Presets + Cross-Covariance module + ISO rework notebooks

[ggalloni]

Summary

Adds a presets layer for ready-to-run SOLikeT configurations, a cross-covariance module to build joint covariances between probes, and an end-to-end ISO simulation pipeline (datasets → cross-covariance → joint analysis), plus reworked tutorials. Builds on the labelled-CrossCov store on master: cross-covariance blocks are aligned to the data by bandpower identity, so they stay correct regardless of build order.

soliket.presets — ready-to-run configurations

• quickstart(preset) → a Session wrapping a Cobaya model at the fiducial point; build_info(preset) for the raw info dict.
• resolve_aliases(model) exposes components by role (.mflike, .lensing, .foreground, .cosmo), matched by class — no fragile model.components[i] indexing.
• Parameters split into per-group files (cosmo / foreground / systematics) plus theory.yaml, loaded via load_params / load_fiducial_map / build_params. A defaults_dir override lets a folder replace any group file (per-file fallback to the package defaults).
• Templates for mflike, lensing, multigaussian. Clear errors when a group file is missing or not a mapping; foreground-drift tripwire and a cobaya mutate-safety guard.

soliket.cross_covariance — joint covariances

• Kernels: cmb_lensing_block, lensing_induced_block, shear_kappa_block, n1_crosscov_block; CAMB lensed-Cl derivatives (camb_lensing_derivatives); builders cmb_lensing_crosscov, lensing_induced_cov, shear_kappa_crosscov.
• CrossCov.from_cmb_lensing(mflike, lensing) computes the CMB×lensing block and labels each row/column with its bandpower identity (CMB rows in mflike's (pol, hasYX_xsp, (t1,t2), leff) vocabulary, kappa columns from the lensing ids). Alignment then happens by identity via CrossCov.to_canonical at assembly — build order is no longer load-bearing.

GaussianLikelihood / data

• bandpower_ids adapter so external likelihoods (mflike) carry per-bandpower identities; _get_gauss_data populates ids/indices.
• Unbinned-theory accessor and memoised bandpower binning.

sacc_tools — smooth datasets

• Builders for smooth (noise-free) MFLike and lensing twins (smooth_mflike_sacc, smooth-twin SACC), top-hat windows, Gaussian covariance.

CCL tracers

• Fix NLA intrinsic-alignment bias normalization; guard non-buildable tracers.

Notebooks

• ISO_sims pipeline (one shared ISO fiducial, SO normal-hierarchy neutrinos via a theory.yaml override): create_datasets → create_cross_covariance → analyse_datasets. Generated sims/ products are git-ignored (reproducible).
• analyse_datasets also writes a committable examples/ISO_sims.yaml — a cobaya-run twin of the Python run with <REPO>/<PACKAGES> placeholders for the machine-specific roots.
• Tutorials: quickstart.ipynb (presets on-ramp) and first_step_tutorial.ipynb (hand-built deep-dive) reworked to use resolve_aliases, source fiducials from the preset (load_fiducial_map), and run the sampler from Python (YAML kept as a serialization demo, cobaya-run noted for clusters).

[ggalloni]

@itrharrison @HTJense @raphkou @MatForconi Hi guys, let me know what you think of this. In principle, this should be the first step towards the reorganization we discussed at the whiteboard. In particular, have a look at the new notebooks, the preset containing defaults, and new tools for different levels of interaction with the repo.

[raphkou]

Thanks @ggalloni for this amazing work! I tested all the notebooks in ISO_sims and they work perfectly! I just had a few questions/comments:

• In create_datasets.ipynb, why not using the same yaml files to create the galaxy x kappa data? I guess we could use the same cosmology yaml file to ensure that we use the same cosmological parameters?
• In ISO_sims/defaults, I guess we will also add the other yaml files regarding foreground, accuracy, mcmc parameters, etc.? I can help with that if needed.
• I also imagine that there is some more cleaning to do in other subfolders of the examples folder or in notebooks/dev, as several files may not be useful anymore.

Many thanks again for this huge piece of work, it's looking very great! I can also handle the grid runs notebook/scripts when this branch becomes more stable or is merged.

[ggalloni]

Hi @raphkou, thanks for the feedback! I should have fixed your first point by converting the cosmo-point into the equivalent settings for CCL.

About the second, yes, absolutely! In this first example, I wanted to show that if only a part of the default files is provided in a folder like that, the rest of the options will be inherited from the package-level defaults in the presets folder.

As for the last comment, yes, for now, I kept them there as a backup. I wanted to wait for your approval before removing the old ones, as maybe there is something I missed, etc.