Add ASV benchmarks for 6 modules changed in v0.9.5 (#1156)

* Add ASV benchmarks for 6 uncovered modules (#1137)

normalize, diffusion, erosion, balanced_allocation, dasymetric,
and reproject all received memory-guard or lazy-reduction fixes
in the v0.9.5 cycle but had no ASV benchmark coverage.

Each new module follows the existing Benchmarking base-class pattern
with appropriate grid sizes and backend params.

* Add Rescale, Standardize, Diffusion, Dasymetric to CI benchmark filter (#1137)

These four are fast enough (<5ms at 300px) for CI. Erosion and Reproject
are excluded because they're too compute-heavy for PR checks.

* Fix code review findings in benchmark modules (#1137)

- dasymetric.py: add call parens to has_cuda_and_cupy() guard
- reproject.py: move import to module level to keep it out of timed methods

Author: brendancol

.github/workflows/benchmarks.yml

@@ -37,15 +37,15 @@ jobs:
         run: |
           asv continuous origin/master HEAD \
             --factor $ASV_FACTOR \
-            --bench "^(Slope|Proximity|Zonal|CostDistance|Focal)" \
+            --bench "^(Slope|Proximity|Zonal|CostDistance|Focal|Rescale|Standardize|Diffusion|Dasymetric)" \
             | tee bench_output.txt
 
       - name: Run benchmarks (push to master)
         if: github.event_name == 'push'
         working-directory: benchmarks
         run: |
           asv run HEAD^! \
-            --bench "^(Slope|Proximity|Zonal|CostDistance|Focal)" \
+            --bench "^(Slope|Proximity|Zonal|CostDistance|Focal|Rescale|Standardize|Diffusion|Dasymetric)" \
             | tee bench_output.txt
 
       - name: Check for regressions

benchmarks/benchmarks/balanced_allocation.py

@@ -0,0 +1,46 @@
+import numpy as np
+import xarray as xr
+
+from xrspatial.balanced_allocation import balanced_allocation
+
+from .common import get_xr_dataarray
+
+
+class BalancedAllocation:
+    # Memory-intensive: holds N_sources cost surfaces simultaneously.
+    # Keep grids small to avoid OOM during benchmarking.
+    params = ([100, 300], ["numpy", "dask"])
+    param_names = ("nx", "type")
+
+    def setup(self, nx, type):
+        ny = nx // 2
+        # Friction surface: positive values everywhere.
+        friction = get_xr_dataarray((ny, nx), type)
+        if type == "dask":
+            import dask.array as da
+            friction.data = da.fabs(friction.data) + 0.1
+        else:
+            friction.data = np.abs(friction.data) + 0.1
+        self.friction = friction
+
+        # Source raster: place 4 source points in corners.
+        sources = np.zeros((ny, nx), dtype=np.float32)
+        margin = max(1, nx // 10)
+        sources[margin, margin] = 1
+        sources[margin, nx - margin - 1] = 2
+        sources[ny - margin - 1, margin] = 3
+        sources[ny - margin - 1, nx - margin - 1] = 4
+
+        x = np.linspace(-180, 180, nx)
+        y = np.linspace(-90, 90, ny)
+
+        if type == "dask":
+            import dask.array as da
+            data = da.from_array(sources, chunks=(max(1, ny // 2), max(1, nx // 2)))
+        else:
+            data = sources
+
+        self.raster = xr.DataArray(data, coords=dict(y=y, x=x), dims=["y", "x"])
+
+    def time_balanced_allocation(self, nx, type):
+        balanced_allocation(self.raster, self.friction, max_iterations=10)

