/**
* \brief If the distance is smaller then update
*/
void apply(icontext_type& context, vertex_type& vertex,
const graphlab::empty& empty) {
changed = false;
if(vertex.data().dist > min_dist) {
changed = true;
vertex.data().dist = min_dist;
}
}
// Change the vertex data if any of its neighbors have a lower data value.
void apply(icontext_type& context, vertex_type& vertex, const gather_type& total) {
// mark if values differ to determine which edges to scatter on.
if (message_value < vertex.data()) {
changed = true;
vertex.data() = message_value;
} else {
changed = false;
}
}
void apply(icontext_type& context, vertex_type& vertex,
const gather_type& total)
{
changed = false;
if (vertex.data().color > total.color) {
changed = true;
vertex.data().color = total.color;
}
}
/**
* \brief The scatter function just signal adjacent pages
*/
void scatter(icontext_type& context, const vertex_type& vertex,
edge_type& edge) const {
const vertex_type other = get_other_vertex(edge, vertex);
distance_type newd = vertex.data().dist + edge.data().dist;
if (other.data().dist > newd) {
const min_distance_type msg(newd);
context.signal(other, msg);
}
} // end of scatter
void apply(icontext_type& context, vertex_type& vertex, const gather_type &total) {
vertex_data_type new_label = most_common(total);
if (new_label != vertex.data()) {
vertex.data() = new_label;
changed = true;
} else {
changed = false;
}
}
// Scatter to scatter_edges edges with the new message value.
void scatter(icontext_type& context, const vertex_type& vertex, edge_type& edge) const {
bool isEdgeSource = (vertex.id() == edge.source().id());
bool hasSameData = isEdgeSource ? (vertex.data() == edge.target().data()) : (vertex.data() == edge.source().data()) ;
if (!hasSameData) {
min_combiner combiner;
combiner.value = message_value;
context.signal(isEdgeSource ? edge.target() : edge.source(), combiner);
}
}
void apply(icontext_type& context, vertex_type& vertex,
const gather_type& total)
{
converged = true;
double new_pagerank = 0.15 + 0.85 * total.pagerank;
double delta = fabs(vertex.data().pagerank - new_pagerank);
vertex.data().pagerank = new_pagerank;
if (delta > EPS) {
converged = false;
}
}
void apply(icontext_type& context, vertex_type& vertex, const gather_type& total)
{
cout << "apply(), vid=" << vertex.id() << endl;
cout << "total=" << total << endl;
// reset incoming messages
vertex.data().multiplied_incoming_messages.clear();
// iterate over message targets
for (set<vertex_id_type>::const_iterator target_it=total.message_targets.begin(); target_it!=total.message_targets.end(); ++target_it)
{
vertex_id_type target_id=*target_it;
// iterate over message sources (and the betas)
for (map<vertex_id_type, VectorXd>::const_iterator source_it=total.message_source_betas.begin(); source_it!=total.message_source_betas.end(); ++source_it)
{
vertex_id_type source_id=source_it->first;
// we dont need to consider the kernel matrix of a edge with itself since this is always omitted in the message scheduling
if (source_id==target_id)
continue;
cout << "adding message from " << source_id << " to " << vertex.id() << " to construct message from " << vertex.id() << " to " << target_id << endl;
// extract K and beta for incoming message
MatrixXd K=vertex.data().kernel_dict[pair<vertex_id_type, vertex_id_type>(target_id,source_id)];
VectorXd beta=source_it->second;
// if beta has zero rows, it is initialised to constant with unit norm (first iteration)
if (!beta.rows())
{
beta=VectorXd::Constant(K.cols(), 1.0);
beta=beta/beta.norm();
}
cout << "K_" << vertex.id() << "^(" << target_id << "," << source_id << "):" << endl << K << endl;
cout << "times" << endl;
cout << "beta_(" << source_id << "," << vertex.id() << "):" << endl << beta << endl;
// for a fixed source and target node, compute incoming kernelbp message
VectorXd message=K*beta;
cout << "message: " << message << endl;
// multiply all messages together
