Xavier Bouthillier

Results 83 issues of Xavier Bouthillier

## 🐛 Bug I am trying to execute the visual bert example with VQA2 but one of the datasets seems to be corrupted. I can download test2015.tar.gz, trainval2014.tar.gz and coco_val2017.tar.gz...

Since release v0.6.0 of pymoo yesterday (https://github.com/anyoptimization/pymoo/releases/tag/0.6.0), HEBO's acq_optimizers fail to import because of an ImportError during the module's own imports: ``` tests/unittests/algo/long/hebo/test_hebo.py:9: in from hebo.models.model_factory import model_dict .tox/algo/lib/python3.7/site-packages/hebo/__init__.py:10: in...

See issue here: https://github.com/huawei-noah/HEBO/issues/19

This [test](https://github.com/Epistimio/orion/blob/develop/tests/functional/algos/test_algos.py#L241) often fails with TPE because it converges faster than the other builtin algorithms and often end up suggesting 2 times the same trial (not exactly the same but...

bug

@Delaunay made a PR (https://github.com/facebookresearch/hydra/pull/2050) on Hydra to add a plugin for Oríon but the maintainers decided to add it to the contrib folder. This means the plugin is not...

documentation

The functions executed by the executor may try to spawn subprocesses. For instance, if the function needs to do wandb.init(), it will try to spawn a subprocess. We need to...

When the working-dir is not specified by the user, a temporary directory is created for each runner (each `orion hunt` call) [[ref](https://github.com/Epistimio/orion/blob/develop/src/orion/client/experiment.py#L782)]. The trial.exp_working_dir is set to this directory when...

bug

If there is an error during the attempt of building a builtin prior (methods implemented in SpaceBuilder), the builder assumes the expression is not a builtin prior and then tries...

enhancement

The users script may not work properly if the config file does not have the expected extension. We should keep the same one instead of turning it into .conf. @Hnekoeiq

bug
low

Users need a way to checkpoint their current storage and restore it. This would also allow migrating from one database backend to another, for example checkpointing from pickleddb, and then...