Optimized ( Rect/Frect ) unionall()/unionall_ip()

[itzpr3d4t0r]

This change is comprised of two changes:

• made unionall/unionall_ip METH_O
• Optimized looping with lists/tuples

Makes the function 80% faster on average (with rect lists or tuples) from my testing.

Test program:

from data_utils import Plotter, Evaluator
from pygame import Rect
from random import randint

def tests_setup(curr_size: int, g: dict):
    r = Rect(0, -10, 40, 23)
    rects = [
        Rect(randint(-100, 100), randint(-100, 100), randint(-100, 100), randint(-100, 100))
        for _ in range(curr_size)
    ]

    g["r"] = r
    g["rects"] = rects


tests = [
    ("unionall new", "r.unionall(rects)"),
    ("unionall_ip new", "r.unionall_ip(rects)"),
]

Evaluator(tests, tests_setup, max_size=2500, reps=1000, num=1).run()

files = [
    ("unionall old", "red"),
    ("unionall_ip old", "white"),

    ("unionall new", "blue"),
    ("unionall_ip new", "lime"),
]

p = Plotter("alphablit", files, mode="MIN")
p.plot_tests()

data_utils file:

import json

from matplotlib import pyplot as plt
import matplotlib.colors as mcolors

from statistics import mean, stdev, median
from timeit import repeat

COLORS = list(mcolors.CSS4_COLORS.keys())

__all__ = ["Plotter", "Evaluator"]

JSON_DIR = "files_json"

class Plotter:

    def __init__(self, title: str, tests: list, mode: str = "MIN",
                 limit_to_range: int = -1):
        plt.style.use(["dark_background"])
        self.title = title
        self.tests = tests

        self.mode = mode
        self.mode_func = None
        self.limit_to_range = limit_to_range
        self.filter(mode)

    def filter(self, mode: str):
        self.mode = mode

        match mode:
            case "MEAN":
                self.mode_func = mean
            case "MIN":
                self.mode_func = min
            case "MAX":
                self.mode_func = max
            case "MEDIAN":
                self.mode_func = median

    def plot_tests(self, scatter=False):
        for file_name, color in self.tests:
            try:
                with open(f"{JSON_DIR}/{file_name}.json", "r") as f:
                    data = json.load(f)
            except FileNotFoundError:
                print(f"File {file_name}.json not found!")
                quit()

            timings = [self.mode_func(dp) for dp in data["data"]][:self.limit_to_range]

            print(f"=== {file_name} ===")
            print(
                f"Total: {sum([sum(data_point) for data_point in data['data']])}\n"
                f"Mean: {mean(timings)}\n"
                f"Median: {median(timings)}\n"
                f"Stdev: {stdev(timings)}"
            )
            print()
            if scatter:
                plt.scatter(range(len(timings)), timings, color=color, label=file_name, s=1)
            else:
                plt.plot(timings, color=color, label=file_name, linewidth=1)
        plt.legend()
        plt.title(self.title)
        plt.xlabel("Surface size (px)")
        plt.ylabel("Time (s)")
        plt.show()

    def compare(self, indices: list[tuple[int, int]], c1="white", c2="lime"):
        for i1, i2 in indices:
            filename_1, _ = self.tests[i1]
            filename_2, _ = self.tests[i2]

            try:
                with open(f"{JSON_DIR}/{filename_1}.json", "r") as f:
                    data_1 = json.load(f)
            except FileNotFoundError:
                print(f"File {filename_1}.json not found!")
                quit()

            try:
                with open(f"{JSON_DIR}/{filename_2}.json", "r") as f:
                    data_2 = json.load(f)
            except FileNotFoundError:
                print(f"File {filename_2}.json not found!")
                quit()

            timings_1 = [self.mode_func(dp) for dp in data_1["data"]][
                        :self.limit_to_range]
            timings_2 = [self.mode_func(dp) for dp in data_2["data"]][
                        :self.limit_to_range]

            plt.figure(figsize=(10, 5))
            curr_index = 1

            plt.subplot(len(indices), 2, curr_index)
            plt.scatter(range(len(timings_1)), timings_1, color=c1, label=filename_1,
                        s=.5)
            plt.scatter(range(len(timings_1)), timings_2, color=c2, label=filename_2,
                        s=.5)
            plt.legend()
            plt.title("Timings")
            plt.xlabel("Surface size (px)")
            plt.ylabel("Time (s)")

            plt.subplot(len(indices), 2, curr_index + 1)
            comparative_data = [
                100 * ((t1 / t2) - 1) for t1, t2 in zip(timings_1, timings_2)
            ]
            plt.scatter(range(len(comparative_data)), comparative_data, color="red", s=1)
            plt.plot([0] * len(comparative_data), color="green", linewidth=2)
            plt.title("Relative % improvement")
            plt.xlabel("Surface size (px)")

        plt.show()


class Evaluator:
    def __init__(self, tests: list, tests_setup, pre_test_setup=None,
                 max_size: int = 1000, reps: int = 30,
                 num: int = 1):
        self.tests = tests
        self.tests_setup = tests_setup
        self.max_size = max_size
        self.reps = reps
        self.num = num
        self.G = {}

        if pre_test_setup:
            try:
                pre_test_setup(self.G)
            except TypeError:
                raise TypeError("pre_test_setup must be a callable")

    def run(self):
        for test_name, statement in self.tests:
            data = {"title": test_name, "data": []}

            print(f"\n========| {test_name.upper()} |========")

            for curr_size in range(1, self.max_size + 1):

                self.print_progress_bar(curr_size)

                self.tests_setup(curr_size, self.G)
                data["data"].append(self.run_test(statement))

            with open(f"{JSON_DIR}/{test_name}.json", "w") as f:
                json.dump(data, f)

    def print_progress_bar(self, curr_size: int):
        amt = 3 * curr_size // 100
        print(
            "\r[" + "▪" * amt + " " * (3 * 10 - amt) + f"] {curr_size} | {self.max_size}", end=""
        )

    def run_test(self, statement):
        return repeat(statement, globals=self.G, number=self.num, repeat=self.reps)

[oddbookworm]

I don't see anything that sticks out to me as a problem, so looks good!

[MyreMylar]

OK, LGTM 👍

I can verify the performance improvements in unioning lots of rectangles. I'm sure there is somebody out there doing something like that who will appreciate it being faster.