[Rivet-announce] Professor 2.2 released

[Holger Schulz]

Dear tuning and parametrisation enthusiasts,

we are very pleased to announce the release of Professor 2.2.


This release brings a much more efficient calculation of
parametrisations allowing to
parametrise 20D parameter spaces in seconds, a task that would have
taken hours
with the previous version. Unfortunately, parametrisations calculated
with previous
versions of professor are no longer supported. There was no way to
ensure backwards
compatibility. We feel that in this case the improvement in speed
significantly outweighs
keeping backward compatibility.

Further, for this release a focus has been improved robustness of our
scripts and much more meaningful
error messages prompted by feedback we got. This concerns for example
the cases of zero
error in data bins. We added a new CL switch for prof2-tune --filter
which automatically
removes such bins from the tuning.

Another addition is prof2-sobol in contrib/ which is another
pseudo-random generator that
gives less clustering than random.uniform in prof2-sample. It works the
same way as prof2-sample
except that it lacks the veto functionality at the moment.

We improved the machine learning algorithms by tuning the support vector
machine parameters
to give better resolutions. We are in contact with an author of
scikit-learn to understand the
resolution pattern observed better.

This release is available as source code and as docker image. We also
provide jupyter
notebooks now for our python api that can be test-driven on binder
(thanks to docker).
There are a few notebooks showing how to generally construct
parametrisations
i.e. not limited to rivet output. We also created a prototype for
interactive explorer
that is meant as replacement for the ageing prof2-I as notebook.


We would like to thank for the feedback we got from our users and hope
that you
enjoy working with this new version of professor.

Thanks,
Andy and Holger