Things to save from protopipe before the full-refactoring #1744
Closed
6 of 7 tasks
Labels
help wanted
new functionality
refactoring
summary issue
An issue which collects information about multiple issues and/or PR
Projects
Milestone
We are near to a complete refactoring of the pipeline based on ctapipe, but if we did it now some things would disappear and that is not good (both from the point-of-view of comparisons with historical pipelines and for any future study).
These features can and should be generalized to account for more use-cases than the ones taken into account currently by protopipe.
This issue is a summary list: I will add new things as I stumble upon them and each related PR should be linked to the respective feature.
This is the equivalent of any other
QualityQuery
-based class we use (inImageProcessor
andShowerProcessor
) but for image extractors.It is needed by extractors like
TwoPassWindowSum
(CTAMARS-based analysis), for which it is fundamental to label some events (in this case images, not showers) which do not survive a step during the total image extraction process (in this case a preliminary image cleaning cut in between the two passes).This will require to modify the
ImageExtractor
class to output a quality query together with charges and peak times.Of course, for single-pass image extractors, this condition will should be always positive by default and thus it would have no visible effect.
In the specific case of protopipe I have implemented the LST-stereo trigger condition, required for Prod3b simulations in which this requirement was not simulated
This should be a simple scale factor for the reconstructed charge extracted from each pixel.
In particular, as reported here,
This allows the pipeline to read from DL2 files this quantity directly in the right frame were telescope-wise impact distances should be computed
It has not been used in a long time and it should be added as an
ImageCleaner
, perhaps from a module itself.This is something that should be added after PR #1726.
There should a function that takes as input
what in protopipe I call
TRAINING
data (aka DL1b + shower geometry)information from the configuration of the model building process, in particular which features have been used
the possibility to eval at runtime features that do not come directly from the
TRAINING
file, but that are functions of those (e.g.np.log10(hillas_intensity)
)some configuration options to compute the final estimated quantity (weighted average and if yes with what weight, e.g. standard deviation of the predictions, some parametrized function of TRAINING variables, etc...)
Shower Impact Parameter(s)
Related issues:
Related Pull Requests:
#1745 will allow to compute it directly in the right frame, but for now it is something that is not done by ctapipe's
ShowerProcessor
.From my side, I think wrapping the first step of the pipeline around
ctapipe-process
just to open the files - calculate impact_dist - close file - use ctapipe ML module (or protopipe.mva) is a bit uncomfortable...This is of course because we need to have impact parameter in our files for benchmarking, otherwise it would be sufficient to compute it directly in the ML module.
Alternatively, the benchmarking code would compute it, but it seems a bit ugly to me.
Should we do allow this operation from ctapipe-process? in the end we already have 1 DL2 telescope-wise variable written to file (core_psi)....
The text was updated successfully, but these errors were encountered: