This PR adds the feature of active learning meaning bayesian optimization with pure exploration to the repo. It includes a new predictive strategy named ActiveLearningStrategy that uses the acquisition function [qNegIntegratedPosteriorVariance]link to choose new candidate points depending on how strongly they will reduce the variance of the posterior. Like proposed in this issue.
This new strategy can handle single and multiple output models. For multiple output models specific weights for each output variable can be supplied.
Furthermore, a tutorial notebook is provided that illustrates the usage of the ActiveLearningStrategy for single and multiple output cases. Please check the code for further adjustments. Hopefully, we can merge the PR soon.
This PR adds the feature of active learning meaning bayesian optimization with pure exploration to the repo. It includes a new predictive strategy named
ActiveLearningStrategy
that uses the acquisition function[qNegIntegratedPosteriorVariance]
link to choose new candidate points depending on how strongly they will reduce the variance of the posterior. Like proposed in this issue.This new strategy can handle single and multiple output models. For multiple output models specific weights for each output variable can be supplied.
Furthermore, a tutorial notebook is provided that illustrates the usage of the
ActiveLearningStrategy
for single and multiple output cases. Please check the code for further adjustments. Hopefully, we can merge the PR soon.