Run bash test.sh in the command line.
ddp containts the ddp algorithm and functions.
environments contain the custom gym environments. If you add one, you need to update the environments/__init__.py
Each folder might have its own params, which should have a specialized name. Please don't use import params or
import blank as params as we have some unlimited imports (from x import *). In general, try not to have the same name as other files.
A simple fully connected model:
python experiment-single-force/train.py --verbose --baseline --total_steps 100
Learn hamiltonian without symplectic structure:
python experiment-single-force/train.py --verbose --total_steps 50
Take advantage of hamiltonian structure:
python experiment-single-force/train.py --verbose --structure --total_steps 50
You can also just run the training script, which will only train naive baseline and the structured symplectic models:
./train_pendulum.sh
You should be using Conda. After installing your new dependency using conda or pip, run at the top level:
conda export env > environment.yml
OR
conda env update --prefix ./env --file environment.yml --prune
and commit the changes. TODO(Rebecca): find out which of these makes more sense to use.
- Install conda https://docs.conda.io/en/latest/miniconda.html
- In a terminal, run:
conda env create -f environment.yml -n sympodenet
conda activate sympodenet
- To test that everything works, run
./test.sh