Ecosyste.ms: Awesome
An open API service indexing awesome lists of open source software.
https://github.com/inoueakimitsu/tune-hve
Hard Vicinal Estimates Hyperparameter Tuning Tool
https://github.com/inoueakimitsu/tune-hve
Last synced: about 2 months ago
JSON representation
Hard Vicinal Estimates Hyperparameter Tuning Tool
- Host: GitHub
- URL: https://github.com/inoueakimitsu/tune-hve
- Owner: inoueakimitsu
- License: mit
- Created: 2022-12-01T02:16:45.000Z (about 2 years ago)
- Default Branch: main
- Last Pushed: 2023-11-16T00:29:40.000Z (about 1 year ago)
- Last Synced: 2024-10-12T04:46:36.066Z (3 months ago)
- Language: Python
- Size: 91.8 KB
- Stars: 0
- Watchers: 1
- Forks: 0
- Open Issues: 0
-
Metadata Files:
- Readme: README.md
- License: LICENSE
Awesome Lists containing this project
README
# tune-hve
Hard Vicinal Estimates Hyperparameter Tuning Tool
## Summary
The $\hat{p}_{g}^{\mathrm{HVE}} (\boldsymbol{x}, y)$ in CcGAN is
influenced by the settings of $\kappa$ and $\sigma$.
If $\kappa$ or $\sigma$ are too small, the vicinity is not taken into account.
Given $y^r$, $\kappa$, and $\sigma$,
this tool visualize the range covered by the the second term of
$\hat{p}_{g}^{\mathrm{HVE}} (\boldsymbol{x}, y)$.
This gives the setting guidelines for $\kappa$ and $\sigma$.
## Approach
For the empirical distribution of labels on real data, we do the following:
- First, convolve a Gaussian kernel with bandwidth $\sigma$.
- Next, we convolve a box kernel with bandwidth $\kappa$.
- This visualizes the range covered by the labels in the real data.
## Usage
```
usage: tunehve [-h] [--max MAX] [--min MIN] [--kappa KAPPA] [--sigma SIGMA] data output
Tune HVE
positional arguments:
data data file
output output file
optional arguments:
-h, --help show this help message and exit
--max MAX max value
--min MIN min value
--kappa KAPPA kappa
--sigma SIGMA sigma
```
The following commands can be run to check the behavior for unbalanced age data.
```bash
python tunehve/hve.py testdata/y1.csv out.png --min 0 --max 90 --kappa 2 --sigma 0.3
```
## License
tune-hve is available under the MIT License.
## Reference
1. Ding, Xin, et al. "CcGAN: continuous conditional generative adversarial networks for image generation." International Conference on Learning Representations. 2020.