Skip to content

Latest commit

 

History

History
20 lines (12 loc) · 1.18 KB

README.md

File metadata and controls

20 lines (12 loc) · 1.18 KB
Raw

pseudo-labelling scripts from Humphrey et al. (2022, MNRAS, submitted)

Scripts used to create results in our MNRAS paper on application of pseudo-labelling to estimation of galaxy properties

If you're not using a GPU, you'll need to modify a keyword for CatBoost, LightGBM, and XGBoost:

  • for CatBoost, remove or change the task_type keyword (if not defined, it will default to CPU);

  • for XGBoost, change tree_method to hist or exact. The latter can be a lot slower, but often gives slightly higher quality results;

  • for LighGBM, remove or change the device keyword (if not defined, it will default to CPU).

Depending on the computer you use, you may also need to reduce the max_depth keyword for some of the learning algorithms.

If you don't have the Intel extension to Sxikit-Learn installed, or don't have an Intel CPU, you'll also need to delete/comment out these lines:

from sklearnex import patch_sklearn
patch_sklearn("knn_regressor")
patch_sklearn('random_forest_regressor')

If you do have an Intel CPU, I really recommend installing the Intel sklearnex package (and an Intel version of Python 3) due to the dramatic speed increases you'll get in some functions.