search

GPflow / GPflowOpt

Bayesian Optimization using GPflow
Apache License 2.0
270 stars 61 forks source link

HvPoI #36

Closed javdrher closed 7 years ago

javdrher commented 7 years ago

Implementation of the HvPoI multi-objective optimization algorithm. Before this can be merged some steps are required.

I'm quite ok with the acquisition implementation, however I can't say the same about the Pareto class. This code is the result of two refactors based on port of the original MATLAB code. I think it already looks better but as several multi-objective strategies are based on hypervolumes, this makes that class quite significant in our framework. Its not up to standards yet. It also needs documentation. To improve this, I'm counting on you @icouckuy . Currently, our test coverage also misses 2 lines in pareto.py, but I understood the levels of improvement can be removed?

In addition, a notebook might be interesting also.

codecov-io commented 7 years ago

Codecov Report

Merging #36 into master will decrease coverage by 0.21%. The diff coverage is 98.87%.

Impacted file tree graph

@@            Coverage Diff             @@
##           master      #36      +/-   ##
==========================================
- Coverage   99.73%   99.52%   -0.22%     
==========================================
  Files          14        9       -5     
  Lines         757      838      +81     
==========================================
+ Hits          755      834      +79     
- Misses          2        4       +2
Impacted Files Coverage Δ
GPflowOpt/__init__.py 100% <100%> (ø) :arrow_up:
GPflowOpt/acquisition.py 100% <100%> (ø)
GPflowOpt/pareto.py 98.5% <98.5%> (ø)
GPflowOpt/bo.py 98.14% <0%> (-0.32%) :arrow_down:
GPflowOpt/domain.py 98.8% <0%> (-0.07%) :arrow_down:
GPflowOpt/transforms.py 100% <0%> (ø) :arrow_up:
GPflowOpt/optim.py 100% <0%> (ø) :arrow_up:
GPflowOpt/scaling.py 100% <0%> (ø) :arrow_up:
GPflowOpt/design.py 100% <0%> (ø) :arrow_up:
... and 9 more

Continue to review full report at Codecov.

Legend - Click here to learn more Δ = absolute <relative> (impact), ø = not affected, ? = missing data Powered by Codecov. Last update 2db20b7...9fe0451. Read the comment docs.

icouckuy commented 7 years ago

Next up: removing obsolute code in the cell bound calculation, and making it faster if possible.

icouckuy commented 7 years ago

While originally for GPflowOpt 0.2, we are thinking of merging this multiobjective acquisition function for 0.1:

If noone has objections I will merge it in a couple of days