Add logging when data range is large enough to trigger slow path.

python/lsst/summit/utils/utils.py

@@ -932,7 +932,7 @@ def getFilterSeeingCorrection(filterName):
             raise ValueError(f"Unknown filter name: {filterName}")
 
 
-def getCdf(data, scale, nBinsMax=131072):
+def getCdf(data, scale, nBinsMax=300_000):
     """Return an approximate cumulative distribution function scaled to
     the [0, scale] range.
 
@@ -1008,8 +1008,10 @@ def getQuantiles(data, nColors):
         A monotonically increasing sequence of size (nColors + 1). These are
         the edges of nColors intervals.
     """
-    if (np.nanmax(data) - np.nanmin(data)) > 131072:
+    if (np.nanmax(data) - np.nanmin(data)) > 300_000:
         # Use slower but memory efficient nanquantile
+        logger = logging.getLogger(__name__)
+        logger.warning("Data range is very large; using slower quantile code.")
         boundaries = np.nanquantile(data, np.linspace(0, 1, nColors + 1))
     else:
         cdf, minVal, maxVal = getCdf(data, nColors)

tests/test_utils.py

@@ -238,7 +238,13 @@ def test_quantiles(self):
         for nColors, (mean, width, decimal) in itertools.product(colorRanges, dataRanges):
             data = np.random.normal(mean, width, (100, 100))
             data[10, 10] = np.nan  # check we're still nan-safe
-            edges1 = getQuantiles(data, nColors)
+            if np.nanmax(data) - np.nanmin(data) > 300_000:
+                with self.assertLogs(level="WARNING") as cm:
+                    edges1 = getQuantiles(data, nColors)
+                self.assertIn("Data range is very large", cm.output[0])
+            else:
+                with self.assertNoLogs(level="WARNING") as cm:
+                    edges1 = getQuantiles(data, nColors)
             edges2 = np.nanquantile(data, np.linspace(0, 1, nColors + 1))  # must check with nanquantile
             np.testing.assert_almost_equal(edges1, edges2, decimal=decimal)