Juan M. Cruz-Martinez

Make the usage of feature scaling + large x a bit clearer

1

Closes #2377 This PR deals with the clarity issues pointed out by #2377 1. Allows the user to drop completely large-x from the runcard 2. If feature scaling is enabled,...

Test positivity at postfit time and with grids instead of fktables

1

Instead of using positivity fktables, at the postfit stage, we can use grids (that are then committed to this repository). This is not a very complicated task but still complex...

Make the diagonal fit the default

8

It has been decided that for 4.1 we will want the diagonal covmat to be the default way in which fits are run, see this comment: https://github.com/NNPDF/nnpdf/pull/2301#issuecomment-2922178907 (it would be...

We can check integrability directly looking at the PDF instead of creating ad-hoc grids. As @giacomomagni and @evagroenendijk just found out, the amount of things thst can go wrong is...

Dynamically create the evolution grid

2

At the moment we always have a fixed Q-grid for the evolution https://github.com/NNPDF/nnpdf/blob/2ee90763374d891277434bf462f5770e285619ed/n3fit/src/evolven3fit/q2grids.py#L127 However, now that we have slightly changed the bottom and charm mass, we need to change the...

When running in GPUs, if any of the replicas fail, the entire ensemble will fail and the fit needs to restart. Not sure how to approach this since I have...