Merge pull request #6 from sjoshistrats/feature/benchmarks

Add benchmarks

.circleci/config.yml

@@ -13,6 +13,7 @@ jobs:
       - run:
           command: |  # Run tests
             pipenv install pytest
+            pipenv install pytest-benchmark
             pipenv run pytest python/fastgrouper/test/ -s -vv
 workflows:
   build_test:

python/fastgrouper/test/test_arr.py → python/fastgrouper/test/test_grouped.py

@@ -1,3 +1,4 @@
+import pytest
 import numpy as np
 from   numpy.testing import assert_almost_equal
 
@@ -14,37 +15,37 @@ def foobar_op(x, y):
     return np.mean(x + y)
 
 def test_arr_grouped():
-    
+
     # Prepare example
     arr_grpr = fastgrouper.arr.Grouped(GIDS)
-    
+
     # Check apply with positional args
     result = arr_grpr.apply(foobar_op, XVALS, YVALS)
-    
+
     # Ensure returned result is a numpy array
     assert isinstance(result, np.ndarray)
-    
+
     # Check values
     assert_almost_equal(result, np.array(EXPECTED_APPLY))
-    
+
     # Check values when using apply with keyword args
     result = arr_grpr.apply(foobar_op, XVALS, y=YVALS)
     assert_almost_equal(result, np.array(EXPECTED_APPLY))
-    
+
     # Check values when using apply_expand
     result = arr_grpr.apply_expand(foobar_op, XVALS, y=YVALS)
     assert_almost_equal(result, np.array(EXPECTED_APPLY_EXPAND))
-    
+
 def test_li_grouped():
-    
+
     # Prepare example
     li_grpr = fastgrouper.li.Grouped(GIDS)
-    
+
     # Check apply with keyword args
     result = li_grpr.apply(foobar_op, XVALS, y=YVALS)
-    
+
     # Ensure returned result is a list
     assert isinstance(result, list)
-    
+
     # Check values
-    assert_almost_equal(result, EXPECTED_APPLY)
+    assert_almost_equal(result, EXPECTED_APPLY)

python/fastgrouper/test/test_grouped_benchmark.py

@@ -0,0 +1,66 @@
+import pytest
+import pandas as pd
+import numpy as np
+from   numpy.testing import assert_almost_equal
+
+import fastgrouper.arr
+import fastgrouper.li
+
+# Create assymetrical groups
+XVALS = np.linspace(500, 1200, 50000)
+YVALS = np.linspace(-230, 177.3, 50000)
+GIDS  = np.tile(np.arange(500), 100)
+GIDS[[14, 19, 230, 87]] = 4
+GIDS[[345, 1270, 63, 1287]] = 12
+
+def beepbop(x, y):
+    return np.min(np.abs(np.sin(x) + np.sin(y)))
+
+def test_fastgrouper_arr_slice_apply_benchmark(benchmark):
+    arr_grpr = fastgrouper.arr.Grouped(GIDS)
+    idx = pd.Index(arr_grpr.dedup_gids, name="gids")
+
+    def apply_fn():
+        result = arr_grpr.apply(beepbop, XVALS, YVALS)
+
+        # Sorting to make a more fair comparison against pure pandas benchmark
+        return pd.Series(result, index=idx).sort_index()
+
+    benchmark(apply_fn)
+
+def test_pure_pandas_slice_apply_benchmark(benchmark):
+    df = pd.DataFrame({
+        "gids": GIDS,
+        "xvals": XVALS,
+        "yvals": YVALS
+    })
+    pdgrpd = df.groupby("gids")
+
+    def apply_fn(r):
+        return beepbop(r["xvals"].values, r["yvals"].values)
+
+    benchmark(pdgrpd.apply, apply_fn)
+
+def test_fastgrouper_all_steps_benchmark(benchmark):
+    def apply_fn():
+        arr_grpr = fastgrouper.arr.Grouped(GIDS)
+        idx = pd.Index(arr_grpr.dedup_gids, name="gids")
+        result = arr_grpr.apply(beepbop, XVALS, YVALS)
+
+        # Sorting to make a more fair comparison against pure pandas benchmark
+        return pd.Series(result, index=idx).sort_index()
+
+    benchmark(apply_fn)
+
+def test_pure_pandas_all_steps_benchmark(benchmark):
+    df = pd.DataFrame({
+        "gids": GIDS,
+        "xvals": XVALS,
+        "yvals": YVALS
+    })
+    pdgrpd = df.groupby("gids")
+
+    def apply_fn():
+        return df.groupby("gids").apply(lambda r: beepbop(r["xvals"].values, r["yvals"].values))
+
+    benchmark(apply_fn)

setup.py

@@ -9,7 +9,7 @@
 
 setuptools.setup(
     name="fastgrouper",
-    version="0.1.1",
+    version="0.1.2",
     author="Shreyas Joshi",
     author_email="[email protected]",
     description="A package for applying efficient groupby operations.",
@@ -26,5 +26,8 @@
     package_dir={"": "python"},
     packages=setuptools.find_packages(where="python"),
     python_requires=">=3.6",
-    install_requires=["numpy", "pandas"]
+    install_requires=["numpy", "pandas"],
+    extras_require={
+        "testing": ["pytest", "pytest-benchmark"]
+    }
 )