Simon Blanke

Performance in high dimensional search-spaces

1

The issue #49 showed a slowdown of optimization algorithms in high dimensional (>1000) search-spaces. So this is aiming at improving this performance (for now) in random-search and hill-climbing. If I...

Add space-filling curves algorithm to grid-search

5

One drawback of the grid-search algorithm is its lack of exploration of the search-space if the number of iterations is much smaller, than the search-space. The grid-search begins in a...

This would add another API "level", which enables more control over the optimization run and will be closer to the original mathematical idea of gradient-free-optimization. This API could improve the...

[ENH] change `_MetaObjectMixin._check_objects` default for `attr_name` to auto-detect

7

#### Reference Issues/PRs Closes #159 #### What does this implement/fix? Explain your changes. Sets `attr_name` to `None` and auto detect from "named_object_parameters" of `attr_name` remains `None` #### Does your contribution...

[ENH] Centralize parallel backend selection in skbase and introduce typed config class

2

**Is your feature request related to a problem? Please describe.** There is no parallel backend API in skbase today, while sktime, skpro, and Hyperactive each carry their own backend selection...

[ENH] pretraining and global forecasting API for forecasters

3

Since this is a big and complex feature to implement I targeted the pretrained-forecasters branch for merging to keep the "main" clean. With more progress on the implementation the viability...

[BUG] set low_memory and maxiter for tests of varmax (fixes #6260)

1

#### Reference Issues/PRs fixes #6260 #### What does this implement/fix? Explain your changes. sets low_memory to True and sets a small value for maxiter to avoid memory failure: https://github.com/sktime/sktime/issues/6260#issuecomment-3606253000 ####...

#### Reference Issues/PRs fixes issue #7921 #### What does this implement/fix? Explain your changes. Just fixes the test-output as described in https://github.com/sktime/sktime/issues/7921#issuecomment-3605956022 #### Does your contribution introduce a new dependency?...

[BUG] Data leakage in `TorchExperiment`

1

The `TorchExperiment` reuses the same `LightningDataModule` instance across all optimization trials. If the datamodule maintains internal state (e.g., data augmentation settings, random state, preprocessing cache), this state persists across trials....

[ENH] Allow `TorchExperiment` metric to set "higher is better"

1

The `TorchExperiment` assumes all metrics should be minimized (lower is better) by default. This can be seen [here in this line](https://github.com/SimonBlanke/Hyperactive/blob/main/src/hyperactive/experiment/integrations/torch_lightning_experiment.py#L114). However, many metrics should be maximized (accuracy, F1, AUC),...