size_t SplitData(
std::vector<CellData<D,C>*>& vdata, SplitMethod sm,
size_t start, size_t end, const Position<C>& meanpos)
{
Assert(end-start > 1);
size_t mid=0;
Bounds<C> b;
for(size_t i=start;i<end;++i) b += vdata[i]->getPos();
int split = b.getSplit();
switch (sm) { // three different split methods
case MIDDLE :
{ // Middle is the average of the min and max value of x or y
double splitvalue = b.getMiddle(split);
DataCompareToValue<D,C> comp(split,splitvalue);
typename std::vector<CellData<D,C>*>::iterator middle =
std::partition(vdata.begin()+start,vdata.begin()+end,comp);
mid = middle - vdata.begin();
} break;
case MEDIAN :
{ // Median is the point which divides the group into equal numbers
DataCompare<D,C> comp(split);
mid = (start+end)/2;
typename std::vector<CellData<D,C>*>::iterator middle = vdata.begin()+mid;
std::nth_element(vdata.begin()+start,middle,vdata.begin()+end,comp);
} break;
case MEAN :
{ // Mean is the weighted average value of x or y
double splitvalue = meanpos.get(split);
DataCompareToValue<D,C> comp(split,splitvalue);
typename std::vector<CellData<D,C>*>::iterator middle =
std::partition(vdata.begin()+start,vdata.begin()+end,comp);
mid = middle - vdata.begin();
} break;
case RANDOM :
{ // Random is a random point from the first quartile to the third quartile
DataCompare<D,C> comp(split);
// The code for RANDOM is same as MEDIAN except for the next line.
// Note: The lo and hi values are slightly subtle. We want to make sure if there
// are only two values, we actually split. So if start=1, end=3, the only possible
// result should be mid=2. Otherwise, we want roughly 1/4 and 3/4 of the span.
mid = select_random(end-3*(end-start)/4,start+3*(end-start)/4);
typename std::vector<CellData<D,C>*>::iterator middle = vdata.begin()+mid;
std::nth_element(vdata.begin()+start,middle,vdata.begin()+end,comp);
} break;
default :
myerror("Invalid SplitMethod");
}
if (mid == start || mid == end) {
xdbg<<"Found mid not in middle. Probably duplicate entries.\n";
xdbg<<"start = "<<start<<std::endl;
xdbg<<"end = "<<end<<std::endl;
xdbg<<"mid = "<<mid<<std::endl;
xdbg<<"sm = "<<sm<<std::endl;
xdbg<<"b = "<<b<<std::endl;
xdbg<<"split = "<<split<<std::endl;
for(size_t i=start; i!=end; ++i) {
xdbg<<"v["<<i<<"] = "<<vdata[i]<<std::endl;
}
// With duplicate entries, can get mid == start or mid == end.
// This should only happen if all entries in this set are equal.
// So it should be safe to just take the mid = (start + end)/2.
// But just to be safe, re-call this function with sm = MEDIAN to
// make sure.
Assert(sm != MEDIAN);
return SplitData(vdata,MEDIAN,start,end,meanpos);
}
Assert(mid > start);
Assert(mid < end);
return mid;
}
