Eero Siivola

The solution @apaleyes posted is maybe not the most optimal one due to it being rather slow. The solution is based on sequential sampling from the posterior and conditioning the...

@apaleyes it needs to be reset always before the optimization routine starts. The reset is done externally always before the acquisition function is being optimized (here: https://github.com/EmuKit/emukit/blob/085291df26962d5579fe8834c98cd09208adf1eb/emukit/examples/preferential_batch_bayesian_optimization/pbbo/acquisitions/acquisition_function.py#L111). I think it...

To add, the acquisition functions used in the example are not exactly as in https://arxiv.org/abs/1704.03651 However, the original paper has limitations in the scalability w.r.t. the number of attributes in...

The IBO approach of Brochu only presents one acquisition strategy which is very exploitative. Some consider this to be a limitation as well (Which Gonzalez et al. solve in their...

So if you only need to compare two items (and not a batch of items) AND your input is not very high dimensional, the approach of Gonzalez et al. might...

This is the only open source implementation of the paper by Gonzalez. et al. that I know of: https://github.com/fcole90/minimal_pbo However, they do not implement any acquisition function, but it might...

Knowing your application, if you can also ask the user for optimal knob position given that other parameters are fixed, you might also be interested in this upcoming ICML 2020...

Hi! Yes, exactly! If so, the whole optimization process would be much faster than with a single preferential query. They actually are inputting the "optimal knob position" as multiple preferential...