Model Optimizer

Fit model of dynamical system to data

Usage

$ python main.py LotkaVolterra result.pkl
$ python analyze_result.py result.pkl

Tests

$ nosetests

Presets

Exemplary function definitions can be found in ./presets/examples.py.

Features

• Initialization
  • Randomized individuals
• Selection
  • Fit individuals are more likely to be chosen for reproduction
• Crossover
  • Coefficient pulling
    • Randomized scale to prevent cycles
  • Coefficient sign switch
  • Exchange subtrees between terms
• Mutation
  • Add scaled Gauss to coefficient
  • Switch function with operator of same arity or different arity with randomized arguments
• Misc
  • Keep elite (fittest individuals) over generations
  • Apply culling technique
  • Pause procedure and inspect generation with [CTRL]+[Z]
  • Optionally fix certain coefficients
• Fitness
  • Root relative squared error of series with multiple initial conditions