), learners are stopped in order to avoid learning from very limited data. Therefore, data generator cannot find the best model. Just wait a few minutes.

Got a error "mkdir: cannot create directory ‘./results/sample/p-0.8-bsize-16-bnum-20-hot-100-parallel-10-4-alpha-1.5-play-20-rollout---4-lr-0.001-l2-0.0001-save-15-cleanup-1-keep-3/lock_data_all’: File exists" in training

In order to handle parallel read and write, we lock the entire training data when accessed by a process. But it seems that this lock is sometimes violated when run on Docker. If you want to avoid this error, please run the program without Docker (but we observed that though it outputted many errors, the generated models actually worked well).