CTB Max Combo Approximation

ctb.py

#!/usr/bin/env python3

# We want to guess how many combo units a CTB slider tends to be worth.
# We'll be looking at all catch-specific beatmaps with an FC as their top play,
# from 2010 to the present (Nov. 2017). For each beatmap, we want to count
# the max combo of the FC (this is guaranteed to be the max combo), the number
# of regular hit objects, and the number of sliders.

# To estimate slider combo, we'll compute:
# slidercombo = ((max combo - normal objects) / sliders)
# A smarter model would account for slider length,
# slider type, slider velocity, and tick rate.

import json
import os
import re
import requests
import statistics

site_url = "https://osu.ppy.sh/b/"
osu_url = "https://osu.ppy.sh/osu/"
api_url = "https://osu.ppy.sh/api/get_beatmaps?k=%s&m=2" % os.environ["OSU_API_KEY"]  # noqa
combo_regex = re.compile("<td><strong>Max Combo</strong></td><td>([0-9]+)</td>")  # noqa
misses_regex = re.compile("<td><strong>Misses</strong></td><td>([0-9]+)</td>")
beatmaps = {}
date = "201x-01-01 00:00:00"

for year in range(7):  # 2010 - 2017
    maps = json.loads(requests.get("%s&since=%s" % (api_url, date.replace("x", str(year)))).text)  # noqa
    l = len(maps)
    for i, bmap in enumerate(maps):
        print("Map %d/%d (201%d)" % (i + 1, l, year))
        if bmap["approved"] not in ["4", "2", "1"]:
            print("Not ranked")
            continue

        b_id = bmap["beatmap_id"]
        if b_id in beatmaps:
            continue

        try:
            text = requests.get("%s%s" % (osu_url, b_id)).text
            website = requests.get("%s%s" % (site_url, b_id)).text
        except:
            print("Request error")
            continue

        if not website.startswith("\n"):
            # Some new maps are only available on the new site,
            # whose HTML I don't feel like looking at.
            print("New site")
            continue

        combo_matches = combo_regex.findall(website)
        if not combo_matches:
            print("No combo match")
            continue
        misses_matches = misses_regex.findall(website)
        if not misses_matches:
            print("No misses match")
            continue
        if misses_matches[0] != "0":
            print("Not FC")
            continue

        for i, line in enumerate(text.split("\n")):
            if "[HitObjects]" in line:
                break
        else:
            print("No hitobjects")
            continue

        beatmaps[b_id] = {}
        beatmaps[b_id]["combo"] = int(combo_matches[0])
        beatmaps[b_id]["sliders"] = []

        objects = 0
        for line in text.split("\n")[i+1:]:
            if "|" in line:  # A slider.
                beatmaps[b_id]["sliders"].append(line.strip())
            else:  # I don't think we can account for spinners unfortunately.
                objects += 1
        if not beatmaps[b_id]["sliders"]:
            print("No sliders")
            beatmaps.pop(b_id)
        else:
            beatmaps[b_id]["objs"] = objects

    with open("maps.json", "w") as f:
        json.dump(beatmaps, f)

slidercombos = [(b["combo"] - b["objs"]) / len(b["sliders"]) for b in beatmaps.values()]  # noqa

print("Mean: %f" % statistics.mean(slidercombos))
print("Median: %f" % statistics.median(slidercombos))
print("STD: %f" % statistics.stdev(slidercombos))

# Mean: 2.442353
# Median: 2.333743
# STD: 0.414389

# What have we learned?
# This tells us that sliders tend to be short.
# The minumum combo of a slider is 2, and so we can see that most
# sliders stick to this minimum. A quick AT playthrough of a few
# CTB maps confirms this as well.
# We can also see that long, slow sliders pull the mean up a little bit,
# while there's no possibility for low outliers.