if (vertex.data().multiplied_incoming_messages.find(target_id)==vertex.data().multiplied_incoming_messages.end())
vertex.data().multiplied_incoming_messages[target_id]=message;
else
{
cout << "old message product: " << vertex.data().multiplied_incoming_messages[target_id] << endl;
vertex.data().multiplied_incoming_messages[target_id]=vertex.data().multiplied_incoming_messages[target_id].cwiseProduct(message);
}
cout << "new message product: " << vertex.data().multiplied_incoming_messages[target_id] << endl;
}
}
}
void apply(icontext_type& context, vertex_type& vertex, const gather_type &total) {
if (context.iteration() == 0) {
vertex.data().neighbors = total.get();
} else {
size_t d = vertex.data().neighbors.size();
size_t t = last_msg;
// Due to rounding errors, the results is sometimes not exactly
// 0.0 even when it should be. Explicitly set LCC to 0 if that
// is the case, other calculate it as tri / (degree * (degree - 1))
double lcc = (d < 2 || t == 0) ? 0.0 : double(t) / (d * (d - 1));
vertex.data().clustering_coef = lcc;
}
}
/* Use the total rank of adjacent pages to update this page */
void apply(icontext_type& context, vertex_type& vertex,
const double& total) {
//printf("Entered apply on node %d value %lg\n", vertex.id(), total);
vertex_data & user = vertex.data();
assert(mi.x_offset >=0 || mi.y_offset >= 0);
assert(mi.r_offset >=0);
/* perform orthogonalization of current vector */
if (mi.orthogonalization){
for (int i=mi.mat_offset; i< mi.vec_offset; i++){
vertex.data().pvec[mi.vec_offset] -= alphas.pvec[i-mi.mat_offset] * vertex.data().pvec[i];
}
return;
}
double val = total;
//assert(total != 0 || mi.y_offset >= 0);
//store previous value for convergence detection
if (mi.prev_offset >= 0)
user.pvec[mi.prev_offset ] = user.pvec[mi.r_offset];
assert(mi.x_offset >=0 || mi.y_offset>=0);
if (mi.A_offset && mi.x_offset >= 0){
if (info.is_square() && mi.use_diag)// add the diagonal term
val += (/*mi.c**/ (user.A_ii+ regularization) * user.pvec[mi.x_offset]);
//printf("node %d added diag term: %lg\n", vertex.id(), user.A_ii);
val *= mi.c;
}
/***** COMPUTE r = c*I*x *****/
else if (!mi.A_offset && mi.x_offset >= 0){
val = mi.c*user.pvec[mi.x_offset];
}
/**** COMPUTE r+= d*y (optional) ***/
if (mi.y_offset>= 0){
val += mi.d*user.pvec[mi.y_offset];
}
/***** compute r = (... ) / div */
if (mi.div_offset >= 0){
val /= user.pvec[mi.div_offset];
}
user.pvec[mi.r_offset] = val;
//printf("Exit apply on node %d value %lg\n", vertex.id(), val);
}
/** The gather function computes XtX and Xy */
gather_type gather(icontext_type& context, const vertex_type& vertex,
edge_type& edge) const {
if(edge.data().role == edge_data::TRAIN) {
const vertex_type other_vertex = get_other_vertex(edge, vertex);
return gather_type(other_vertex.data().factor, edge.data().obs);
} else return gather_type();
} // end of gather function
void aggregate(icontext_type& context, const vertex_type& vertex){
if (!marked) {
marked = true;
saedb::IAggregator* active_node = context.getAggregator("active_node");
float t = vertex.data();
active_node->reduce(&t);
}
}
edge_dir_type scatter_edges(icontext_type& context, const vertex_type& vertex) const {
if( vertex.data().changed == 1 ) {
return OUT_EDGES;
}
else {
return NO_EDGES;
}
}
void scatter(icontext_type& context, const vertex_type& vertex,
edge_type& edge) const {
const vertex_type other = edge.target();
pagerank_type value = vertex.data().pagerank / vertex.num_out_edges();
assert(other.id() != vertex.id());
const sum_pagerank_type msg(value);
context.signal(other, msg);
}
void scatter(icontext_type& context, const vertex_type& vertex,
edge_type& edge) const
{
const vertex_type other = edge.target();
distance_type newd = vertex.data().dist + edge.data().dist;
const min_distance_type msg(newd);
context.signal(other, msg);
}
/**
* \brief Synchronizes all copies of this vertex
*
* If the current vertex value has changed, copy the vertex value to
* all mirrors. This is for advanced use!
