Setup benchmark suite

Signed-off-by: Tim Paine <[email protected]>

.gitignore

@@ -104,6 +104,11 @@ csp/lib/
 *.so
 *.tsbuildinfo
 
+# Benchmarks
+.asv
+ci/benchmarks/*
+!ci/benchmarks/benchmarks.json
+
 # Jupyter / Editors
 .ipynb_checkpoints
 .autoversion

Makefile

@@ -118,6 +118,28 @@ dockerps:  ## spin up docker compose services for adapter testing
 dockerdown:  ## spin up docker compose services for adapter testing
 	$(DOCKER) compose -f ci/$(ADAPTER)/docker-compose.yml down
 
+##############
+# BENCHMARKS #
+##############
+.PHONY: benchmark benchmarks benchmark-regen benchmark-view benchmarks-regen benchmarks-view
+benchmark:  ## run benchmarks
+	python -m asv run --config csp/benchmarks/asv.conf.jsonc --verbose `git rev-parse --abbrev-ref HEAD`^!
+
+# https://github.com/airspeed-velocity/asv/issues/1027
+# https://github.com/airspeed-velocity/asv/issues/488
+benchmark-regen:
+	python -m asv run --config csp/benchmarks/asv.conf.jsonc --verbose v0.0.4^!
+	python -m asv run --config csp/benchmarks/asv.conf.jsonc --verbose v0.0.5^!
+
+benchmark-view:  ## generate viewable website of benchmark results
+	python -m asv publish --config csp/benchmarks/asv.conf.jsonc
+	python -m asv preview --config csp/benchmarks/asv.conf.jsonc
+
+# Alias
+benchmarks: benchmark
+benchmarks-regen: benchmark-regen
+benchmarks-view: benchmark-view
+
 ###########
 # VERSION #
 ###########

ci/benchmarks/benchmarks.json

@@ -0,0 +1,25 @@
+{
+    "stats.basic.StatsBenchmarkSuite.time_stats": {
+        "code": "class StatsBenchmarkSuite:\n    def time_stats(self, function):\n        def g():\n            data = csp.curve(typ=np.ndarray, data=self.data)\n            value = getattr(csp.stats, function)(data, interval=self.interval, **self.function_args.get(function, {}))\n            csp.add_graph_output(\"final_value\", value, tick_count=1)\n    \n        timer = Timer(\n            lambda: csp.run(g, realtime=False, starttime=self.start_date, endtime=timedelta(seconds=self.num_rows))\n        )\n        elapsed = timer.timeit(1)\n        return elapsed\n\n    def setup(self, _):\n        self.start_date = datetime(2020, 1, 1)\n        self.num_rows = 1_000\n        self.array_size = 100\n        self.test_times = [self.start_date + timedelta(seconds=i) for i in range(self.num_rows)]\n        self.random_values = [\n            np.random.normal(size=(self.array_size,)) for i in range(self.num_rows)\n        ]  # 100 element np array\n        self.data = list(zip(self.test_times, self.random_values))\n        self.interval = 500",
+        "min_run_count": 2,
+        "name": "stats.basic.StatsBenchmarkSuite.time_stats",
+        "number": 0,
+        "param_names": [
+            "function"
+        ],
+        "params": [
+            [
+                "'median'",
+                "'quantile'",
+                "'rank'"
+            ]
+        ],
+        "rounds": 2,
+        "sample_time": 0.01,
+        "type": "time",
+        "unit": "seconds",
+        "version": "f57f3ee288b0805597f9edee91b4d1dddf41046d34fbd46cfbd7135f459e62e3",
+        "warmup_time": -1
+    },
+    "version": 2
+}

conda/dev-environment-unix.yml

@@ -3,6 +3,7 @@ channels:
   - conda-forge
   - nodefaults
 dependencies:
+ - asv
  - bison
  - brotli
  - build

conda/dev-environment-win.yml

@@ -3,6 +3,7 @@ channels:
   - conda-forge
   - nodefaults
 dependencies:
+ - asv
  - brotli
  - build
  - bump2version>=1

csp/benchmarks/__init__.py

@@ -0,0 +1 @@
+from .common import *

csp/benchmarks/asv.conf.jsonc

@@ -0,0 +1,33 @@
+// https://asv.readthedocs.io/en/v0.6.3/asv.conf.json.html
+{
+    "version": 1,
+    "project": "csp",
+    "project_url": "https://github.com/Point72/csp",
+    "repo": "../..",
+    "branches": ["main", "tkp/bm"],
+    "dvcs": "git",
+
+    "install_command": ["in-dir={env_dir} python -mpip install {wheel_file}"],
+    "uninstall_command": ["return-code=any python -mpip uninstall -y {project}"],
+    "build_command": [
+        "python -m pip install build",
+        "python -m build --wheel -o {build_cache_dir} {build_dir}"
+    ],
+    "environment_type": "virtualenv",
+    "install_timeout": 600,
+    "show_commit_url": "http://github.com/point72/csp/commit/",
+
+    "pythons": ["3.11"],
+
+    // "environment_type": "mamba",
+    // "conda_channels": ["conda-forge"],
+    // "conda_environment_file": "conda/dev-environment-unix.yml",
+
+    "benchmark_dir": "../../csp/benchmarks",
+    "env_dir": "../../.asv/env",
+    "results_dir": "../../ci/benchmarks",
+    "html_dir": "../../.asv/html",
+
+    "hash_length": 8,
+    "build_cache_size": 2
+}

