Francesco149 / pyttanko

osu! pp and difficulty calculator, pure python implementation of https://github.com/Francesco149/oppai-ng
The Unlicense
20 stars 8 forks source link

.. image:: https://travis-ci.org/Francesco149/pyttanko.svg?branch=master :target: https://travis-ci.org/Francesco149/pyttanko

osu! pp and difficulty calculator. pure python implementation of https://github.com/Francesco149/oppai-ng

this is meant to be a standalone single-file module that's as portable as possible using only python 2.6+ builtins with no extra dependencies

if you need a command line interface, check out oppai-ng <https://github.com/Francesco149/oppai-ng>_

if you need a more object oriented implementation, check out oppadc <https://github.com/The-CJ/oppadc.py>_

usage

pyttanko is a single-file module, so the simplest way to use it is to simply drop it in your project's folder:

.. code-block:: sh

cd my/project
curl https://raw.githubusercontent.com/Francesco149/pyttanko/master/pyttanko.py > pyttanko.py

this way, anyone who clones your project won't have to install pyttanko as it will be bundled

if you prefer, it's also available on pip

.. code-block:: sh

pip install pyttanko

or you can also manually install like so:

.. code-block:: sh

curl -L https://github.com/Francesco149/pyttanko/archive/HEAD.tar.gz -o HEAD.tar.gz
cd pyttanko-*
python setup.py install --user

check out

.. code-block:: sh

pydoc pyttanko

or

.. code-block:: sh

python -c "help('pyttanko')"

for the full documentation

minimal example:

.. code-block:: python

#!/usr/bin/env python

import sys
import pyttanko as osu

p = osu.parser()
bmap = p.map(sys.stdin)

stars = osu.diff_calc().calc(bmap)
print("%g stars" % stars.total)

pp, _, _, _, _ = osu.ppv2(stars.aim, stars.speed, bmap=bmap)
print("%g pp" % pp)

which you can run with:

.. code-block:: sh

cat /path/to/file.osu | ./example.py

performance

pyttanko runs the test suite over 10 times slower than the original C implementation and uses ~8 times more memory, so if you need to batch process thousands of scores, you should consider writing native bindings for the C version.

tests were performed on linux 4.9.38, python 2.7.10 on a i7-4790k

this is still a pretty respectable speed considering python is interpreted

.. code-block:: sh

$ cd ~/src/pyttanko/
$ time -v ./run_test
...
    Command being timed: "./run_test"
    User time (seconds): 101.68
    System time (seconds): 0.61
    Percent of CPU this job got: 99%
    Elapsed (wall clock) time (h:mm:ss or m:ss): 1m 42.34s
    Average shared text size (kbytes): 0
    Average unshared data size (kbytes): 0
    Average stack size (kbytes): 0
    Average total size (kbytes): 0
    Maximum resident set size (kbytes): 88688
    Average resident set size (kbytes): 0
    Major (requiring I/O) page faults: 0
    Minor (reclaiming a frame) page faults: 631637
    Voluntary context switches: 1
    Involuntary context switches: 4116
    Swaps: 0
    File system inputs: 0
    File system outputs: 56
    Socket messages sent: 0
    Socket messages received: 0
    Signals delivered: 0
    Page size (bytes): 4096
    Exit status: 0

$ cd ~/src/oppai-ng/test
$ ./build
$ time -v ./oppai_test
...
    Command being timed: "./oppai_test"
    User time (seconds): 9.09
    System time (seconds): 0.06
    Percent of CPU this job got: 99%
    Elapsed (wall clock) time (h:mm:ss or m:ss): 0m 9.15s
    Average shared text size (kbytes): 0
    Average unshared data size (kbytes): 0
    Average stack size (kbytes): 0
    Average total size (kbytes): 0
    Maximum resident set size (kbytes): 11840
    Average resident set size (kbytes): 0
    Major (requiring I/O) page faults: 0
    Minor (reclaiming a frame) page faults: 304
    Voluntary context switches: 1
    Involuntary context switches: 39
    Swaps: 0
    File system inputs: 0
    File system outputs: 0
    Socket messages sent: 0
    Socket messages received: 0
    Signals delivered: 0
    Page size (bytes): 4096
    Exit status: 0