·
24 commits
to master
since this release
This release encompasses a few different versions - the changes made to ParBayesianOptimization in the last few months are all being included in these notes.
Major Changes
- Gaussian Process package is now DiceKriging. Faster predictions and doesn't use S6 methods.
- bayesOpt() and its methods/functions has replaced BayesianOptimization().
- Added time limit, utility requirement, and improved other early stopping situations (errors, no unique parameters, etc).
- Iterations can be added to an existing bayesOpt object, and the parameters of the search (acq, gsPoints, etc.) can be changed.
Minor Changes
- Plotting has been improved. Now uses ggplot2 and includes S3 method for bayesOpt objects.
- Now includes a print method.
- Errors can be included in scoreSummary along with parameters and elapsed time to help debug problems in FUN.
- A set limit on the number of errors can be set before returning results (errorHandling parameter)
Meta Changes
- Improvements to Documentation
- Added comprehensive unit tests for successful and unsuccessful scenarios.