Skip to content

Instantly share code, notes, and snippets.

@smoogipoo

smoogipoo/doc.md

Last active October 6, 2021 10:53
Show Gist options
  • Save smoogipoo/2533f3872237623d2f0fdb890b02de7e to your computer and use it in GitHub Desktop.
Save smoogipoo/2533f3872237623d2f0fdb890b02de7e to your computer and use it in GitHub Desktop.
Download ZIP
Dockerized diffcalc runner doc
Raw
doc.md

Inputs

./run.sh      [sr|pp|all]     [--local-only]     [osu|taiko|catch|mania] [osu|taiko|catch|mania]
               ^^^^^^^^^       ^^^^^^^^^^^^       ^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^
               arg:mode       opt:local-only               arg:ruleset (one-or-more)

Outputs

For each ruleset, a set of .csv file outputs generated from the database via queries described at the bottom of this document, containing:

  1. If the mode arg is sr or all:
    • [ruleset]_sr_all_gains.csv - Star rating diffs for all mods sorted by gains.
    • [ruleset]_sr_all_losses.csv - Star rating diffs for all mods sorted by losses.
    • [ruleset]_sr_nomod_gains.csv - Star rating diffs for no-mod sorted by gains.
    • [ruleset]_sr_nomod_losses.csv - Star rating diffs for no-mod sorted by losses.
  2. If the mode arg is pp or all:
    • [ruleset]_pp_all_gains.csv - PP diffs for all mods sorted by gains.
    • [ruleset]_pp_all_losses.csv - PP diffs for all mods sorted by losses.
    • [ruleset]_pp_nomod_gains.csv - PP diffs for no-mod sorted by gains.
    • [ruleset]_pp_nomod_losses.csv - PP diffs for no-mod sorted by losses.

The user should additionally be able to access the database by themselves without

Structure

Inside a cloned repo should primarily exist three repository submodules and a run script:

top_level
    | /osu (submodule -> https://github.com/ppy/osu)
    | /osu-performance (submodule -> https://github.com/ppy/osu-performance)
    | /osu-difficulty-calculator (submodule -> https://github.com/ppy/osu-difficulty-calculator)
    | run.sh

In order to use this tool, users should navigate into each of the submodules and checkout the appropriate commits. For example, a change to performance calculators in the osu submodule will require a respective change in the osu-performance submodule.
This is because the tool should be able to be used to test changes before they're merged.

Run

Setup

  1. Set up a MYSQL database.
  2. Download, extract, and import data from https://data.ppy.sh containing top 10000 scores into two databases.
    • curl https://data.ppy.sh/ | grep performance_${{ matrix.ruleset.name }}_top_1000 | tail -1 | awk -F "\"" '{print $2}' | sed 's/\.tar\.bz2//g'
    • Call the databases osu_master and osu_local respectively.
    • This should be included in the database setup image.
  3. Download and extract beatmap data from https://data.ppy.sh to ./beatmaps.
    • curl https://data.ppy.sh/ | grep osu_files | tail -1 | awk -F "\"" '{print $2}' | sed 's/\.tar\.bz2//g'
    • This should perist to the checkout rather than be included in a docker image (it acts as a cache and is mutable between runs).

Now we'll run osu-difficulty-calculator and osu-performance in sequence. This should be done in docker containers of their own so as to isolate the build toolchains from the host system.

It's important to note which database the processes are running against:

  • We always need to run the local checkouts from the structure above against the osu_local table.
  • If the --local-only option is not set, then need to checkout the HEADs of ppy/osu, ppy/osu-difficulty-calculator, and ppy/osu-performance, and repeat the process from those checkouts against the osu_master table.
    • This should be done by cloning the repositories into a temporary directory.

Global environment variables

CONCURRENCY=4

Diffcalc

  1. Pass ./beatmaps as a volume.
  2. Run ./UseLocalOsu.sh
  3. Set environment variables:
    • DB_HOST={docker database}
    • DB_NAME=osu_local
    • BEATMAPS_PATH={beatmaps volume path}
    • SAVE_DOWNLOADED=1
    • ALLOW_DOWNLOAD=1
  4. Map each ruleset to their respective ID (osu:0, taiko:1, catch:2, mania:3).
  5. Run dotnet run -c:Release -- all -ac -c {CONCURRENCY}.
    • Add options to the run string for all relevant ruleset IDs as -m [ruleset_id].
    • If possible, add a global environment variable like INCLUDE_CONVERTS to enable/disable the -ac option.

