Way to recover the coordinates from the PLOTTING specification. Hopefully some f(k1,k2,k3, process_type) after the kinematics are transformed should suffice.
Otherwise, we would add kin_x, kin_q to the plotting specification for nondefault cases.
Just need to hook up the proper calls to scipy.integrate and logic to sum over flavors. Should be based on an already existing code.
From there, we can make the usual lumi plots, as well as the Karlsruhe plot.
Done. Needs to be compared to the SMPDF. Some minor improvements could be:
Need to decide how to group them We don’t want a 4000x4000 table. Maybe by default take percentile(abs(rho), 90) or similar.
Trivial to do the PDF distances. Do we want a better measure of distance such as Kolmogorov?
The UI for using them should look like:
replica_filters:
- bad_dataset:
threshold: 2
dataset_input:
dataset: CMSDY2D12
theoryid: 65
pdf: NNPDF30_nlo_as_0118
fit: myfit
use_cuts:True
actions_:
- - plot_pdfreplicas
- plot_training_validationand what this does is mark all the replicas that are above the threshold in all the plots with replicas.
Similar to filters, there should be an interface for colors, so that we can e.g. do the kinematic plot with a custom color scale.
Important missing feature? Decide after some more experience with reports produced with the existing features. Maybe more functionality to group figures.
To late to hook with C++?
Otherwise implement the same logic in Python, which would surely be faster.
Length and so on.
ϕ and so on.