AnovaTable RegSuf::anova()const{
AnovaTable ans;
double nobs = n();
double p = size(); // p+1 really
ans.SSE = SSE();
ans.SST = SST();
ans.SSM = ans.SST - ans.SSE;
ans.df_total = nobs-1;
ans.df_error = nobs - p;
ans.df_model = p-1;
ans.MSE = ans.SSE/ ans.df_error;
ans.MSM = ans.SSM/ans.df_model;
ans.F = ans.MSM/ans.MSE;
ans.p_value = pf(ans.F, ans.df_model, ans.df_error, false, false);
return ans;
}
int main(int argc, char *argv[]){
/* Itteration Variables */
int i = 1;
int gen = 0;
double bestSSE = DBL_MAX;
/* Creating performance variables */
double genDiv = 0;
double genSSE[3] = {0,0,0}; /* worst, avg, best */
double sseError[MAXPOP];
double val[NUMPOINTS][2];
/* Genetic Paramenters */
int populationSize = 500; /* Population size (constant) */
int treeDepth = 15; /* Maximum Tree Depth */
int maxGenerations = 50; /* Maximum Number of generations */
double sseGoal = 0.5; /* SSE error goal (for spartan) */
double pruneFactor = 0.0; /* Probability that a branch will be prunned */
double constProb = 0.60; /* Probability that a leaf will be a randomly choose (0,1) constant */
double mutationRate = 0.20; /* Mutation Rate (probability that a node will be mutated */
double swapRate = 0.7; /* Probability that crossover will occur */
double tournamentFraction = 0.40; /* fraction of individuals selected by tournament */
double rankFraction = 0.50; /* fraction of individual selected by rank */
double spartanFraction = 0.1; /* fraction of individuals selected by rank, copied as is */
struct geneticParam genParam; /* compat representation of generation */
node *forest[MAXPOP]; /* Forest of trees */
node *bestTree;
/* Output Variables */
int train = 1;
int performance = 1;
double evalPoint = 0;
FILE *out = stdout;
char bestTreeName[128];
char response[128] = "Y";
strcpy(bestTreeName,"bestTree");
/* Processing Command Line Inputs */
for (i = 1; i < argc; i++){
if (strcmp(argv[i],"--maxDepth")==0){
sscanf(argv[++i],"%d",&treeDepth);
}
else if (strcmp(argv[i],"--test")==0){
test();
return EXIT_SUCCESS;
}
else if (strcmp(argv[i],"--popSize")==0){
populationSize = atoi(argv[++i]);
}
else if (strcmp(argv[i],"--pruneFactor")==0){
pruneFactor = (double)atof(argv[++i]);
}
else if (strcmp(argv[i],"--constProb")==0){
constProb = (double)atof(argv[++i]);
}
else if (strcmp(argv[i],"--sseGoal")==0){
sseGoal = (double)atof(argv[++i]);
}
else if (strcmp(argv[i],"--mutationRate")==0){
mutationRate = (double)atof(argv[++i]);
}
else if (strcmp(argv[i],"--swapRate")==0){
swapRate = (double)atof(argv[++i]);
}
else if (strcmp(argv[i],"--tournamentFraction")==0){
tournamentFraction = (double) atof(argv[++i]);
rankFraction = 1.0 - tournamentFraction;
}
else if (strcmp(argv[i],"--rankFraction")==0){
rankFraction = (double) atof(argv[++i]);
}
else if (strcmp(argv[i],"--spartanFraction")==0){
spartanFraction = (double) atof(argv[++i]);
}
else if (strcmp(argv[i],"--maxGen")==0){
maxGenerations = atoi(argv[++i]);
}
else if(strcmp(argv[i],"--bestTreeName")==0){
strcpy(bestTreeName,argv[++i]);
}
else if (strcmp(argv[i],"--help")==0){
usage(stdout,argv[0]);
exit(EXIT_SUCCESS);
}
else{
fprintf(stderr,"Argument %s is not reconized\n",argv[i]);
usage(stderr,argv[0]);
exit(EXIT_FAILURE);
}
}
/* Checking Input Arguments */
if (populationSize > MAXPOP){
fprintf(stderr,"population size is set to the max at %d. Change define for larger population\n",MAXPOP);
populationSize = MAXPOP;
}
if (mutationRate > 1)
mutationRate = 0.1;
if (swapRate > 1)
swapRate = 0.1;
if (tournamentFraction + rankFraction > 1.0){
fprintf(stderr,"Tournament Fraction %5.3f and rank fraction %5.3f are greater than 1.0\n",
tournamentFraction,rankFraction);
fprintf(stderr,"Both are set to default values\n");
tournamentFraction = 0.9;
rankFraction = 1.0 - tournamentFraction;
}
genParam.mutationRate = mutationRate;
genParam.swapRate = swapRate;
genParam.touramentFraction = tournamentFraction;
genParam.rankFraction = rankFraction;
genParam.spartanFraction = spartanFraction;
genParam.constProb = constProb;
genParam.maxDepth = treeDepth;
genParam.pruneFraction = pruneFactor;
splash(out);
/* Train or Run? */
if (argc <= 1){
fprintf(stdout,"Preform symbolic regression with genetic programming [y/n]:\n");
fscanf(stdin,"%s",response);
if (strcmp(response,"y")==0 || strcmp(response,"Y")==0)
train = 1;
else
train = 0;
fprintf(stdout,"Do you want to print the performance of the best tree [y/n]: \n");
fscanf(stdin,"%s",response);
if (strcmp(response,"y")==0 || strcmp(response,"Y")==0)
performance = 1;
else
performance = 0;
}
/* Run Information */
fprintf(out,"Parameters:\n");
fprintf(out,"\tPopulation Size: %d\n\tMax Tree Depth: %d\n",populationSize,treeDepth);
fprintf(out,"\tPrune factor: %3.2f\n\tConstant Probability: %3.2f\n",pruneFactor,constProb);
fprintf(out,"\tSSE Goal: %3.2f\n\tMax Generations: %d\n",sseGoal,maxGenerations);
fprintf(out,"\tSpartan Fraction: %3.2f\n",spartanFraction);
fprintf(out,"\tTournament Fraction: %3.2f\n\tRank Fraction: %5.2f\n",tournamentFraction,rankFraction);
fprintf(out,"\tMutation Rate: %3.2f\n\tSwap Rate: %3.2f\n",mutationRate,swapRate);
printSet();
/* Reading in the data file */
importData("proj2-data.dat",val);
/* Creating the intitial population */
if (train){
srand( time(NULL));
fprintf(stdout,"Creating Intial Population\n");
rampedHalfHalf(forest,populationSize,&genParam);
/* Running Generations */
fprintf(out,"Generation\tDiversity\tMean SSE\tBest SSE\n");
while(gen < maxGenerations && bestSSE > sseGoal){
/* Diversity and SSE */
genDiv = diversity(forest,populationSize);
bestSSE = SSE(forest,populationSize,val,genSSE,sseError,bestTreeName);
fprintf(out,"\t%d\t%3.2f\t\t%3.2e\t%3.2e\n",gen,genDiv,genSSE[1],genSSE[2]);
/* Genetic Operations */
breedGeneration(forest,populationSize,sseError,&genParam);
gen++;
}
/* Diversity and SSE */
genDiv = diversity(forest,populationSize);
bestSSE = SSE(forest,populationSize,val,genSSE,sseError,bestTreeName);
fprintf(out,"\t%d\t%3.2f\t\t%3.2e\t%3.2e\n",gen,genDiv,genSSE[1],genSSE[2]);
/* Clean up, clean up, everybody do your share */
deleteForest(forest,populationSize);
