Franck Charras

Add return_generator functionality

93

This PR is derived from #587, it keeps only the part that enable `__call__` to return a generator. Some additional notes: - the keyword is now named `return_generator` to match...

[Do not merge] empty pr for pipelines

1

Going to add a few empty commits to this PR to repeat current master pipelines and hunt failures.

ENH Use Array API in `r2_score`

10

#### Reference Issues/PRs The PR builds on preliminary explorations done by @elindgren in https://github.com/scikit-learn/scikit-learn/pull/27102 It tackles one of the items outlined in https://github.com/scikit-learn/scikit-learn/issues/26024. #### Any other comments? This PR proposes...

DOC Add note in Array API doc regarding support for devices without float 64 support

2

#### What does this implement/fix? Explain your changes. Adds a note in the array api documentation that documents scikit-learn policy regarding support of devices that do not support float64 precision...

WIP: add an example of self sustaining parallel compute

1

I initiated the `return_as` project with this particular use-case in mind: creating a compute model where the outputs of the tasks that are processed in parallel are used to feed...

KNN on max series seems slower than cuda-based implementation on comparable devices ?

4

Initial report contained an error, please follow through the first comment for a better explanation. ```python import numpy as np from sklearn.neighbors import NearestNeighbors import sklearn device = " #...

KMeans is slow on gpu

10

The following snippet ```python import numpy as np import sklearn device = "cpu" # device = "gpu:0" from sklearnex import patch_sklearn patch_sklearn() sklearn.set_config(target_offload=f"{device}") from sklearn.cluster import KMeans seed = 123...

[Feature request] Could it be possible to customize lookup paths in the CLI

1

Hello, thanks for the benchopt initiative, it is great and after I started using following @tomMoral advices I've found the main concepts really well articulated and I'm adopting it. One...

[Feature request] Allow arbitrary naming of datasets, objectives and solvers classes ?

1

This is mostly a cosmetic suggestion, currently benchopt expects that the classes are named like this: ```python class Solver(BaseSolver): ... ``` ```python class Objective(BaseObjective): ... ``` ```python class Dataset(BaseDataset): ......

[Feature Request] CLI that prints all datasets, objectives, solvers, and available parameters, parameters than run by defaults, and accepted values for each parameters

2

For instance `benchmark run --help` could print ``` Datasets - my_first_dataset: - n_samples: desc: how many samples in the dataset - default: [50, 100, 1000] - valid values: N/A -...