All notable changes to this project will be documented in this file.
The format is based on Keep a Changelog.
Fixed bug with PL (Thanks @kingjr).
Use job id based seed to avoid systematic failures with port allocation for distributed.
Remove automatic export of WORLD_SIZE inside submitit job target,
use dora.distrib.set_distrib_env if you relied on it.
Fixed version_base parameter support that appeared in Hydra.
Updated and simplified PyTorch Lightning distributed integration. Improved overall integration with PL, in particular with PLLogProgress and simplified Dora logger.
Adding HiPlot support out of the box.
Fixed bug with nested grid searches.
Set use_rendezvous=False by default.
More reliable passing of arguments of Hydra (before, setting None would actually fail). I hope this wont break any existing XP sig...
Allow for empty mem constraint in Slurm.
Fixing callbacks default value in PL.
Extra "keys" in Hydra config files are now allowed (i.e. overrides with +something=12).
The package where Dora looks for grids can be customized, in Hydra with dora.grid_package in the base config or passing grid_package='...' to argparse_main.
Better doc for launcher API.
Fix dict support with Hydra. Okay it is time that I release a new version now...
Reliable rmtree used to avoid --clear being blocked by some locking issues on NFS.
Fix bug with PL.
Early deletion of rendezvous file to avoid errors on job requeue. This might actually lead to bugs in the future as this is not officially supported but from a discussion with PyTorch engineers, "it should be okay".
Actually, because rendezvous file are not that reliable, added using Slurm to find
the master addr. Port is decided based on the XP signature (running twice the same XP on the
same machine will crash, but anyway this is probably a bad idea). Set dora.use_rendezvous: false to test out. This will soon become the default value.
Always export RANK and WORLD_SIZE as env variable, so that they can be consumed by Hydra config resolver.
dora.log.LogProgress.update now returns True if logging will happen at the end of the iteration.
Add silent option for grid API, which suppress all printing.
Adding process_sheep method in Explorer, that can replace process_history and provide access to the sheep and XP (sheep.xp)
in order to allow for processing that can depend on the config of the XP.
Automatically simplifies argv list for Hydra experiments when the same parameter is repeated multiple times.
Better error message when making a typo in the grid name. Always show the traceback when getting an import error.
Easier sharing of XP hyper params. Added import/export command to easily share XP hyper-params in text form.
Added shared repository option (shared option in Dora config). No metrics or
checkpoints can be shared, this is still a bit dangerous, but this will act as a shared
database for mappings from SIG -> hyper params, so that you can just pass a SIG
to your teammate and launch the same XP. See the README section on sharing for more details.
Adding support for type arrays.
Disabling automatic loading of PyTorch Lightning if installed, as this trigger a warning with distutils/setuptools.
Add py.typed file to source distribution. Fixed bug in LogProgress for very slow speed.
Added possiblity to log always the time per iteration rathen than iterations per seconds. PR.
Fixed a bug with DoraConf, making sure that the dir attribute is always
absolute, even when reset after creation, which could happen when using Hydra
with a relative path for dora.dir.
Initial release (first versions were private betas).