// Find adjacent intervals with smallest $\chi^2$
IntervalList::iterator find_min_chi_chi()
{
// Compute $\chi^2$ value for each adjacent interval,
// keeping track of minimum value
IntervalList::iterator lit = g_intervals.begin(),
lend = g_intervals.end();
IntervalList::iterator next = lit, min_lit = lend;
float min_chisquared = 1e6;
bool first = true;
for ( ; lit != lend; ++lit) {
next = lit;
++next;
if (next == lend) break;
float chisquared = compute_chisquared(*lit, *next);
// Debug
print_interval_set(*lit);
cout << "\\chi^2 = " << chisquared << endl;
if (first || chisquared < min_chisquared) {
min_lit = lit;
min_chisquared = chisquared;
}
first = false;
}
cout << "min_chisquared = " << min_chisquared << endl;
return min_lit;
}
void print_interval_summary(ostream& os, int dimIndex)
{
IntervalList::iterator lit = g_intervals.begin(),
lend = g_intervals.end();
IntervalList::iterator next = lit;
vector<DataType> split_points;
os << "\n" "Feature " << (dimIndex+1) << ":" << endl;
os << "Ranges: ";
for ( ; lit != lend; ++lit) {
next = lit;
++next;
pair<DataType, DataType> range, next_range;
range = get_range(*lit, dimIndex);
if (next != lend) {
next_range = get_range(*next, dimIndex);
float average = (range.second + next_range.first) * 0.5f;
split_points.push_back(average);
}
os << "[" << range.first << ", " << range.second << "] ";
}
os << endl;
os << "Split points: ";
copy(split_points.begin(), split_points.end(),
ostream_iterator<DataType>(os, ", "));
os << endl;
}
// Debugging -- print all intervals
void print_all_intervals()
{
IntervalList::iterator lit = g_intervals.begin(),
lend = g_intervals.end();
cout << "[intervals]" << endl;
for ( ; lit != lend; ++lit) {
IntervalSet& indices = *lit;
print_interval_set(indices);
}
}
// $\chi^2$ just for one dimension
void chi_chi_dim_analysis(int dimIndex)
{
// Sort and initialize one interval per unique attribute value
sort(g_data.begin(), g_data.end(), tuple_less_than<Tuple>(dimIndex));
TupleVec::iterator tit = g_data.begin(), tend = g_data.end();
cout << "[sort]" << endl;
int index = 0;
for ( ; tit != tend; ++tit, ++index) {
cout << index << ":";
copy(tit->first.begin(), tit->first.end(), ostream_iterator<float>(cout, ", "));
cout << tit->second << endl;
}
initialize_intervals(dimIndex);
// Count instances of all classes
count_classes();
while ((int)g_intervals.size() > g_max_intervals) {
// Find adjacent intervals with smallest $\chi^2$
IntervalList::iterator min_lit = find_min_chi_chi();
assert(min_lit != g_intervals.end());
IntervalList::iterator min_lit_next = min_lit;
++min_lit_next;
cout << "[before merge] ";
print_all_intervals();
// Merge
IntervalSet& interval_1 = *min_lit;
IntervalSet& interval_2 = *min_lit_next;
interval_1.insert(interval_2.begin(), interval_2.end());
g_intervals.erase(min_lit_next);
cout << "[after merge] ";
print_all_intervals();
}
// Debugging
print_interval_summary(cout, dimIndex);
// Logged output
print_interval_summary(olog, dimIndex);
}
