How to save trained model to a file and apply the model?

[zhangjingwei713]

I tested the test_smac_env.py and although it can run, the trained model is not saved. how can I save the model to a file, and how can I use the trained model like sl_model.pth for DI-star? Thanks!

[PaParaZz1]

• test_smac_env.py (link) is just a unittest file for smac environment without any training code.
• For SMAC training, you can enter ding -m serial -c dizoo/smac/config/smac_3s5z_qmix_config.py -s 0 in your terminal, and model will be saved automatically in default_experiment/ckpt directory when better final reward is acquired.
• For more details of loading and saving model, you can refer to doc.

[zhangjingwei713]

The model file ckpt_best.pth.tar is generated in the default_experiment/ckpt folder. Use ding -m eval -c dizoo/smac/config/smac_2s3z_qmix_config.py -s 0 --load-path default_ experiment/ckpt/ckpt_best.pth.tar command, I can see the battle between reinforcement-learning ai and the StarCratf 2 built-in ai, but the step speed is very fast like training. How can I set step speed to normal, and how to play the game with human vs agent mode? Thanks

[PaParaZz1]

Maybe you can try to enable this argument in SC2Env, and find additional information in original pysc2 repo.

[upia99]

Add a time.sleep() here if you want to slow down the game speed. Human vs agent is a bit of challenging. First, you need to be familiar with StartCraft2 editor, then you can edit the map which you trained your model with, replace the built-in AI with human player, then you can play with the model.