Ecosyste.ms: Awesome
An open API service indexing awesome lists of open source software.
https://github.com/luthaf/farthest-sampler
https://github.com/luthaf/farthest-sampler
Last synced: 18 days ago
JSON representation
- Host: GitHub
- URL: https://github.com/luthaf/farthest-sampler
- Owner: Luthaf
- Created: 2021-02-18T16:21:01.000Z (almost 4 years ago)
- Default Branch: master
- Last Pushed: 2022-05-03T10:34:27.000Z (over 2 years ago)
- Last Synced: 2024-11-09T04:39:08.090Z (2 months ago)
- Language: Rust
- Size: 15.2 MB
- Stars: 2
- Watchers: 3
- Forks: 0
- Open Issues: 0
-
Metadata Files:
- Readme: README.md
Awesome Lists containing this project
README
# Farthest sampler
Some code to perform FPS selections of features or samples.
## Installation
```bash
pip install git+https://github.com/Luthaf/farthest-sampler
```
## Example
```py
import numpy as np
from farthest_sampler import select_fps_voronoi, select_fps_standard
points = np.load("my-file.npy") # or generate the data dynamically
assert len(points.shape) == 2 # only 2D array
assert points.dtype == np.float64 # only array of float64 values
selected_voronoi = select_fps_voronoi(
points,
300, # number of points to select
0, # first selected point
)
selected_standard = select_fps_standard(
points,
300, # number of points to select
0, # first selected point
)
assert np.all(selected_voronoi == selected_standard)
# selected_{voronoi,standard} contain the indexes of the selected points
```
## Performance
Here are the result of the benchmarks included in this repository on the
[Boston](https://scikit-learn.org/stable/modules/generated/sklearn.datasets.load_boston.html#sklearn.datasets.load_boston)
house price dataset (shape 506x13), and a dataset of SOAP features (shape
2520x1362).
```
group main/Standard FPS/ main/Voronoi FPS/
----- ------------------ -----------------
100 samples/Boston dataset 1.00 3.4±0.51ms 1.17 4.0±0.78ms
200 samples/Boston dataset 1.00 7.0±0.45ms 1.25 8.8±1.88ms
100 samples/SOAP 1.61 78.3±7.48ms 1.00 48.6±4.12ms
200 samples/SOAP 1.91 155.6±9.72ms 1.00 81.5±4.38ms
10 features/Boston dataset 1.00 171.6±28.89µs 1.41 242.5±41.00µs
100 features/SOAP 1.00 75.8±7.63ms 1.21 92.0±3.14ms
200 features/SOAP 1.00 153.6±18.42ms 1.17 179.2±35.07ms
```
Overall, VoronoiFPS is faster than the standard full algorithm when selecting
samples in a reasonably uniform space (i.e. SOAP feature space, but not the
Boston dataset).
To run these benchmarks for yourself, you'll need
[critcmp](https://github.com/BurntSushi/critcmp):
```bash
cargo bench -- --save-baseline main
critcmp main -g ".*FPS/(.*)"
```