SMTorg / smt / 25323246775

90%

Build Type

push

github

Committed by web-flow

Commit Message

Implement specialized predict methods in MFCK/SMFCK (#809)

Addition of wrapper functions to predict the conditional mean and
conditional variance in MFCK.

- predict_values()
- predict_variances_all_levels()
- predict_variances()

The function now returns the predictions in a numpy array instead of
returning them in a list (this change keeps consistency with MFK).
Useful for benchmarks where you want to compare the predictive
performance with (MFK, SMFK).

Update of the notebooks of SMFK, MFCK, and SMFCK with the last update;
the notebooks had some problems regarding the predictive mean.

-In SMFCK is now necessary to define "theta0" exactly as in MFK. You
need to define it in the same dimensional space of your inputs (before,
you just needed to put an initial value, and MFCK automatically assigned
the same initial theta to all the dimensions)

-As SMFK integrates sparse approximations for some experiments is
necessary to increase the nugget (near 1e-10) to guarantee the inversion
of the matrix.

---------

Co-authored-by: Rémi Lafage <remi.lafage@onera.fr>

Coverage Stats

38 of 67 new or added lines in 1 file covered. (56.72%)

10 existing lines in 1 file now uncovered.

7825 of 8718 relevant lines covered (89.76%)

0.9 hits per line