do 10-fold cross-validation using libsvm and the iris dataset.

crossfold.sh

#!/bin/bash

libsvmpath="./libsvm-3.21"

# 1. build the libsvm
cd $libsvmpath
make

if [ $? -ne 0 ]; then
    echo "failed to build, exit";
    exit 1;
fi

cd ../

# 2. set the config variables
dataset=./iris.scale.all
model=$libsvmpath/iris.model
trainbin=$libsvmpath/svm-train
predictbin=$libsvmpath/svm-predict
output=./iris.predicted.out

fold=10 # do 10-fold cross validation
gamma_choices="0.001 0.005 0.01 0.05 0.1 0.5 1 2 5 10 20 50 100 200 1000"

# 3. find the parameter that achieves the best 10-fold accuracy
maxacc=0
maxgamma=0
for gamma in $gamma_choices; do
    acc=$($trainbin -g $gamma -v $fold $dataset $model | tail -1 | awk -F '[ \t%]' '{print $(NF - 1)}')
    echo "gamma=$gamma	acc=$acc"
    
    cond=$(echo "$acc > $maxacc" | bc)
    if [ $cond -gt 0 ]; then
        maxacc=$acc
        maxgamma=$gamma
        
        echo "max gamma changed: $gamma"
    fi
done

# 4. retrain the model using the best parameter
$trainbin -g $gamma $dataset $model

# 5. use the model to do testing
$predictbin $dataset $model $output