cosmo_val + Snakemake migration to the SACC writers (PRD #241 row 4)

README.md

@@ -87,6 +87,25 @@ docker run --rm -it ghcr.io/cosmostat/sp_validation:develop python -c "import sp
 
 We do not currently build images for Apple Silicon/arm64; however the amd64 images should work on these systems, albeit with reduced performance.
 
+## Local Installation
+
+Requires Python ≥ 3.12. With [uv](https://docs.astral.sh/uv/):
+
+```bash
+uv venv --python 3.12
+uv pip install -e '.[test]'
+```
+
+To also install the data-vector blinding stack (Smokescreen + firecrown, PRD
+[#241](https://github.com/CosmoStat/sp_validation/issues/241)), pass the
+dependency-override file — firecrown is not pip-resolvable without it (see
+`uv-overrides.txt` for why):
+
+```bash
+uv pip install --overrides uv-overrides.txt -e '.[test,blinding]'
+python scripts/patch_firecrown.py  # make pip-installed firecrown importable without NumCosmo
+```
+
 
 
 ## Flow chart

papers/bmodes/rules/claims.smk

@@ -86,10 +86,10 @@ def _xi_reporting_path(version):
 
 
 def _xi_integration_path(version):
-    """Path to fine-binned 2PCF integration file."""
+    """Path to fine-binned 2PCF integration file. Unpatched: values only, no covariance."""
     return (
         f"{COSMO_VAL_OUTPUT}/{version}_xi_minsep={FIDUCIAL['min_sep_int']}"
-        f"_maxsep={FIDUCIAL['max_sep_int']}_nbins={FIDUCIAL['nbins_int']}_npatch={FIDUCIAL['npatch']}.txt"
+        f"_maxsep={FIDUCIAL['max_sep_int']}_nbins={FIDUCIAL['nbins_int']}_npatch=1.txt"
     )
 
 
@@ -355,25 +355,21 @@ rule pure_eb_covariance:
 
 
 rule calculate_pure_eb_ptes:
-    """Calculate PTE matrices for Pure E/B mode scale cut robustness.
+    """PTE matrices for pure E/B-mode scale-cut robustness.
 
-    Per-blind: Uses blind-specific integration covariance for PTE calculation.
-    The pure_eb_data vectors are identical across blinds; only covariance differs.
-
-    In practice, BB covariance is blind-independent (validated by
-    bb_covariance_blind_independence), so downstream consumers (config_space_pte_matrices)
-    only request blind A. The per-blind wildcard is retained for the blind independence test.
+    Nothing here varies with the blind: the data vectors come from the blind-A
+    gather and the PTEs are Hartlap-debiased by the MC draw count, not by a
+    per-blind covariance. The wildcard survives as the filename slot the
+    consumer (config_space_pte_matrices) reads, and only blind A is ever built.
     """
     input:
         pure_eb_data="results/paper_plots/intermediate/{version}_A_pure_eb_semianalytic.npz",
-        cov_integration=lambda w: _cov_integration_path(w.version, w.blind),
     output:
         "results/paper_plots/intermediate/{version}_{blind}_pure_eb_ptes.npz",
     wildcard_constraints:
         blind=r"[ABC]",
     params:
         version="{version}",
-        npatch=FIDUCIAL["npatch"],
         n_samples=config["covariance"]["n_samples"],
     resources:
         mem_mb=16000,

papers/bmodes/scripts/calculate_pure_eb_ptes.py

@@ -4,14 +4,13 @@
 pure-E/B ``semianalytic.npz`` (data vectors + MC covariance), evaluates the
 ξ_+^B / ξ_-^B / joint ξ_tot^B χ² PTE matrices over the scale-cut grid via
 ``sp_validation.b_modes.calculate_eb_statistics`` (Hartlap-corrected inverse
-MC covariance), and writes the PTE matrices to
+MC covariance, debiased by the draw count), and writes the PTE matrices to
 ``{out}/{version}_{blind}_pure_eb_ptes.npz``.
 
     python calculate_pure_eb_ptes.py \
         --version SP_v1.4.6.3_leak_corr --blind A \
         --pure-eb-data <..._pure_eb_semianalytic.npz> \
-        --cov-integration <cov ..._processed.txt> \
-        --npatch 1 --n-samples 2000 --out <output_dir>
+        --n-samples 2000 --out <output_dir>
 """
 
 import argparse
@@ -22,31 +21,21 @@
 from sp_validation.b_modes import calculate_eb_statistics
 
