Backport PR #2771 on branch v3.10.x (Speed up histogram in plot options (alternative to #2763))

CHANGES.rst

@@ -4,6 +4,9 @@
 Bug Fixes
 ---------
 
+- Histogram in Plot Options now uses random sampling to better
+  represent the data without sacrificing performance. [#2771]
+
 Cubeviz
 ^^^^^^^
 

jdaviz/configs/default/plugins/plot_options/plot_options.py

@@ -4,9 +4,8 @@
 import matplotlib
 import numpy as np
 
-from astropy.visualization import (
-    ManualInterval, ContrastBiasStretch, PercentileInterval
-)
+from astropy.visualization import ManualInterval, ContrastBiasStretch
+
 from echo import delay_callback
 from traitlets import Any, Dict, Float, Bool, Int, List, Unicode, observe
 
@@ -36,6 +35,8 @@
 
 __all__ = ['PlotOptions']
 
+RANDOM_SUBSET_SIZE = 10_000
+
 
 def _register_random_cmap(
     cmap_name,
@@ -973,13 +974,8 @@
                 x_max = x_limits.max()
                 y_min = max(y_limits.min(), 0)
                 y_max = y_limits.max()
-                arr = comp.data[y_min:y_max, x_min:x_max]
-                if self.config == "imviz":
-                    # Downsample input data to about 400px (as per compass.vue) for performance.
-                    xstep = max(1, round(arr.shape[1] / 400))
-                    ystep = max(1, round(arr.shape[0] / 400))
-                    arr = arr[::ystep, ::xstep]
-                sub_data = arr.ravel()
+
+                sub_data = comp.data[y_min:y_max, x_min:x_max]
 
             else:
                 # spectrum-2d-viewer, for example.  We'll assume the viewer
@@ -996,28 +992,29 @@
                                 (y_data >= viewer.state.y_min) &
                                 (y_data <= viewer.state.y_max))
 
-                sub_data = comp.data[inds].ravel()
+                sub_data = comp.data[inds]
 
         else:
-            if self.config == "imviz":
-                # Downsample input data to about 400px (as per compass.vue) for performance.
-                xstep = max(1, round(data.shape[1] / 400))
-                ystep = max(1, round(data.shape[0] / 400))
-                arr = comp[::ystep, ::xstep]
-            else:
-                # include all data, regardless of zoom limits
-                arr = comp.data
-            sub_data = arr.ravel()
-
-        # filter out nans (or else bqplot will fail)
-        if np.any(np.isnan(sub_data)):
-            sub_data = sub_data[~np.isnan(sub_data)]
+            # include all data, regardless of zoom limits
+            sub_data = comp.data
 
+        self.stretch_histogram.viewer.state.random_subset = RANDOM_SUBSET_SIZE
         self.stretch_histogram._update_data('histogram', x=sub_data)
 
         if len(sub_data) > 0:
-            interval = PercentileInterval(95)
-            hist_lims = interval.get_limits(sub_data)
+
+            # Use glue to compute the statistics since this allows us to use
+            # a random subset of the data to compute the histogram.
+            # The 2.5 and 97.5 hardcoded here is equivalent to
+            # PercentileInterval(95).get_limits(sub_data)
+            glue_data = self.stretch_histogram.app.data_collection['histogram']
+            hist_lims = (
+                glue_data.compute_statistic('percentile', glue_data.id['x'],
+                                            percentile=2.5, random_subset=RANDOM_SUBSET_SIZE),
+                glue_data.compute_statistic('percentile', glue_data.id['x'],
+                                            percentile=97.5, random_subset=RANDOM_SUBSET_SIZE)
+            )
+
             # set the stepsize for vmin/vmax to be approximately 1% of the range of the
             # histogram (within the percentile interval), rounded to 1-2 significant digits
             # to avoid random step sizes.  This logic is somewhat arbitrary and can be safely

jdaviz/core/template_mixin.py

@@ -4631,13 +4631,13 @@ def _update_data(self, label, reset_lims=False, **kwargs):
         data = self.app.data_collection[label]
 
         # if not provided, fallback on existing data
-        length_mismatch = False
+        shape_mismatch = False
         for component in self._viewer_components:
             kwargs.setdefault(component, data[component])
-            if len(kwargs[component]) != len(data[component]):
-                length_mismatch = True
+            if np.asarray(kwargs[component]).shape != data[component].shape:
+                shape_mismatch = True
 
-        if not length_mismatch:
+        if not shape_mismatch:
             # then we can update the existing entry
             components = {c.label: c for c in data.components}
             data.update_components({components[comp]: kwargs[comp]

jdaviz/core/tests/test_tools.py

@@ -74,6 +74,11 @@ def test_stretch_bounds(imviz_helper):
 
 
 def test_stretch_bounds_and_spline(imviz_helper):
+
+    # As the histogram randomly samples the array, we should make sure the
+    # values used here are reproducible
+    np.random.seed(42)
+
     image_1 = NDData(make_4gaussians_image(), unit=u.nJy)
     imviz_helper.load_data(image_1)
     po = imviz_helper.plugins["Plot Options"]
@@ -93,7 +98,7 @@ def test_stretch_bounds_and_spline(imviz_helper):
 
         knots_after_drag_move = (
             [0.0, 0.1, 0.21712585033417825, 0.7, 1.0],
-            [0.0, 0.05, 0.2900993441358025, 0.9, 1.0],
+            [0.0, 0.05, 0.2852214046563617, 0.9, 1.0],
         )
 
         stretch_tool.on_mouse_event(knot_move_msg)

pyproject.toml

@@ -12,8 +12,8 @@ dependencies = [
     "traitlets>=5.0.5",
     "bqplot>=0.12.37",
     "bqplot-image-gl>=1.4.11",
-    "glue-core>=1.18.0",
-    "glue-jupyter>=0.20",
+    "glue-core>=1.20.0",
+    "glue-jupyter>=0.21.0",
     "echo>=0.5.0",
     "ipykernel>=6.19.4",
     "ipyvue>=1.6",