Fitness - Githubissues

alan-turing-institute / p2lab-pokemon

A Python library for running genetic algorithms to optimize Pokemon teams!

BSD 3-Clause "New" or "Revised" License

8 stars 1 forks source link

Fitness #8

Closed philswatton closed 1 year ago

philswatton commented 1 year ago

This PR implements two candidate functions for the concept of 'fitness' in the genetic algorithm, both in src/genetic/fitness.py

They are:

BTmodel(): implements the Bradley-Terry model with an iterative estimation procedure summarised here: https://en.wikipedia.org/wiki/Bradley%E2%80%93Terry_model
- I had initially thought this would need to be estimated via a logit model, but this seems like a nicer solution
- The advantage of this model is that even if all teams don't face all teams, so long as we have a good number of matches (and teams are reasonably well-connected to each other in a graph structure) then it should take into account e.g. a team facing only strong teams
- Because of the way this has been computed, all fitness values will sum to 1
- win_percentages(): Implements a the ratio of matches a team has won to matches a team has played
- Nice and straightforward
- Results will however need normalising - this could be added as a parameter to the function
- Some stuff renamed in the genetic algorithm psuedocode to avoid clashes in argument names

philswatton commented 1 year ago

Also adds the venv to gitignore

phinate commented 1 year ago

Nice! I'll merge this now and coerce it with the other PRs :)