 
-class FakeGG:
-    """Minimal GGCorrelation-like object for calculate_eb_statistics."""
-
-    def __init__(self, nbins, npatch):
-        self.nbins = nbins
-        self.npatch1 = npatch
-        self.npatch2 = npatch
-
-
 def calculate_ptes(
     version,
     blind,
     pure_eb_data,
-    cov_integration,
-    npatch,
     n_samples,
     output_dir,
 ):
     dataset = np.load(pure_eb_data)
 
     theta = dataset["theta"]
-    nbins = len(theta)
 
     results = {
-        "gg": FakeGG(nbins, int(npatch)),
+        "theta": theta,
+        # The MC draws are the realisations behind this covariance.
+        "n_eff": int(n_samples),
         "xip_E": dataset["xip_E"],
         "xim_E": dataset["xim_E"],
         "xip_B": dataset["xip_B"],
@@ -57,11 +46,7 @@ def calculate_ptes(
     }
 
     print(f"Calculating PTE matrices for {version}...")
-    results = calculate_eb_statistics(
-        results,
-        cov_path_int=cov_integration,
-        n_samples=int(n_samples),
-    )
+    results = calculate_eb_statistics(results)
 
     pte_matrices = results["pte_matrices"]
     output_data = {
@@ -83,21 +68,13 @@ def _from_cli(argv=None):
     ap.add_argument("--version", required=True)
     ap.add_argument("--blind", default="A")
     ap.add_argument("--pure-eb-data", required=True, help="Gathered semianalytic .npz")
-    ap.add_argument(
-        "--cov-integration",
-        default=None,
-        help="Integration-grid covariance _processed.txt (optional)",
-    )
-    ap.add_argument("--npatch", type=int, default=1)
     ap.add_argument("--n-samples", type=int, default=2000)
     ap.add_argument("--out", required=True, help="Output directory (lc {output})")
     a = ap.parse_args(argv)
     calculate_ptes(
         version=a.version,
         blind=a.blind,
         pure_eb_data=a.pure_eb_data,
-        cov_integration=a.cov_integration,
-        npatch=a.npatch,
         n_samples=a.n_samples,
         output_dir=a.out,
     )

papers/bmodes/scripts/run_xi_sweep.py

@@ -2,8 +2,8 @@
 
 Loops the [non-fiducial version list](sweep_versions.nonfiducial_versions) and
 runs the same ``run_2pcf.run_2pcf`` compute the fiducial two_point recipes call,
-once per version, writing every version's ξ± text dump (+ ξ+/ξ- FITS) into one
-lc ``{output}`` dir under run_2pcf's native, already-canonical name
+once per version, writing every version's ξ± text dump into one lc ``{output}``
+dir under run_2pcf's native, already-canonical name
 ``{ver}_xi_minsep={min}_maxsep={max}_nbins={nbins}_npatch={npatch}.txt`` — the
 exact pattern ``cosebis_version_comparison._xi_integration`` reconstructs.
 
@@ -76,7 +76,7 @@ def _from_cli(argv=None):
                 ver=ver,
                 cat_config=a.cat_config,
                 output_dir=a.out,
-                save_fits=True,
+                grid=grid,
                 **GRIDS[grid],
             )
 

papers/cosmo_val/config/config.yaml

@@ -58,11 +58,11 @@ cosmo_val:
         kmax: 20
         kmax_extrapolate: 500
 
-  # Pure E/B-mode decomposition (config space)
-  pure_eb:
-    min_sep_int: 0.08
-    max_sep_int: 300
-    nbins_int: 1000
+  # The fine ξ± grid the B-mode integrals run over.
+  integration:
+    min_sep: 0.08
+    max_sep: 300
+    nbins: 1000
 
   # COSEBIs decomposition (config space, fine integration binning)
   cosebis:

pyproject.toml

@@ -79,7 +79,8 @@ dependencies = [
     "pymaster",
     "regions",
     "reproject",
-    "sacc>=0.12",
+    # Floor matches the resolved lock; needs sacc's 2.x rewrite (BlockDiagonalCovariance).
+    "sacc>=2.4,<3",
     # scipy 1.18 ported FITPACK from Fortran to C, changing the return shape of
     # RectBivariateSpline(scalar, scalar, grid=False) from 0-d `array(x)` to
     # shape-(1,) `array([x])`. camb's BBN Y_He predictor (bbn.py) wraps the
@@ -152,8 +153,8 @@ glass = [
 # Snakemake workflow and cross-validation runners are available.
 workflow = [
     "snakemake",
-    # run_2pcf_highres.py drives the MPI convergence run; the container ships
-    # OpenMPI (/opt/ompi) so mpi4py builds against it.
+    # Optional MPI runners (the container ships OpenMPI at /opt/ompi, so mpi4py
+    # builds against it).
     "mpi4py",
     # NOTE: workflow/scripts/cv_*.py also import `cv_runner`, which is not
     # published or resolvable (no public repo found) — left undeclared pending