optuna

optuna copied to clipboard

Hyperband bracket assignment using hash in parallelization

19

I think there could be a problem when using Hyperband pruner in parallelization. In Hyperband, each trial is assigned to the bracket by computing hash("{}{}".format(study.studyname, trial.number). However, the hash gives...

donny8

Dask / Optuna Integration

30

Hello Optuna maintainers, I work on Dask and I would like to improve integration between Optuna and Dask. I did a brief investigation a couple of days ago (see https://github.com/dask/dask/issues/6571)...

mrocklin

Print `study.user_attr` by `optuna studies` command

10

### Motivation Using the cli to query studies for their user_attr would be quite useful. There doesn't seem to be a way to do this, though there does seem to...

blazespinnaker

Avoid PyTorch version mismatch between BoTorch and Optuna.

1

### Motivation BoTorch used in [the integration module of Optuna](https://optuna.readthedocs.io/en/stable/reference/generated/optuna.integration.BoTorchSampler.html) has dropped Python 3.7 support since its version 0.9.0. With this change, the PyTorch version specified by Optuna will not...

nzw0301

Failing assertion left in 'unreachable line' breaks HyperbandPruner

9

### Expected behavior Handle the error gracefully. For instance, remove the assertion and instead let `_get_bracket_id` return zero (most frequently returned value of `bracket_id`). Alternatively, introduce a debug flag to...

mirekphd

Add a tutorial of Wilcoxon pruner

1

## Motivation I want to add a tutorial of wilcoxon pruner. ## Description of the changes Added a 013_wilcoxon_pruner.py file.

eukaryo

Enhance performance of GPSampler

5

## Motivation This PR enhances the performance of `GPSampler` to the same level of (if not above) BoTorch. ## Description of the changes This PR introduces the following changes: *...

contramundum53

Fixing potential unbound local variable in `Study._optimize_sequential`

8

### Motivation `Study._optimize_sequential` encounters an issue if `_run_trial` errors and callbacks are present, as seen in the following relevant code block: ```python try: frozen_trial = _run_trial(study, func, catch) finally: #...

lumip

TPESampler constant_liar=True with constrained optimization spams "Trial does not have constraint values. It will be treated as a lower priority than other trials."

5

### Expected behavior When I don't pass in `constant_liar=True` this warning doesn't show up, and I don't think having `constant_liar=True` should be warning either. Also when I rerun the same...

rzhang-at-hrt

Introduce lazy greedy algorithm to `_solve_hssp`

3

### Motivation Practically speed-up `_solve_hssp()` using lazy greedy algorithm. ### Description The current implementation of [`_solve_hssp()`](https://github.com/optuna/optuna/blob/2f279f321664bf8793914f442ca13b1b97495b2b/optuna/_hypervolume/hssp.py#L8) relies on the greedy algorithm for submodular maximization problems with a cardinality constraint. the...

Alnusjaponica