Data

[alecandido]

Roadmap for closure and actual data implementation.

• 0: PDF level fake data (PDF uncertainty)
• 1: closure test data through actual DIS theory (PDF uncertainty)
• 2: actual data (actual covmat)

[alecandido]

At PDF level it is done for:

• MC replicas sets: i.e. NNPDF ones (essentially), this is the easiest: just sample x and flavor with a sensible distribution, and then sample replica by sampling uniformly its ID
• hessian sets: i.e. everything else, in this case in order to associate each sample to a "replica", I had sample a point in a space with dimension equal to the number of eigenvectors, from a multi-Gaussian, then combine the eigenvectors in that direction, and sample x and flavor with the same distribution of above

I believe this is fine, but it's much messier to sample from hessian sets, so everything else will be done for MC sets only.

[alecandido]

For actual data I made a leap forward, since Roy explained me how to use vp.api.API 😃