Connect the RAIL creation and estimation modules and evaluate p(z) metrics on the simplest end-to-end test thereof
Contacts: @aimalz @jbkalmbach @jfcrenshaw @johannct @sschmidt23
Day/Time: TBD, probably the later time slot on Monday, Tuesday, Thursday, Friday
Main communication channel: #desc-pz-rail on Slack
GitHub repo: https://github.com/LSSTDESC/RAIL
Goals and deliverable
[ ] baseline creation module -- finalize the open PR (likely done before Sprint Week)
[X] baseline estimation module -- even includes a few familiar estimators
[ ] baseline evaluation module -- gather the PZ DC1 metrics scripts to run on the outputs of p(z) estimators, maybe draft class structure
[ ] write the script connecting the three RAIL modules
Resources and skills needed
Python aficionados: This is a good time to jump in on pioneering the class structure for the evaluation module!
Proponents of p(z) estimators: We want to integrate your p(z) estimator into the RAIL framework.
Collectors of training sets and/or template libraries for p(z) estimation: We want to model the p(z, photometry) space realistically from your data.
Detailed description
RAIL is the back-end code for running experiments on photo-z estimators under increasingly realistically imperfect conditions. The baseline version, project MonoRAIL, has three main modules, described in the repo's README. They've been evolving slowly toward an end-to-end pipeline, making this sprint the golden spike! After this, the code will be ready for folks to develop in different directions simultaneously to accomplish a variety of PZ and XWG goals.
Golden Spike: baseline RAIL end-to-end pipeline
Connect the RAIL creation and estimation modules and evaluate p(z) metrics on the simplest end-to-end test thereof
Contacts: @aimalz @jbkalmbach @jfcrenshaw @johannct @sschmidt23 Day/Time: TBD, probably the later time slot on Monday, Tuesday, Thursday, Friday Main communication channel:
#desc-pz-railon Slack GitHub repo: https://github.com/LSSTDESC/RAILGoals and deliverable
creationmodule -- finalize the open PR (likely done before Sprint Week)estimationmodule -- even includes a few familiar estimatorsevaluationmodule -- gather the PZ DC1 metrics scripts to run on the outputs of p(z) estimators, maybe draft class structureResources and skills needed
evaluationmodule!Detailed description
RAIL is the back-end code for running experiments on photo-z estimators under increasingly realistically imperfect conditions. The baseline version, project MonoRAIL, has three main modules, described in the repo's README. They've been evolving slowly toward an end-to-end pipeline, making this sprint the golden spike! After this, the code will be ready for folks to develop in different directions simultaneously to accomplish a variety of PZ and XWG goals.