[WIP] Random Feature-based CES

[odunbar]

Purpose

Adds the ability to use (Scalar-valued) RF with uncertainty in place of GP within CES. using RandomFeatures.jl

• See Random Feature development #164 for a statement of need.

TODO

• higher dim example
• inteface for passing in hparam algorithm options
• ...

Current content

• Interfaces with the currently unregistered RandomFeatures.jl (thus users will need to dev CalibrateEmulateSample.jl and RandomFeatures.jl locally rather than using the package management systems)
• adds ScalarRandomFeatureInterface as a MachineLearningTool
• new example examples/Emulator/RandomFeature/optimize_and_plot_RF.jl

RF results

Results show similar means, and RF has smaller (but not dissimilar) uncertainties in forecasts

GP results

from examples/Emulator/GaussianProcess/plot_GP.jl

• new example examples/Lorenz/Lorenz_example_RF.jl

RF results

We see parameter uncertainties slightly inflated but not too dissimilar compared to GP.

GP results

from examples/Lorenz/Lorenz_example.jl (with GP)

[codecov]

Codecov Report

Base: 88.65% // Head: 65.02% // Decreases project coverage by -23.63% ⚠️

Coverage data is based on head (c8252c1) compared to base (8980d60).
Patch coverage: 0.00% of modified lines in pull request are covered.

Additional details and impacted files

@@             Coverage Diff             @@
##           master     #183       +/-   ##
===========================================
- Coverage   88.65%   65.02%   -23.64%     
===========================================
  Files           4        5        +1     
  Lines         388      529      +141     
===========================================
  Hits          344      344               
- Misses         44      185      +141     

Impacted Files	Coverage Δ
src/Emulator.jl	93.63% <ø> (ø)
src/RandomFeature.jl	0.00% <0.00%> (ø)

Help us with your feedback. Take ten seconds to tell us how you rate us. Have a feature suggestion? Share it here.

☔ View full report at Codecov.
📢 Do you have feedback about the report comment? Let us know in this issue.