double MultiClassExp::getConfidence(const FeatureVec& feature_vec, const std::vector<double>& weights_plus, const std::vector<double>& weights_minus)
{
double confidence_plus = std::inner_product(feature_vec.begin(), feature_vec.end(),
weights_plus.begin(), 0.0);
double confidence_minus = std::inner_product(feature_vec.begin(), feature_vec.end(),
weights_minus.begin(), 0.0);
double confidence = confidence_plus-confidence_minus;
return confidence;
}
/*the document is index:value format because there is less features,centorid is array format for quick access
*now it's changed into cosine!2011-07-22
* */
float KNN::InnerProduct(FeatureVec & document,vector<float> & centorid, int catIndex)
{
/*part of cosine*/
float similarity=0.0;
float finalScore=0.0;
float totalDoc=0.0;
FeatVecIte it;
for(it=document.begin();it!=document.end();it++)
{
similarity+=((*it).weight*centorid[(*it).index]);//direct access the vector
}
//float docLen=inner_product(centorid.begin(),centorid.end(),centorid.begin(),0.0);//tip:可以提前计算好。@zhqm
return similarity;
}
int KNN::MaxProbCat(FeatureVec & document,vector<vector<float> > & Centoroid,float & score)
{
float MaxScoreNow=0.0;
int MaxCatIndex=-1;
int currentCatIndex=0;
float docDotProduct=0.0;
FeatVecIte dit;
for(dit=document.begin();dit!=document.end();dit++)
{
docDotProduct+=(*dit).weight*(*dit).weight; //tip:@zhqm totalDoc +=
}
docDotProduct = sqrt(docDotProduct);
if (docDotProduct == 0){
score = 0.0;
return MaxCatIndex;
}
vector<vector<float> >::iterator it;
for(it=Centoroid.begin();it!=Centoroid.end();it++,currentCatIndex++)
{
float score=InnerProduct(document,(*it), currentCatIndex);
score=(float)score/(docDotProduct * m_centoroidFeatureDotProduct.at(currentCatIndex));
//cout<<"cat="<<currentCatIndex<<"\tscore="<<score<<endl;
if(score>MaxScoreNow)//a better class
{
MaxScoreNow=score;
MaxCatIndex=currentCatIndex;
}
}
score=MaxScoreNow;
return MaxCatIndex;
}