csp/benchmarks/common.py

@@ -0,0 +1,63 @@
+from asv_runner.benchmarks import benchmark_types
+from asv_runner.benchmarks.mark import SkipNotImplemented
+from logging import getLogger
+
+__all__ = ("ASVBenchmarkHelper",)
+
+
+class ASVBenchmarkHelper:
+    """A helper base class to mimic some of what ASV does when running benchmarks, to
+    test them outside of ASV.
+
+    NOTE: should be removed in favor of calling ASV itself from python, if possible.
+    """
+
+    def __init__(self, *args, **kwargs):
+        self.log = getLogger(self.__class__.__name__)
+
+    def run_all(self):
+        # https://asv.readthedocs.io/en/v0.6.3/writing_benchmarks.html#benchmark-types
+        benchmarks = {}
+
+        for method in dir(self):
+            for cls in benchmark_types:
+                if cls.name_regex.match(method):
+                    benchmark_type = cls.__name__.replace("Benchmark", "")
+                    if benchmark_type not in benchmarks:
+                        benchmarks[benchmark_type] = []
+
+                    name = f"{self.__class__.__qualname__}.{method}"
+                    func = getattr(self, method)
+                    benchmarks[benchmark_type].append(cls(name, func, (func, self)))
+
+        def run_benchmark(benchmark):
+            skip = benchmark.do_setup()
+            try:
+                if skip:
+                    return
+                try:
+                    benchmark.do_run()
+                except SkipNotImplemented:
+                    pass
+            finally:
+                benchmark.do_teardown()
+
+        for type, benchmarks_to_run in benchmarks.items():
+            if benchmarks_to_run:
+                self.log.warn(f"Running benchmarks for {type}")
+            for benchmark in benchmarks_to_run:
+                if len(getattr(self, "params", [])):
+                    # TODO: cleaner
+                    param_count = 0
+                    while param_count < 100:
+                        try:
+                            benchmark.set_param_idx(param_count)
+                            params = benchmark._current_params
+                            self.log.warn(f"[{type}][{benchmark.name}][{'.'.join(str(_) for _ in params)}]")
+                            run_benchmark(benchmark=benchmark)
+                            param_count += 1
+                        except ValueError:
+                            break
+                else:
+                    self.log.warn(f"Running [{type}][{benchmark.func.__name__}]")
+                    run_benchmark(benchmark=benchmark)

csp/benchmarks/stats/basic.py

@@ -0,0 +1,50 @@
+import numpy as np
+from datetime import datetime, timedelta
+from timeit import Timer
+
+import csp
+from csp.benchmarks import ASVBenchmarkHelper
+
+__all__ = ("StatsBenchmarkSuite",)
+
+
+class StatsBenchmarkSuite(ASVBenchmarkHelper):
+    """
+    python -m csp.benchmarks.stats.basic
+    """
+
+    params = (("median", "quantile", "rank"),)
+    param_names = ("function",)
+
+    rounds = 5
+    repeat = (100, 200, 60.0)
+
+    function_args = {"quantile": {"quant": 0.95}}
+
+    def setup(self, _):
+        self.start_date = datetime(2020, 1, 1)
+        self.num_rows = 1_000
+        self.array_size = 100
+        self.test_times = [self.start_date + timedelta(seconds=i) for i in range(self.num_rows)]
+        self.random_values = [
+            np.random.normal(size=(self.array_size,)) for i in range(self.num_rows)
+        ]  # 100 element np array
+        self.data = list(zip(self.test_times, self.random_values))
+        self.interval = 500
+
+    def time_stats(self, function):
+        def g():
+            data = csp.curve(typ=np.ndarray, data=self.data)
+            value = getattr(csp.stats, function)(data, interval=self.interval, **self.function_args.get(function, {}))
+            csp.add_graph_output("final_value", value, tick_count=1)
+
+        timer = Timer(
+            lambda: csp.run(g, realtime=False, starttime=self.start_date, endtime=timedelta(seconds=self.num_rows))
+        )
+        elapsed = timer.timeit(1)
+        return elapsed
+
+
+if __name__ == "__main__":
+    sbs = StatsBenchmarkSuite()
+    sbs.run_all()

pyproject.toml

@@ -83,6 +83,8 @@ develop = [
     "sqlalchemy",  # db
     "threadpoolctl",  # test_random
     "tornado",  # profiler, perspective, websocket
+    # benchmarks
+    "asv",
 ]
 showgraph = [
     "graphviz",