Performance?

[michalrus]

Hello!

Thank you for this wonderful piece of code. =)

Have you tried tuning the performance?

E.g. for 10,000 random matches (each of max. 5 teams of max. 10 members each) it consistently takes:

• 79.41 s to calculate new ratings (so ~8 ms per parallelized/amortized rate()),
• 16.52 s to calculate match qualities (~2 ms per parallel quality())

… on 4 cores (and 4 parallel python3 processes).

[michalrus]

There’s also some C++ version that doesn’t do teams but is ~200× faster → https://github.com/JesseBuesking/trueskill#benchmark .

[bernd-wechner]

At a random guess I'd say you're just looking at the difference between Python and C++ in terms of performance. But I'm guessing if you were keen you could profile both of them and look for the bottlenecks, try doing a flame graph, very informative:

http://www.brendangregg.com/flamegraphs.html

[michalrus]

Very cool, I’ll try that! Thanks. =)

[sublee]

When I implemented it, performance was not one of the goals. The goal is to implement the TrueSkill algorithm with clean code for ease of understanding.

If you need better performance, another implementation (perhaps, with another language) will be helpful rather than this project.