search

EmuKit / emukit

A Python-based toolbox of various methods in decision making, uncertainty quantification and statistical emulation: multi-fidelity, experimental design, Bayesian optimisation, Bayesian quadrature, etc.
https://emukit.github.io/emukit/
Apache License 2.0
605 stars 128 forks source link

Nonlinear Multi-Fidelity Problem #445

Closed pzhcdh closed 1 year ago

pzhcdh commented 1 year ago

In https://github.com/EmuKit/emukit/blob/main/notebooks/Emukit-tutorial-multi-fidelity.ipynb, it is only mentioned that linear problems are only modeled using high-precision data. How can nonlinear problems be modeled using only high-precision data? It would be great if I could get a reply.

apaleyes commented 1 year ago

Hi! Can you clarify which section of the notebook you are referring to? There is nothing inherently special about linear/non-linear in that respect. In both cases it is expected that high fidelity data is of better quality.

pzhcdh commented 1 year ago

In the linear problem, there are results of using only high-precision data for GPR for comparison, but for nonlinear problems, there is no result of using only high-precision data for GPR. I would like to ask how to use only high-precision data for GPR in nonlinear cases, so that it can be seen that the low-precision data has improved the results.

apaleyes commented 1 year ago

Assuming you are referring to the plot in section "1.2 Comparison to standard GP", there is no reason same cannot be done with non-linear model.

apaleyes commented 1 year ago

I am going to assume this question was resolved.