Sensitivity analysis

[mastoffel]

Global sensitivity analysis using sobol indices, and plotting functions.

[mastoffel]

@MaxBalmus @aranas just tagging you because of the tutorial you are writing. Feel free / let me know if you'd like to give comments / suggestions / review. Otherwise I'd merge it tomorrow or Thursday. Here's how it will work:

This is how it currently works:

em = AutoEmulate()
em.setup(X, y)
em.compare()

emulator = em.get_model("gp")

# sensitivity analysis
problem = {
        'num_vars': 2,
        'names':  ['x1', 'x2]
        'bounds': [[0.1, 0.5]. [0.01, 0.2]]
}
Si = sensitivity_analysis(mod, problem,)
plot_sensitivity_analysis(Si, index="S1")

[MaxBalmus]

At a glance it looks great. I will try to look more in depth.

Just a small feature request (sorry if is is a bit pedantic): since we already have X defined, and that is usually a data frame or derived from a dataframe, can we have a problem_generator function that read X and writes problem for you? The bounds could be derived from the min/max of X.

[mastoffel]

At a glance it looks great. I will try to look more in depth.

Just a small feature request (sorry if is is a bit pedantic): since we already have X defined, and that is usually a data frame or derived from a dataframe, can we have a problem_generator function that read X and writes problem for you? The bounds could be derived from the min/max of X.

Yes, that sounds good!

[codecov-commenter]

Codecov Report

Attention: Patch coverage is 78.87931% with 49 lines in your changes missing coverage. Please review.

Project coverage is 92.45%. Comparing base (48129e4) to head (580f340).

Files with missing lines	Patch %	Lines
autoemulate/sensitivity_analysis.py	62.03%	41 Missing ⚠️
autoemulate/compare.py	38.46%	8 Missing ⚠️

Additional details and impacted files

@@            Coverage Diff             @@
##             main     #260      +/-   ##
==========================================
- Coverage   94.26%   92.45%   -1.82%     
==========================================
  Files          58       62       +4     
  Lines        3313     3617     +304     
==========================================
+ Hits         3123     3344     +221     
- Misses        190      273      +83     

☔ View full report in Codecov by Sentry.
📢 Have feedback on the report? Share it here.

[github-actions]

Coverage report

Click to see where and how coverage changed

File	Statements	Missing	Coverage	Coverage (new stmts)	Lines missing
autoemulate
compare.py					557-566, 588
sensitivity_analysis.py					44-49, 57, 60, 62, 67, 95, 127-130, 171, 227-241, 264-304
tests
test_sensitivity_analysis.py
Project Total

This report was generated by python-coverage-comment-action

[mastoffel]

Ok, so now if no arguments are given to sensitivity analysis, it just takes the best model from cv, refits on the full data and infers the problem definition as @MaxBalmus suggested. However, possible to do all this manually too.

em = AutoEmulate()
em.setup(X, y)
em.compare()

si = em.sensitivity_analysis()
em.plot_sensitivity_analysis(si)