/xcsf

XCSF learning classifier system (evolved hyperrectangle, MLP neural network, GP-tree, or DGP-graph conditions, and least squares or backpropagation neural network predictions) for function approximation/regression.

Primary LanguageCGNU General Public License v2.0GPL-2.0

=========================================================== XCSF learning classifier system

Evolved Conditions:
CON = 0: real-valued hyperrectangle intervals
CON = 1: multilayer perceptron neural networks
CON = 2: GP trees
CON = 3: dynamical GP graphs

Mutation for conditions:
SAM = 0: fixed rate
SAM = 1: self-adaptive rate

Computed Predictions:
PRE = 0: linear least squares (aka modified Delta update or Widrow-Hoff).
PRE = 1: quadratic least squares
PRE = 2: linear recursive least squares
PRE = 3: quadratic recursive least squares
PRE = 4: backpropagation multilayer perceptron neural networks

An updated GNUPlot of the current system error can be enabled by compiling with
GNUPLOT=1 (on GNU/Linux gnuplot-x11 must be installed.)

The matching and set prediction functions (where most processing occurs) can be
parallelised using OpenMP by compiling with PARALLEL=1.

Example usage:

To initiate the learning of in/sine_1var_train.dat and test performance on in/sine_1var_test.dat run:

xcsf sine_1var