* Under most circumstances you should not need to use
* this function directly.
*/
void synchronize() {
if (vtx_set && graph.l_is_master(vtx.local_id())) {
std::string new_value = serialize_to_string(vtx.data());
if (original_value != new_value) {
// synchronize this vertex's value
engine.synchronize_one_vertex_wait(vtx);
}
std::swap(original_value, new_value);
}
}
/**
* \brief If the distance is smaller then update
*/
void apply(icontext_type& context, vertex_type& vertex,
const gather_type& total) {
changed = false;
if(context.iteration() == 0) {
changed = true;
vertex.data().dist = 0;
context.setUpdateFlag(changed);
return;
//lastchange = vertex.data().dist;
}
if(vertex.data().dist > total.dist) {
changed = true;
vertex.data().dist = total.dist;
//lastchange = vertex.data().dist;
}
context.setUpdateFlag(changed);
//std::cout << "vid=" << vertex.id() << ", val=" << vertex.data().dist << "\n";
}
void apply(icontext_type& context, vertex_type& vertex,
const factor_type& sum) {
const size_t num_neighbors = vertex.num_in_edges() + vertex.num_out_edges();
ASSERT_GT(num_neighbors, 0);
// There should be no new edge data since the vertex program has been cleared
vertex_data& vdata = vertex.data();
ASSERT_EQ(sum.size(), NTOPICS);
ASSERT_EQ(vdata.factor.size(), NTOPICS);
vdata.nupdates++; vdata.nchanges = 0;
vdata.factor = sum;
} // end of apply
vertex_type operator * (vertex_type const& vex) const
{
vertex_type ret;
typedef Eigen::Matrix<value_type, 4, 1> impl_vex_type;
impl_vex_type* impl_vex = (impl_vex_type*)vex.data();
impl_vex_type res = _mat * *impl_vex;
ret.x = res[0];
ret.y = res[1];
ret.z = res[2];
ret.w = res[3];
return ret;
}
pair<size_t, size_t> count_triangles(const vertex_type& a, const vertex_type& b, const bool inverted=false) const {
typedef neighbors_type::const_iterator neighbors_iterator_type;
const vertex_id_type a_id = a.id();
const vertex_id_type b_id = b.id();
const neighbors_type& a_adj = a.data().neighbors;
const neighbors_type& b_adj = b.data().neighbors;
const bool directed = global_directed;
if (a_adj.at(b_id) > 1 && a.id() < b.id() && !inverted) {
return make_pair(0, 0);
}
if (a_adj.size() > b_adj.size()) {
return reverse(count_triangles(b, a, true));
}
size_t a_count = 0;
size_t b_count = 0;
for (neighbors_iterator_type it_a = a_adj.begin(); it_a != a_adj.end(); it_a++) {
const vertex_id_type c_id = it_a->first;
neighbors_iterator_type it_b = b_adj.find(it_a->first);
if (it_b != b_adj.end()) {
if (b_id < c_id) {
a_count += directed ? it_b->second : 2;
}
if (a_id < c_id) {
b_count += directed ? it_a->second : 2;
}
}
}
return make_pair(a_count, b_count);
}
void apply(icontext_type& context, vertex_type& vertex,
const graphlab::empty& empty) {
if (context.iteration() < ROUND)
{
context.signal(vertex);
}
if(context.iteration() == 0)
{
return;
}
pagerank_type new_pagerank = 0.15 + 0.85 * sum_pagerank;
vertex.data().pagerank = new_pagerank;
}
/** apply collects the sum of XtX and Xy */
void apply(icontext_type& context, vertex_type& vertex,
const gather_type& sum) {
// Get and reset the vertex data
vertex_data& vdata = vertex.data();
// Determine the number of neighbors. Each vertex has only in or
// out edges depending on which side of the graph it is located
if(sum.Xy.size() == 0) { vdata.residual = 0; ++vdata.nupdates; return; }
