search

hackingmaterials / rocketsled

plug-n-play black box optimizer for high-throughput computing
Other
43 stars 11 forks source link

Consider "backend GP" for uncertainty estimation on non-GP models #68

Open ardunn opened 5 years ago

ardunn commented 5 years ago

A GP can be trained on the residuals of predictions in order to estimate uncertainty. This may be superior to bootstrap estimation in terms of computation time at least.

ardunn commented 4 years ago

alternatively, just additionally fitting a gp and using the uncertainty from that.