kernel-quantile-regression 0.0.15

Kernel quantile regression

Kernel quantile regression

The kernel_quantile_regression package is an open source implementation of the quantile regressor technique introduced in [1].

This repo contains the code for reproducing the research paper Probabilistic energy forecasting through quantile regression in reproducing kernel Hilbert spaces. This article has been published in the ACM SIGEnergy Energy Informatics Review.

Example of kernel quantile regression on the Melbourne temperature data [2].

Installation

Use the package manager pip to install kernel_quantile_regression.

pip install kernel-quantile-regression

Usage

from kernel_quantile_regression.kqr import KQR

# create model instance
# specify your quantile q and hyperparameters C and gamma
kqr_1=KQR(alpha=q, C=100, gamma=0.5)

# train model
kqr_1.fit(X_train, y_train)

# predict
kqr_1.predict(X_test)

Repo files

• Data/ The Data directory contains the raw files for the GEFCom2014 challenge [3], data can be accessed from Dr. Tao Hong blog http://blog.drhongtao.com/2017/03/gefcom2014-load-forecasting-data.html. The Data folder contains also the transformed raw data, those constitute the input for our probabilistic forecasting study.

• plots/ Plots for the tutorial and experiments.

• src/kernel_quantile_regression Source code.

• train_test scripts to train the models, saved and test them.

  • models contains , for each quantile, the pickled trained models.

• utils Utility functions for extracting, loading and transforming raw data of the GEFCom2014 challenge.

• kqr_tutorial.py Getting started example, where our method is compared against other valid quantile regressors.

References

[1] Ichiro Takeuchi, Quoc V. Le, Timothy D. Sears, and Alexander J. Smola. 2006. Non- parametric Quantile Estimation. Journal of Machine Learning Research 7, 45 (2006), 1231–1264. https://www.jmlr.org/papers/volume7/takeuchi06a/takeuchi06a.pdf

[2] Rob J Hyndman, David M Bashtannyk, and Gary K Grunwald. 1996. Estimating and visualizing conditional densities. Journal of Computational and Graphical Statistics 5, 4 (1996), 315–336. https://www.jstor.org/stable/1390887

[3] Tao Hong, Pierre Pinson, Shu Fan, Hamidreza Zareipour, Alberto Troccoli, and Rob J.Hyndman. 2016b. Probabilistic energy forecasting: Global Energy Forecasting Competition 2014 and beyond. International Journal of Forecasting 32, 3 (2016), 896–913. https://www.sciencedirect.com/science/article/abs/pii/S0169207016000133?via%3Dihub

License

MIT