mat_type XtX = sum.XtX;
vec_type Xy = sum.Xy;
// Add regularization
for(int i = 0; i < XtX.rows(); ++i) XtX(i,i) += LAMBDA; // /nneighbors;
// Solve the least squares problem using eigen ----------------------------
const vec_type old_factor = vdata.factor;
vdata.factor = XtX.selfadjointView<Eigen::Upper>().ldlt().solve(Xy);
// Compute the residual change in the factor factor -----------------------
vdata.residual = (vdata.factor - old_factor).cwiseAbs().sum() / XtX.rows();
++vdata.nupdates;
} // end of apply
void init(icontext_type& context, vertex_type& vertex,
const message_type& msg) {
vertex.data() = msg;
}
void scatter(icontext_type& context, const vertex_type& vertex, edge_type& edge) const
{
if (vertex.data() + edge.data() < edge.target().data())
context.signal(edge.target());
}
void apply(icontext_type& context, vertex_type& vertex, const gather_type& total)
{
vertex.data() = min(vertex.data(), total.dis);
}
void apply(icontext_type& context, vertex_type& vertex,
const gather_type& total) {
const double newval = total + RESET_PROB;
vertex.data() = newval;
}
void scatter(icontext_type& context, const vertex_type& vertex, edge_type& edge) const
{
cout << "scatter(), edge=" << edge.source().id() << "->" << edge.target().id() << ", called from vid=" << vertex.id() << endl;
cout << "computing message from vid=" << vertex.id() << " to vid=" << edge.source().id() << endl;
vertex_id_type message_target=edge.source().id();
// find out whether full rank or incomplete Cholesky mode
// distinguish case this node being observed or not
VectorXd new_beta;
if (edge.target().data().is_observed)
{
cout << "observed target" << endl;
// extract system solutions and observation kernel vector, base on full rank or incomplete Cholesky
if (edge.data().full_rank)
{
cout << "full rank case" << endl;
MatrixXd L_s=edge.data().solution_matrices["L_s"];
cout << "L_s:" << L_s << endl;
MatrixXd L_t=edge.data().solution_matrices["L_t"];
cout << "L_t:" << L_t << endl;
VectorXd k=vertex.data().kernel_dict_obs.at(message_target);
cout << "k:" << k << endl;
// L_{s}^{-T}(L_{s}^{-1}(L_{t}^{-T}(L_{t}^{-1}k_{t}^{s}), from right to left, 4 solver calls
new_beta=k;
new_beta=L_t.triangularView<Lower>().solve(new_beta);
new_beta=L_t.transpose().triangularView<Upper>().solve(new_beta);
new_beta=L_s.triangularView<Lower>().solve(new_beta);
new_beta=L_s.transpose().triangularView<Upper>().solve(new_beta);
}
else
{
cout << "incomplete Cholesky case" << endl;
MatrixXd Q_s=edge.data().solution_matrices["Q_s"];
cout << "Q_s:" << Q_s << endl;
MatrixXd R_s=edge.data().solution_matrices["R_s"];
cout << "R_s:" << R_s << endl;
MatrixXd P_s=edge.data().solution_matrices["P_s"];
cout << "P_s:" << P_s << endl;
MatrixXd Q_t=edge.data().solution_matrices["Q_t"];
cout << "Q_t:" << Q_t << endl;
MatrixXd R_t=edge.data().solution_matrices["R_t"];
cout << "R_t:" << R_t << endl;
MatrixXd P_t=edge.data().solution_matrices["P_t"];
cout << "P_t:" << P_t << endl;
MatrixXd W=edge.data().solution_matrices["W"];
cout << "W:" << W << endl;
VectorXd k=vertex.data().kernel_dict_obs.at(message_target);
cout << "k:" << k << endl;
// R_{s}^{-1}(Q_{s}^{T}((P_{s}(W_{s}W_{t}^{T}))(R_{t}^{-1}(Q_{t}^{T}(P_{t}k_{\mathcal{I}_{t}}^{(s)})))
new_beta=k;
new_beta=P_t.transpose()*new_beta;
