Add LogSAS binning for 3pt correlations

.github/workflows/ci.yml

@@ -55,18 +55,18 @@ jobs:
                       CXX: g++
 
         steps:
-            - uses: actions/checkout@v2
+            - uses: actions/checkout@v4
               with:
                 # Helpful for a reliable codecov upload.
                 fetch-depth: 0
 
             - name: Set up Python ${{ matrix.py }}
-              uses: actions/setup-python@v2
+              uses: actions/setup-python@v4
               with:
                 python-version: ${{ matrix.py }}
 
             - name: Cache pip
-              uses: actions/cache@v2
+              uses: actions/cache@v3
               with:
                 path: ~/.cache/pip
                 key: ${{ runner.os }}-${{ matrix.py }}-pip-${{ hashFiles('requirements.txt') }}

CHANGELOG.rst

@@ -24,6 +24,10 @@ API Changes
 - Renamed the base classes BinnedCorr2 -> `Corr2` and BinnedCorr3 -> `Corr3`.  These are not
   normally used directly by users, so it shouldn't be noticeable in user code. (#155)
 - Removed all deprecations from the 4.x series. (#156)
+- Removed support for reading back in output files from the 3.x series. (#165)
+- Removed the 3pt CrossCorrelation classes.  See the new ``ordered=True`` option to the 3pt
+  ``process`` method instead to get correlations where the order of the three catalogs is fixed.
+  This is simpler and more intuitive, and for many use cases it is more efficient. (#165)
 
 
 Performance improvements
@@ -46,6 +50,9 @@ New features
 - Added spin-3 and spin-4 correlations using the letters T (for Trefoil) and Q (for Quatrefoil)
   respectively, including `NTCorrelation`, `KTCorrelation`, `TTCorrelation`, `NQCorrelation`,
   `KQCorrelation` and `QQCorrelation`. (#160)
+- Added ``ordered=True`` option to the 3pt ``process`` methods for keeping the order of the
+  catalogs fixed in the triangle orientation. (#165)
+- Added ``bin_type='LogSAS'`` for 3pt correlations. (#165)
 
 
 Bug fixes

devel/.gitignore

@@ -3,3 +3,4 @@
 junk*
 sva1_gold*.fits
 des_sv.fits
+data

docs/metric.rst

@@ -44,6 +44,12 @@ will be in terms of great circle distances.  However, they will not necessarily
 precisely uniformly in log(r), since the original bin spacing will have been set up in terms
 of the chord distances.
 
+Similarly, for three-point correlation functions with ``bin_type="LogSAS"``, the phi values
+will have been accumulated according to the regular Euclidean triangles made from the three
+chord distances.  The correction to true spherical-geometric angles between the great circles
+happens at the end, so the ``meanphi`` values are approximately correct, but the binning
+will have been done using the angles between the chords.
+
 "Arc"
 -----
 
@@ -62,6 +68,8 @@ chord calculations are in any way problematic for your particular use case.
 
 Also, unlike the "Euclidean" version, the bin spacing will be uniform in log(r) using the
 actual great circle distances, rather than being based on the chord distances.
+And for three-point correlation functions with ``bin_type="LogSAS"``, the phi values
+will be the true spherical-geometric angles between the great circles.
 
 
 .. _Rperp: