Helmholtz-AI-Energy / Helmholtz-AI-Energy/propulate
Unintended restarting from checkpoint
- Dominant language
- Python
- Stars
- 46
- Forks
- 9
- PR merge metrics
- No merged PRs in 30d
Description
It would very beneficial to have a flag for if the fold given is to be used as holding checkpoints or not.
Furthermore, I think it would be great to have a unique name for the runs and save them into the folder if the user does not was a checkpoint.
example:
```python
islands = propulate.Islands(
objective,
propagator,
rng,
generations=num_generations,
num_islands=islands,
migration_probability=0.9,
resume=False,
checkpoint_path=Path("/dir/for/checkpoints/"),
)
```
This would stop the run from trying to load any checkpoints, and it would create a directory at `/dir/for/checkpoints/unique-name` which is where the new checkpoints will be saved
Contributor guide
Research direction
Start at the propulate.Islands entry point and trace how resume and checkpoint_path are currently handled. Define the expected behavior for resume=False and unique run directories, then add coverage showing that existing checkpoints are not loaded and new checkpoints are saved in the selected run directory.
Written by the indexing model from the issue text.
Assessment
- Tech stack
- python
- Domain
- distributed-systems
- Issue type
- Feature
- Difficulty
- 4/5
- Estimated time
- 3-5 days
- Activity status
- Stale
- Clarity
- Mostly clear
- Newbie friendliness
- 35/100