CLMM v2.0 desired features

[combet]

This is an issue to discuss what functionalities we'd like to see in CLMM v2.0

[combet]

Top of my head:

• Stack analysis (Issue/127/write cluster ensemble #457, WL weights for individual background galaxies (Issue/443/stacked_data) #460, Mock cluster ensemble #462)
• Einasto/Hernquist profiles for CCL backend (Einasto + Hernquist for CCL backend + free Einasto slope parameter for NC backend #461)
• 2 halo term (2 halo term for CCL and NC backends #459 )
• Include mass fitting in CLMM (Start a minimal fitting module. #456)
• Handle shear calibration and metadetect outputs
• Use qp to interact with photoz

[cavestruz]

Questions/Comments:

• What do folks feel the scope of the minimal fitting module should be? I remember there being previous discussions on how much the fitting should rely on the user. But, if this is framed as a wrapper function within which folks can swap in/out their favorite fitter with 1-2 defaults, I could see this being useful.
• [convener hat on] Can we define the CLMM v2.0 project, and can folks present this in the next CL General? :-)

[cavestruz]

Initial suggestions from CLMM meeting for fitter:

• Scipy optimizers (LGB....) specify box constraints instead of open ended paramaters, and set pseudo-priors (Angus)
• emcee (can do things in parallel; all we need to provide is likelihood) - need to be able to do things like ensure that chains converged, etc. Note, can be tricky if we want automation - will require user to have more control. (Michel)

[combet]

Other points raised during the CLMM discussion:

• Regarding MCMC, package like ChainConsumer is very useful to check properties of chains, convergence, etc. (Angus)
• Modeling: allow for triaxial/ellipsoidal density profiles (Calum)
• Covariance: we have covariance computed for a cluster ensemble, but might want to also include functionality for single cluster analysis (Constantin). (Also linked to that paper from Heidi)
• Boost estimation:
  • Parametric implementation ongoing but how do we actually define the values of those parameters?
  • Calum: this is actually tricky and can be very dependent on photoz algorithm used
  • Michel: but can probably be infer from N(z) of the cluster versus background
  • Angus: only if you assume you have the true N(z), which we don't.
  • ...[Céline: not sure I've had all this conversation right, please edit accordingly @m-aguena, @calumhrmurray]
• Interfacing with qp
  • qp will be the way photoz information is stored and pass around in DESC
  • ACTION ITEM: need to get in touch with qp folks and start adapting CLMM to handle photoz through qp

[combet]

Adding a new point to the list discussed above.

• Shear calibration: we need to be able to handle metadetect outputs

[marina-ricci]

The previous comments cover what I had in mind. I add the following :

• Would be nice to have methods to compute analytical covariances
• The fitting module, using MCMC, is a “must have” for v2.2. As a user story, I want to be able to infer a posterior on the mass of an ensemble of clusters (and individual clusters), given a data vector containing the profiles + covariance.
• Having a method to predict reduced shear profiles, given a cosmology, survey specifications (for the source sample) and cluster properties (redshift + observable). This would be the first brick for a forward modeling approach.

[combet]

Regarding the fitting module "must have", @vitenti just mentioned at the end of tag-up that the "connector" between FireCrown and CLMM is almost done, so we should be able to use FireCrown for this. Sandro, please correct me if I misunderstood.

[combet]

Oh and have CLMM pip-installable!