Present: Enrico, Javi, Enric, Stefan, Stephan, Guilherme, Danilo, Stephan, Axel, Patrick, Muhammad

Decisions:

- We want both implementations proposed by Stefan, provided that we find a better name for AsDict and discuss further the flattening.

Additional News:

- On the 28th a RDF tutorial will take place for the Ship experiment.

RooFit optimisation plan and its status

ACTION: the alignement of data once the batch evaluation is implemented must be taken care of. Progress needs to be made in this direction while the solution is implemented.

- Binned Vs. Unbinned: the usage of both exists. For example LHCb use mainly unbinned fits while ATLAS/CMS binned. Both have big opportunities to reduce computational load linked to fits.

- It is observed that a vector<T*> may not be much better than a linked list since pointees would be scattered in memory. On the other hand, RooFit uses a memory pool that allocates these pointees in a contiguos region of memory. We'll see the overall effect once the upgrade is complete.

NumPy Interoperability

ACTION: figure out what we want to do in terms of flattening.

- The name AsDict is perhaps not optimal. We need to discuss this.

- The solutions proposed also allow to export arrays of custom types and not only fundamental types.