Changeset to separate the cluster theory and models logic from the internal workings of Firecrown. As an added bonus, also updated the binned cluster number counts likelihood statistic to use ModelingTools in the expected firecrown workflow.
Cluster module
Added an AbundanceData class to wrap the sacc file and to be used by the likelihood
The sacc file data needs to be manipulated to extract the relevant cluster number counts (and masses) for the likelihood
Put this code within this wrapper class to keep the likelihood more clear from this sacc-specific logic
Removed firecrown infrastructure code and integration code from the ClusterAbundance class.
Added a new Kernel class to represent any ingredients in the cluster likelihood
Implemented kernels to represent the probability distribution P(z_spec|z_true), P(lambda|mass), etc
Created a MassRichnessGaussian of Kernel, anticipating having many of these. This could potentially fall into yagni territory (you aren't gonna need it).
Added a new Integrator class for the cluster module
Each integrator passes arguments to the cluster integrated differently, so we added this abstraction to allow specific integrators wrap and call the cluster abundance with their custom mapping
Implemented a NumCosmo and SciPy integrator as an example.
Cluster likelihood factory function
Modified the cluster likelihood to use dependency injection for all of the choices for our likelihood. Or in other words, moved all of the specific choices (integrator, cluster kernel ingredients, data vector) to the top of the call stack, so the user can swap in and out components to the likelihood.
Other firecrown changes
Added ClusterAbundance to ModelingTools.
As a part of this, I have ParamsMap passing through prepare.
This may need to change now that ModelingTools is an updatable...?
Added to setup.cfg instructions to tell mypy to ignore tests. Not sure if we want to keep this.
Tests
Upped test coverage of all cluster code to 100%
Ensured previous tests and new tests are all still passing
Total runtime of all tests is < .5 seconds. All tests except two take < 0.01 seconds. Those two tests are almost regression, maybe we want to flag them as such?
Summary
Changeset to separate the cluster theory and models logic from the internal workings of Firecrown. As an added bonus, also updated the binned cluster number counts likelihood statistic to use
ModelingToolsin the expectedfirecrownworkflow.Cluster module
AbundanceDataclass to wrap the sacc file and to be used by the likelihoodClusterAbundanceclass.Kernelclass to represent any ingredients in the cluster likelihoodMassRichnessGaussianofKernel, anticipating having many of these. This could potentially fall into yagni territory (you aren't gonna need it).Integratorclass for the cluster moduleCluster likelihood factory function
Modified the cluster likelihood to use dependency injection for all of the choices for our likelihood. Or in other words, moved all of the specific choices (integrator, cluster kernel ingredients, data vector) to the top of the call stack, so the user can swap in and out components to the likelihood.
Other firecrown changes
ClusterAbundancetoModelingTools.ParamsMappassing throughprepare.ModelingToolsis an updatable...?setup.cfginstructions to tell mypy to ignore tests. Not sure if we want to keep this.Tests