feat: add conformal Bayesian prediction

lsorber / neo-ls-svm

Neo LS-SVM is a modern Least-Squares Support Vector Machine implementation

MIT License

19 stars 1 forks source link

Closed lsorber closed 8 months ago

lsorber commented 8 months ago

Adds asymmetric conformal Bayesian prediction for both classification and regression.

Details:

Asymmetric means we estimate the positive and negative residual quantiles separately.
Bayesian means the nonconformity measure is that of Gaussian Processes.
To construct the confidence interval we apply linear quantile regression on the leave-one-out residuals given the leave-one-out nonconformity measures.

lsorber commented 8 months ago

Test set performance on the Ames housing dataset: