Skip to content

scott.notes.md

Latest commit

 

History

History
121 lines (83 loc) · 4.6 KB

scott.notes.md

File metadata and controls

121 lines (83 loc) · 4.6 KB

newer notes re: dec 16

python3.9 -m venv iris_env_scott_local_3_9_20
#python3.9 -m venv iris_env_scott_cluster_loginnode_3_9_20
source iris_env_scott_local_3_9_20/bin/activate
pip install -r requirements.txt

scott getting an error on gym=0.21.0 on macbook

this seemed to work just fine on the login node (linux)

Collecting gym==0.21.0 (from gym[accept-rom-license]==0.21.0->-r requirements.txt (line 3))
  Using cached gym-0.21.0.tar.gz (1.5 MB)
  Installing build dependencies ... done
  Getting requirements to build wheel ... error
  error: subprocess-exited-with-error
  
  × Getting requirements to build wheel did not run successfully.
  │ exit code: 1
  ╰─> [3 lines of output]
      /private/var/folders/_l/_09gzxn97x137yclkhgjn7680000gs/T/pip-build-env-5hs1l5rq/overlay/lib/python3.9/site-packages/setuptools/_distutils/dist.py:261: UserWarning: Unknown distribution option: 'tests_require'
        warnings.warn(msg)
      error in gym setup command: 'extras_require' must be a dictionary whose values are strings or lists of strings containing valid project/version requirement specifiers.
      [end of output]
  
  note: This error originates from a subprocess, and is likely not a problem with pip.
****

OLD NOTES FOLLOW

overview

The original requirements.txt file won't build as-is (without indeterminate hacking).

Even downgrading to python 3.9 didn't do the trick.

Regardless: our clusters run 3.11.5, so I've updated the requirements to work with that version.

I've tried to update only the packages that are necessary to get the code to run with 3.11.5.

Next step: pull down to the cluster and see if I can get our training of iris to work in a reproducible fashion.

steps

install dependencies

set up python 3.11.5 venv, activate it

requires: python 3.11.5 and the changes I've made to the repo (requirements) e.g.

brew install pyenv
pyenv install 3.11.5

~/.pyenv/versions/3.11.5/bin/python -m venv iris_env_3.11.5
source iris_env_3.11.5/bin/activate
pip install --upgrade pip setuptools
pip install -r requirements.txt

download roms

requires a directory called roms

mkdir roms
AutoROM --install-dir roms --accept-license

install roms

ale-import-roms roms

train it

note you can use 'online' instead of 'disabled' for wandb.mode

using cuda (doesn't work on mac)

python src/main.py env.train.id=BreakoutNoFrameskip-v4 common.device=cuda:0 wandb.mode=disabled

using cpu (works on mac)

note: this is slow, but it works. For m1/m2 we could use mps but that doesn't seem to work well. TODO check this again for python 3.11; this was true for 3.9... python src/main.py env.train.id=BreakoutNoFrameskip-v4 common.device=cpu wandb.mode=disabled

using mps (mac m series chips)

First, you need to the mps/metal version of pytorch: pip install torch==1.13.1 torchvision torchaudio --extra-index-url https://download.pytorch.org/whl/metal.html

then, you can train: python src/main.py env.train.id=BreakoutNoFrameskip-v4 common.device=mps wandb.mode=disabled

NOTE: this doesn't actually work for me

I get this error:

Epoch 5 / 600

Experience collection (train_dataset): 100%|███████████████████████████████████████████████████████████████████████████████████████████████████████████████████████| 200/200 [00:06<00:00, 33.11it/s]
Experience collection (test_dataset): 100%|██████████████████████████████████████████████████████████████████████████████████████████████████████████████████████████| 16/16 [00:22<00:00,  1.40s/it]
Assertion failed: (0 <= mpsAxis && mpsAxis < 4 && "Runtime canonicalization must simplify reduct[1]    25954 abort      python src/main.py env.train.id=BreakoutNoFrameskip-v4 common.device=mps 
(iris_env) ➜  scott_iris_hack git:(main) /opt/homebrew/Cellar/python@3.9/3.9.20/Frameworks/Python.framework/Versions/3.9/lib/python3.9/multiprocessing/resource_tracker.py:216: UserWarning: resource_tracker: There appear to be 7 leaked semaphore objects to clean up at shutdown
  warnings.warn('resource_tracker: There appear to be %d '

you can get back to basic torch via

pip uninstall torch torchvision torchaudio
pip install -r requirements.txt