PPcalc

  1. Apply the osu-performance patch listed at the bottom of this document.
  2. cd build && cmake .. && make -j && cd ../bin
  3. Update the contents of config.json as follows: 
        {
        "mysql.master.host" : "{docker database}",
        "mysql.master.port" : 3306,
        "mysql.master.username" : "root",
        "mysql.master.password" : "",
        "mysql.master.database" : "osu_local"
    }
  4. Once per ruleset name, run ./osu-performance all -t {CONCURRENCY} -m [ruleset].

Generate

Generate outputs by running the SQL queries listed at the bottom of this document.

Example usage

# I want to test the changes out and generate a full overview of SR + PP changes made in the osu and catch rulesets.

# First I'll clone the tester repository,
git clone --recurse-submodules ppy/diffcalc-tester

# Checkout the relevant commits containing the changes
cd osu
git remote add smoogipoo https://github.com/smoogipoo/osu
git fetch smoogipoo && git checkout XXX

cd ../osu-difficulty-calculator
git remote add smoogipoo https://github.com/smoogipoo/osu-difficulty-calculator
git fetch smoogipoo && git checkout YYY

cd ../osu-performance
git remote add smoogipoo https://github.com/smoogipoo/osu-performance
git fetch smoogipoo && git checkout ZZZ

# A bit of setup. Since I have a beefy PC, I can expend 16 cores.
export CONCURRENCY=16

# Run the process
cd ..
./run.sh all osu catch

# Wait for process to complete (runs diffcalc, ppcalc, generates .csv data)...

# Upload .csv data to Google Sheets, or add as artifacts in GH actions workflow, or output to command line.
# Can be a separate user step, depending on usage scenario. Most common for me will be the Google Sheets path.

# New changes have been made. I need to re-run and generate more data!

# Since the DB is already in the correct state, I'll run only the local checkouts
./run.sh all --local-only osu catch

# And once again do stuff with the .csv data.

Queries

SR diffs

SELECT
    m.beatmap_id,
    m.mods,
    b.filename,
    m.diff_unified as 'sr_master',
    p.diff_unified as 'sr_pr',
    (p.diff_unified - m.diff_unified) as 'diff',
    (p.diff_unified / m.diff_unified - 1) as 'diff%'
FROM osu_master.osu_beatmap_difficulty m
JOIN osu_local.osu_beatmap_difficulty p
    ON m.beatmap_id = p.beatmap_id
    AND m.mode = p.mode
    AND m.mods = p.mods
JOIN osu_local.osu_beatmaps b
    ON b.beatmap_id = p.beatmap_id
WHERE
    # Comment/uncomment the following line to generate combined or nomod-only diffs.
    m.mods = 0 AND
    abs(m.diff_unified - p.diff_unified) > 0.1
ORDER BY p.diff_unified - m.diff_unified
    # If sort-by-gains, use DESC. If sort-by-losses, use ASC.
    DESC
LIMIT 10000;

PP diffs

SELECT
    m.score_id,
    m.beatmap_id,
    m.enabled_mods,
    b.filename,
    m.pp as 'pp_master',
    p.pp as 'pp_pr',
    (p.pp - m.pp) as 'diff',
    (p.pp / m.pp - 1) as 'diff%'
FROM osu_master.osu_scores_fruits_high m
JOIN osu_local.osu_scores_fruits_high p
    ON m.score_id = p.score_id
JOIN osu_local.osu_beatmaps b
    ON b.beatmap_id = p.beatmap_id
WHERE
    # Comment/uncomment the following line to generate combined or nomod-only diffs.
    m.enabled_mods = 0 AND
    abs(m.pp - p.pp) > 0.1
ORDER BY p.pp - m.pp
    # If sort-by-gains, use DESC. If sort-by-losses, use ASC.
    DESC
LIMIT 10000;

osu-performance patch

diff --git a/src/performance/Score.cpp b/src/performance/Score.cpp
index b2d4916..3bd16f6 100644
--- a/src/performance/Score.cpp
+++ b/src/performance/Score.cpp
@@ -48,7 +48,7 @@ void Score::AppendToUpdateBatch(UpdateBatch& batch) const
 		_scoreId
 	));
 
-	batch.AppendAndCommitNonThreadsafe(StrFormat("UPDATE `score_process_queue` SET `status` = 1 WHERE `mode` = {0} AND `score_id` = {1};", static_cast<int>(_mode), _scoreId));
+	//batch.AppendAndCommitNonThreadsafe(StrFormat("UPDATE `score_process_queue` SET `status` = 1 WHERE `mode` = {0} AND `score_id` = {1};", static_cast<int>(_mode), _scoreId));
 }
 
 PP_NAMESPACE_END
Sign up for free to join this conversation on GitHub. Already have an account? Sign in to comment