float correlation_coeff(const R &user1, const R &user2)
{
float numer = 0;
float denom = 0;
float user1r_sq_sum = 0;
float user2r_sq_sum = 0;
#if defined(ALG_REF_IMPL)
//for each product
for (int i = 0; i < user1.size(); ++i)
{
float user1r = user1[i] - mean(user1);
float user2r = user2[i] - mean(user2);
numer += user1r * user2r;
user1r_sq_sum += user1r * user1r;
user2r_sq_sum += user2r * user2r;
}
#else //ALG_ITPP_IMPL
R user1r = user1 - mean(user1);
R user2r = user2 - mean(user2);
//element-wise multiplication of user1r and user2r followed by summation of resultant elements
numer = elem_mult_sum(user1r, user2r);
//element-wise square of user1r followed by summation of resultant elements
user1r_sq_sum = sum_sqr(user1r);
//element-wise square of user2r followed by summation of resultant elements
user2r_sq_sum = sum_sqr(user2r);
#endif
denom = std::sqrt(user1r_sq_sum) * std::sqrt(user2r_sq_sum);
//denom = std::sqrt(user1r_sq_sum * user2r_sq_sum); //in Recommender Systems Handbook
return numer/denom;
}
void Worker::cart_product(const Vc& in, Vc& result, const int circ) {
/* I have an array of numbers from 0 to some radius saved in "in"
* At first I need the cartesian product of these values
* for Globals::dimm2 = Globals::dim - 2 instances
*/
/* for (unsigned int i=0; i<result.size(); i++) {
result[i]=0.0;
}*/
int idx;
//the indices
//squares_sum
ANNcoord ss;
//Start all of the iterators at the beginning
for (int i=0; i<Globals::dimm2; i++) {
vd1[i].begin=in.begin();
vd1[i].end=in.end();
vd1[i].me=in.begin();
}
while(1) {
// Increment the rightmost one, and repeat.
// When you reach the end, reset that one to the beginning and
// increment the next-to-last one. You can get the "next-to-last"
// iterator by pulling it out of the neighboring element in your
// vector of iterators.
idx = 0;
for(Vd::iterator it = vd1.begin(); ; ) {
// okay, I started at the left instead. sue me
++(it->me);
if (it->me == it->end) {
//not needed all the time obviously
if(it + 1 == vd1.end()) {
// I'm the last digit, and I'm about to roll
return;
} else {
// cascade
it->me = it->begin;
result[idx] = *(it->me);
++it;
}
} else {
// normal
result[idx] = *(it->me);
break;
}
++idx;
}
//if the current cartesian product is valid for a new sphere of radius circ:
ss=sum_sqr(Globals::dimm2, result);
if (ss <= radiuses_sqr[circ]) {
raster_circle_wrapper(result, circ, ss);
//depth_all_reprs(Globals::qbs[id]);
}
cart_prod_counter++;
if (cart_prod_counter >= (int) in.size()) {
ov_nodes=1;
ov_leaves=0;
rec_count(Globals::qbs[id]);
// std::cout << "in here"<< std::endl;
if (ov_nodes+ov_leaves > pow(2,20)) {
//find_all_reprs(Globals::qbs[id], &Globals::hypercube_center, 1.0, kdtree, asdf, id);
depth_all_reprs(Globals::qbs[id]);
// std::cout << "find_all_reprs" << std::endl;
}
cart_prod_counter=0;
}
}
}
/**
* calc distance between two items.
* Let a be all the users rated item 1
* Let b be all the users rated item 2
*
* 3) Using Pearson correlation
* Dist_12 = (a - mean)*(b- mean)' / (std(a)*std(b))
*
* 4) Using cosine similarity:
* Dist_12 = (a*b) / sqrt(sum_sqr(a)) * sqrt(sum_sqr(b)))
*
* 5) Using chebychev:
* Dist_12 = max(abs(a-b))
*
* 6) Using manhatten distance:
* Dist_12 = sum(abs(a-b))
*
* 7) Using tanimoto:
* Dist_12 = 1.0 - [(a*b) / (sum_sqr(a) + sum_sqr(b) - a*b)]
*
* 8) Using log likelihood similarity
* Dist_12 = 1.0 - 1.0/(1.0 + loglikelihood)
*
* 9) Using slope one:
* Dist_12 = sum_(u in intersection (a,b) (r_u1-ru2 ) / size(intersection(a,b)))
*/
double calc_distance(CE_Graph_vertex<uint32_t, float> &v, vid_t pivot, int distance_metric) {
//assert(is_pivot(pivot));
//assert(is_item(pivot) && is_item(v.id()));
dense_adj &pivot_edges = adjs[pivot - pivot_st];
int num_edges = v.num_edges();
//if there are not enough neighboring user nodes to those two items there is no need
//to actually count the intersection
if (num_edges < min_allowed_intersection || nnz(pivot_edges.edges) < min_allowed_intersection)
return 0;
dense_adj item_edges;
for(int i=0; i < num_edges; i++)
set_new(item_edges.edges, v.edge(i)->vertexid, v.edge(i)->get_data());
double intersection_size = item_edges.intersect(pivot_edges);
//not enough user nodes rated both items, so the pairs of items are not compared.
if (intersection_size < (double)min_allowed_intersection)
return 0;
if (distance_metric == PEARSON){
if (debug){
std::cout<< pivot -M+1<<" Pivot edges: " <<pivot_edges.edges << std::endl;
std::cout<< "Minusmean: " << minus(pivot_edges.edges,mean) << std::endl;
std::cout<< v.id() -M+1<<"Item edges: " <<item_edges.edges << std::endl;
std::cout<< "Minusmean: " << minus(item_edges.edges, mean) << std::endl;
}
double dist = minus(pivot_edges.edges, mean).dot(minus(item_edges.edges, mean));
if (debug)
std::cout<<"dist " << pivot-M+1 << ":" << v.id()-M+1 << " " << dist << std::endl;
return dist / (stddev[pivot-M] * stddev[v.id()-M]);
}
else if (distance_metric == TANIMOTO){
return calc_tanimoto_distance(pivot_edges.edges,
item_edges.edges,
sum_sqr(pivot_edges.edges),
sum_sqr(item_edges.edges));
}
else if (distance_metric == CHEBYCHEV){
return calc_chebychev_distance(pivot_edges.edges,
item_edges.edges);
}
else if (distance_metric == LOG_LIKELIHOOD){
return calc_loglikelihood_distance(pivot_edges.edges,
item_edges.edges,
sum_sqr(pivot_edges.edges),
sum_sqr(item_edges.edges));
}
else if (distance_metric == COSINE){
return calc_cosine_distance(pivot_edges.edges,
item_edges.edges,
sum_sqr(pivot_edges.edges),
sum_sqr(item_edges.edges));
}
else if (distance_metric ==MANHATTEN){
return calc_manhatten_distance(pivot_edges.edges,
item_edges.edges);
}
else if (distance_metric == SLOPE_ONE){
return calc_slope_one_distance(pivot_edges.edges, item_edges.edges) / intersection_size;
}
return NAN;
}