benchmarks/benchmarks/dasymetric.py

@@ -0,0 +1,60 @@
+import numpy as np
+import xarray as xr
+
+from xrspatial.dasymetric import disaggregate
+
+from .common import get_xr_dataarray
+
+
+class Dasymetric:
+    params = ([100, 300, 1000], ["numpy", "cupy", "dask"])
+    param_names = ("nx", "type")
+
+    def setup(self, nx, type):
+        ny = nx // 2
+
+        # Zones: 4 rectangular blocks.
+        zones_np = np.zeros((ny, nx), dtype=np.int32)
+        zones_np[: ny // 2, : nx // 2] = 1
+        zones_np[: ny // 2, nx // 2 :] = 2
+        zones_np[ny // 2 :, : nx // 2] = 3
+        zones_np[ny // 2 :, nx // 2 :] = 4
+
+        x = np.linspace(-180, 180, nx)
+        y = np.linspace(-90, 90, ny)
+
+        if type == "dask":
+            import dask.array as da
+            zdata = da.from_array(zones_np, chunks=(max(1, ny // 2), max(1, nx // 2)))
+        elif type == "cupy":
+            from xrspatial.utils import has_cuda_and_cupy
+            if not has_cuda_and_cupy():
+                raise NotImplementedError()
+            import cupy
+            zdata = cupy.asarray(zones_np)
+        else:
+            zdata = zones_np
+
+        self.zones = xr.DataArray(zdata, coords=dict(y=y, x=x), dims=["y", "x"])
+
+        # Values: one total per zone.
+        self.values = {1: 1000.0, 2: 2000.0, 3: 1500.0, 4: 2500.0}
+
+        # Weight surface: use the standard Gaussian bump.
+        weight = get_xr_dataarray((ny, nx), type)
+        # Make weights non-negative.
+        if type == "dask":
+            import dask.array as da
+            weight.data = da.fabs(weight.data) + 0.01
+        elif type == "cupy":
+            import cupy
+            weight.data = cupy.abs(weight.data) + 0.01
+        else:
+            weight.data = np.abs(weight.data) + 0.01
+        self.weight = weight
+
+    def time_disaggregate_weighted(self, nx, type):
+        disaggregate(self.zones, self.values, self.weight, method="weighted")
+
+    def time_disaggregate_binary(self, nx, type):
+        disaggregate(self.zones, self.values, self.weight, method="binary")

benchmarks/benchmarks/diffusion.py

@@ -0,0 +1,17 @@
+from xrspatial.diffusion import diffuse
+
+from .common import Benchmarking
+
+
+class Diffusion(Benchmarking):
+    params = ([100, 300, 1000], ["numpy", "cupy", "dask"])
+    param_names = ("nx", "type")
+
+    def __init__(self):
+        super().__init__(func=None)
+
+    def time_diffuse_1step(self, nx, type):
+        diffuse(self.xr, diffusivity=1.0, steps=1)
+
+    def time_diffuse_10steps(self, nx, type):
+        diffuse(self.xr, diffusivity=1.0, steps=10)

benchmarks/benchmarks/erosion.py

@@ -0,0 +1,21 @@
+from xrspatial.erosion import erode
+
+from .common import get_xr_dataarray
+
+
+class Erosion:
+    # Erosion is a global operation that materialises dask arrays.
+    # Keep grid sizes small and iteration counts low so benchmarks
+    # finish in reasonable time.
+    params = ([100, 300], ["numpy", "cupy", "dask"])
+    param_names = ("nx", "type")
+
+    def setup(self, nx, type):
+        ny = nx // 2
+        self.xr = get_xr_dataarray((ny, nx), type)
+
+    def time_erode_500(self, nx, type):
+        erode(self.xr, iterations=500, seed=42)
+
+    def time_erode_5000(self, nx, type):
+        erode(self.xr, iterations=5000, seed=42)

benchmarks/benchmarks/normalize.py

@@ -0,0 +1,19 @@
+from xrspatial.normalize import rescale, standardize
+
+from .common import Benchmarking
+
+
+class Rescale(Benchmarking):
+    def __init__(self):
+        super().__init__(func=rescale)
+
+    def time_rescale(self, nx, type):
+        return self.time(nx, type)
+
+
+class Standardize(Benchmarking):
+    def __init__(self):
+        super().__init__(func=standardize)
+
+    def time_standardize(self, nx, type):
+        return self.time(nx, type)

benchmarks/benchmarks/reproject.py

@@ -0,0 +1,27 @@
+from .common import get_xr_dataarray
+
+try:
+    from xrspatial.reproject import reproject as _reproject
+    _HAS_PYPROJ = True
+except ImportError:
+    _HAS_PYPROJ = False
+
+
+class Reproject:
+    params = ([100, 300, 1000], ["numpy", "dask"])
+    param_names = ("nx", "type")
+
+    def setup(self, nx, type):
+        if not _HAS_PYPROJ:
+            raise NotImplementedError("pyproj required")
+
+        ny = nx // 2
+        self.xr = get_xr_dataarray((ny, nx), type)
+        # Tag with WGS84 so reproject knows the source CRS.
+        self.xr.attrs["crs"] = "EPSG:4326"
+
+    def time_reproject_to_mercator(self, nx, type):
+        _reproject(self.xr, "EPSG:3857")
+
+    def time_reproject_to_utm(self, nx, type):
+        _reproject(self.xr, "EPSG:32610")