denisyarats / proto Goto Github PK

Thanks for open-sourcing this! In the paper (arXiv 2102.11271), there are experiments in multi-task setting (Walker-Run Forward&Run Backward&Flip Forward&Flip Backward, and Reach-Duplo Top Left&Top Right&Bottom Left&Bottom Right). How can I obtain these environments? Thanks a lot!

I keep getting the error
Error in magma_getdevice_arch: MAGMA not initialized (call magma_init() first) or bad device
when running the example training code. I switched to different machines and the error persists. Any ideas on how to resolve this? Thanks!

By installing hydra by
pip install hydra-core --upgrade The following error arises:

Key 'name' not in 'HydraConf'
        full_key: hydra.name
        reference_type=Optional[HydraConf]
        object_type=HydraConf

Thanks for open sourcing this! I wanted to replicate the results on the walker run task.
To reproduce the results of the paper, it is sufficient to run
python train.py env=walker_run num_expl_steps=250000 num_train_steps=250000?