go-gsgp

go-gsgp

Geometric Semantic Genetic Programming ported to Go

The original code (by Mauro Castelli) is available at http://gsgp.sf.net

This version is compatible with the version 1.0 and features:

• shorter, safer Go code;
• better reading of configuration files;
• better handing of command line arguments.

Building

To generate protobuf code, get protobuf on your host, then

go get github.com/golang/protobuf/protoc-gen-go

To include git version number, compile using:

go build -ldflags "-X main.gitCommit=`git describe --long --dirty --tags --always`"

The commit will show up when running with -version flag.

Usage

go get github.com/akiross/go-gsgp
$GOPATH/bin/go-gsgp -train_file train_dataset -test_file test_dataset

To change parameters, edit the configuration.ini file.

The train and test files have the following format:

n_VARS
m_EXAMPLES
V11 V12 V3 ... V1n T
V21 V22 V3 ... V2n T
...
Vm1 Vm2 V3 ... Vmn T

Where, the first line contains the number n of variables, the second line contains the number m of cases in the dataset. Then, follow m lines of n+1 space-separated columns, where the last column is the target value.

Initialization via semantic feeding

Instead of randomly generating all the initial individuals, and computing their semantic, it is possible to feed some pre-computed semantics via files.

To do so, provide a list of files as positional arguments to the program:

$ go-gsgp [options] semantic1 semantic2 ... semanticN

The files consist in a semantic vector with one value per line. The file shall contain the semantic values for the training set followed by the test values.