the overall framework is set up, the ActiveLearningStrategy inherits from the BotorchStrategy and can be used in combination with AnyActiveLearningAcquisitionFunction, which is currently only qNegIntPosVar and can be only used with one output feature.
to make it actually working you have to implement the _get_acqfs method within the functional model of the ActiveLearningStrategy.
feel free to add more acqfs, if you want to use the ones from @hvarfner PR https://github.com/pytorch/botorch/pull/2163, you have to add a validator that they can be used only in combination with a SaasSingleTaskGPSurrogate surrogate.
I would recomment that you just create a branch based on this branch and then create a new PR. Thank you very much for your support.
This PR is still work in progress and implements an
ActiveLearningStrategyinBoFireas discussed in issue https://github.com/experimental-design/bofire/issues/331.@jdridder: this is my current status:
ActiveLearningStrategyinherits from theBotorchStrategyand can be used in combination withAnyActiveLearningAcquisitionFunction, which is currently onlyqNegIntPosVarand can be only used with one output feature._get_acqfsmethod within the functional model of theActiveLearningStrategy.SaasSingleTaskGPSurrogatesurrogate.I would recomment that you just create a branch based on this branch and then create a new PR. Thank you very much for your support.