Rouletter implements roulette wheel, or fitness proportionate selection.

This implementation is built for performance - the critical parts are in C++, accessed from Python via Cython.

A roulette wheel provides a way of making a random selection from 37 (or 38) slots, all equally likely. If you roll the ball many times, these samples build up a picture of the distribution of slot sizes (in the casino, a uniform distribution). A roulette wheel selector is a generalisation of this process - you can set up a wheel with as many slots as you like, and they don't have to all be of equal size.

As a simple example, let's imagine a wheel with four slots, and with the second slot twice the width of the others, i.e., [1,2,1,1]. The wheel has slots corresponding to these sizes, and when we repeatedly spin the wheel and release a ball, over time it should land twice as often in the second slot. Rouletter simulates this kind of selection.

To use, initialise the Rouletter object with an array of integers that define the slot widths, then call spin with a random number between 0.0 and 1.0.

  import random, Rouletter
  widths = [1,2,1,1]
  rw = Rouletter.RouletteSelector(widths)
  print rw.spin(random.random())
  #test distribution looks OK
  hits = [0]*len(widths)
  for i in xrange(1000): hits[rw.spin(random.random())] +=1
  print hits

To build the code from source, ensure you have Cython and suitable compilers installed and call the setup.py script. For my local windows environment, I use:

  python setup.py build_ext --inplace