}
/* Looking at the performance of the best tree */
if (performance){
sprintf(response,"%s.postfix",bestTreeName);
bestTree = bestTreeSummary(out,response,val);
if (argc <= 1){
fprintf(stdout,"Enter value on which to test the tree [n to escape]: \n");
while(fscanf(stdin,"%lf",&evalPoint)==1){
fprintf(stdout,"Tree(%5.3f) = %5.3f\n",evalPoint,evalTree(bestTree,evalPoint));
}
deleteTree(bestTree);
}
}
return EXIT_SUCCESS;
}
double exp1Var(unsigned n, double *x, double *grad, void *my_func_data) {
HWModel *model=(HWModel*)my_func_data;
int fup,fdown;
double *level=model->super.level;
double *trend=model->super.trend;
double *season=model->super.season;
double dotrend=(model->super.dotrend);
double doseasonal=(model->super.doseasonal);
int period=(model->super.period);
int seasonal=(model->super.seasonType); //Season additive oder multiplikative
int xl=(model->obsCount);
double a=(model->super.a);
double b=(model->super.b);
double alpha=x[0];
double beta=model->super.beta;
double gamma=model->super.gamma;
int start_time=(model->super.start_time);
double smape=0;
double ga=1e-5;
double xhat = 0, stmp = 0;
int i, i0, s0;
model->expectedValueXSquared=0;
model->expectedValueX=0;
/* copy start values to the beginning of the vectors */
level[0] = a;
if (dotrend == 1) trend[0] = b;
for (i = start_time; i < xl; i++) {
/* indices for period i */
i0 = i - start_time + 1;
s0 = i0 + period - 1;
/* forecast *for* period i */
xhat = level[i0 - 1] + (dotrend == 1 ? trend[i0 - 1] : 0);
stmp = doseasonal == 1 ? season[s0 - period] : (seasonal != 1);
if (seasonal == 1)
xhat += stmp;
else
xhat *= stmp;
switch(model->super.optflag[3]){
case 0: smape += SMAPE(xhat,model->data[i]);
break;
case 1: smape+= SSE(xhat,model->data[i]);
break;
case 2: smape+= ABS(xhat,model->data[i]);
break;
}
model->expectedValueXSquared+=pow((xhat-model->data[i]),2);
model->expectedValueX+=(abs(xhat-model->data[i]));
/* estimate of level *in* period i */
if (seasonal == 1)
level[i0] = alpha * (model->data[i] - stmp)
+ (1 - alpha) * (level[i0 - 1] + trend[i0 - 1]);
else
level[i0] = alpha * (model->data[i] / stmp)
+ (1 - alpha) * (level[i0 - 1] + trend[i0 - 1]);
/* estimate of trend *in* period i */
if (dotrend == 1)
trend[i0] = beta * (level[i0] - level[i0 - 1])
+ (1 - beta) * trend[i0 - 1];
/* estimate of seasonal component *in* period i */
if (doseasonal == 1) {
if (seasonal == 1)
season[s0] = gamma * (model->data[i] - level[i0])
+ (1 - gamma) * stmp;
else
season[s0] = gamma * (model->data[i] / level[i0])
+ (1 - gamma) * stmp;
}
}
if(grad && model->gradflag)
{
model->gradflag=0;
for(i=0;i<n;i++)
{
if(model->super.optflag[i]==0)
{
grad[i]=0;
continue;
}
x[i]+=ga;
fup=exp1Var(n, x,grad,my_func_data);
x[i]-=2*ga;
fdown=exp1Var(n, x,grad,my_func_data);
x[i]+=ga;
grad[i]=(fup-fdown)/(2*ga);
}
model->gradflag=1;
}
//model->oneStepAheadForecastVariance=(1.0/(model->obsCount-1))*(expectedValueXSquared-((1.0/model->obsCount)*pow(expectedValueX,2)));
return smape/(model->obsCount-(model->super.start_time-1));
}
