WIP: Add basic Polynomail Chaos Expansion implementation

[FriesischScott]

Can compute PCE using least squares minimization or full quadrature rules using legendre basis functions.

The code should be fully extendible to other basis function, e.g. hermite by overloading a few methods in pcebases.jl.

Unfortunately SparseGrids.jl doesn't seem to support mixing nodes/weights for different families of polynomials. Having sparse quadrature rules will be very important going forward.

I'll add hermite polynomials before merging this.

[codecov[bot]]

Codecov Report

Merging #62 (ee0dd9b) into master (97fc031) will increase coverage by 0.59%. The diff coverage is 100.00%.

@@            Coverage Diff             @@
##           master      #62      +/-   ##
==========================================
+ Coverage   98.39%   98.98%   +0.59%     
==========================================
  Files          16       18       +2     
  Lines         373      590     +217     
==========================================
+ Hits          367      584     +217     
  Misses          6        6              

Impacted Files	Coverage Δ
src/models/pce/pcebases.jl	100.00% <100.00%> (ø)
src/models/pce/polynomialchaosexpansion.jl	100.00% <100.00%> (ø)
src/sensitivity/sobolindices.jl	100.00% <100.00%> (ø)
src/simulations/subset.jl	100.00% <0.00%> (ø)
src/reliability/probabilityoffailure.jl	100.00% <0.00%> (ø)
src/models/externalmodel.jl	98.07% <0.00%> (+1.30%)	:arrow_up:

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

[FriesischScott]

I need to find a few more examples to check hermite polynomials and also hybrid bases.

[AnderGray]

Looks great to me, a very nice contribution and implementation. Only (minor) suggestion is that evaluate! could allow for raw samples as well as a DF. But we would have to change this for the evaluate functions

[FriesischScott]

Looks great to me, a very nice contribution and implementation. Only (minor) suggestion is that evaluate! could allow for raw samples as well as a DF. But we would have to change this for the evaluate functions

Maybe create an issue for it so we can discuss this generally?