shahineb / collider-regression Public

Notifications You must be signed in to change notification settings
Fork 2
Star 2

This repo contains material for the paper "Returning the Favour: When Regression Benefits from Probabilistic Causal Knowledge" ICML 2023 paper

arxiv.org/abs/2301.11214

2 stars 2 forks Branches Tags Activity

Star

Notifications

Branches Tags

Name		Name	Last commit message	Last commit date
Latest commit History 37 Commits
config		config
notebooks		notebooks
repro		repro
src		src
utils		utils
.gitignore		.gitignore
README.md		README.md
aggregate_grid_search_results.py		aggregate_grid_search_results.py
aggregate_hyperparms_search_results.py		aggregate_hyperparms_search_results.py
run_hyperparams_search_kernel_model_rotation_data.py		run_hyperparams_search_kernel_model_rotation_data.py
run_kernel_model_FaIR_data.py		run_kernel_model_FaIR_data.py
run_kernel_model_mvn_data.py		run_kernel_model_mvn_data.py
run_kernel_model_rotation_data.py		run_kernel_model_rotation_data.py

Repository files navigation

rkhs-collider

Running a grid search experiment

Setup configuration file of the experiment + search grid under config/search/my_config.yaml
Run grid search script. For example, KRR with polynomial data generating process would be run as

$ python run_grid_search_kernel_model_polynomial_data.py --cfg=config/search/my_config.yaml --o=my/output/directory/

Aggregate search results together into an xarray dataset

$ python aggregate_grid_search_results.py --i=my/output/directory --o=another/output/directory

This last script will dump a cv-search-scores.nc file which can then be loaded in the notebook notebooks/score-analysis.ipynb to visualise metrics against searched hyperparameters values.