void ConvNet::learn(dmatrix4 &stimulus, dmatrix2 &target, int lessons)
{
assert(stimulus.size()==target.size());
if(lessons==0) lessons=stimulus.size();
for(int n=0;n<lessons;n++) {
if (n%10 == 0) std::cout << "Learning lesson #" << n+1 << std::endl;
learn(stimulus[n], target[n]);
}
}
real ConvNet::validation(dmatrix4 &lessons, dmatrix2 &targets)
{
assert(lessons.size()==targets.size());
double rounds = (double)lessons.size();
int result; // Holds the networks' prediction of the category
int answer; // Holds the actual number of the right category
int right = 0; // Holds the number of right answers
for(int n=0;n<rounds;n++) {
result = predict(lessons[n]);
answer = *std::find(targets[n].begin(),
targets[n].end(), 1.0); //index of 1.0 in targets
if (result==answer) right++;
}
return ((double)right/(double)lessons.size());
}
ThoughtBubble PoolLayer::max_pool(dmatrix2 slice) const
{
dvec fslice(4);
ivec bslice(4, 0);
double maxi;
ThoughtBubble output;
maxi = *std::max_element(slice.point(), slice.point()+slice.size());
bslice[*std::find(slice.point(), slice.point()+slice.size(), maxi)] = 1;
output.excitation = bslice;
output.activation = maxi;
return output;
}
