Right now we have multiple ways to run an experiment with sheeprl, namely:
lightning run model sheeprl.py COMMAND ...
python sheeprl.py COMMAND ...
It would be nice to have a single entrypoint so to launch every experiment with python sheeprl.py exp=..., where the exp= specifies the experiment config (predefined configs can be found under the sheeprl/configs/exp/ directory). This let us decorate only the entrypoint with the @hydra.main() decorator.
The only requirement is that algo.name=algorithm_name, where algorithm_namecomes from _algorithmname.py, where algorithm_name.py is the file containing the algorithm implementation (it must contain at least one method decorated with the register_algorithm decorator)
belerico
commented
11 months ago
Right now we have multiple ways to run an experiment with sheeprl, namely:
lightning run model sheeprl.py COMMAND ...python sheeprl.py COMMAND ...It would be nice to have a single entrypoint so to launch every experiment with
python sheeprl.py exp=..., where theexp=specifies the experiment config (predefined configs can be found under thesheeprl/configs/exp/directory). This let us decorate only the entrypoint with the@hydra.main()decorator.The only requirement is that
algo.name=algorithm_name, wherealgorithm_namecomes from _algorithmname.py, wherealgorithm_name.pyis the file containing the algorithm implementation (it must contain at least one method decorated with the register_algorithm decorator)