new_beta=Q_t.transpose()*new_beta;
new_beta=R_t.triangularView<Upper>().solve(new_beta);
new_beta=W*new_beta;
new_beta=P_s.transpose()*new_beta;
new_beta=Q_s.transpose()*new_beta;
new_beta=R_s.triangularView<Upper>().solve(new_beta);
}
}
else
{
cout << "non-observed target" << endl;
cout << "multiplied_incoming_messages: " << vertex.data().multiplied_incoming_messages << endl;
// extract system solutions, depending on full rank or incomplete Cholesky
if (edge.data().full_rank)
{
cout << "full rank case" << endl;
MatrixXd L_s=edge.data().solution_matrices["L_s"];
cout << "L_s:" << L_s << endl;
VectorXd k;
if (!vertex.data().multiplied_incoming_messages.size())
{
cout << "no incoming messages, using constant unit norm vector" << endl;
k=VectorXd::Constant(L_s.cols(), 1.0/sqrt(L_s.cols()));
}
else
{
k=vertex.data().multiplied_incoming_messages.at(message_target);
}
cout << "k:" << k << endl;
// (K_{s}+\lambda I){}^{-1}k_{ut}^{(s)}=L_{s}^{-T}(L_{s}^{-1}k_{ut}^{(s)}) from right to left, 2 solver calls
new_beta=k;
new_beta=L_s.triangularView<Lower>().solve(new_beta);
new_beta=L_s.transpose().triangularView<Upper>().solve(new_beta);
}
else
{
cout << "incomplete Cholesky case" << endl;
MatrixXd Q_s=edge.data().solution_matrices["Q_s"];
cout << "Q_s:" << Q_s << endl;
MatrixXd R_s=edge.data().solution_matrices["R_s"];
cout << "R_s:" << R_s << endl;
MatrixXd P_s=edge.data().solution_matrices["P_s"];
cout << "P_s:" << P_s << endl;
MatrixXd W=edge.data().solution_matrices["W"];
cout << "W:" << W << endl;
VectorXd k;
if (!vertex.data().multiplied_incoming_messages.size())
{
cout << "no incoming messages, using constant unit norm vector" << endl;
k=VectorXd::Constant(W.cols(), 1.0/sqrt(W.cols()));
}
else
{
k=vertex.data().multiplied_incoming_messages.at(message_target);
}
cout << "k:" << k << endl;
// R_{s}^{-1}(Q_{s}^{T}(P_{s}^{T}k_{t}^{(s)}))
new_beta=k;
new_beta=W*new_beta;
new_beta=P_s.transpose()*new_beta;
new_beta=Q_s.transpose()*new_beta;
new_beta=R_s.triangularView<Upper>().solve(new_beta);
}
}
// normalise
new_beta=new_beta/new_beta.norm();
// check whether has changed or not yet existed
double difference;
if (!edge.data().beta.rows())
difference=numeric_limits<double>::infinity();
else
difference=(new_beta-edge.data().beta).norm();
cout << "beta norm difference is " << difference << endl;
if (difference>BETA_EPSILON)
{
// store new message and signal depending node if beta has changed or has not yet existed
edge.data().beta=new_beta;
context.signal(edge.source());
cout << "beta has changed, new_beta=" << new_beta << "\nhas norm=" << new_beta.norm() << ", signalling vid=" << edge.source().id() << endl;
}
else
{
cout << "converged!\n";
}
cout << "beta: " << edge.source().id() << "->" << edge.target().id() << ": " << edge.data().beta << endl;
}
void apply(icontext_type& context, vertex_type& vertex, const gather_type& total) {
if( total < KCORE_DEGREE ) {
vertex.data().in_core = false;
vertex.data().changed += 1;
}
}
/**
* \internal
* \brief Flags that this vertex was synchronized.
*/
void set_synchronized() {
if (vtx_set && graph.l_is_master(vtx.local_id())) {
original_value = serialize_to_string(vtx.data());
}
}
void apply(icontext_type& context, vertex_type& vertex, const gather_type& total){
vertex.data().actived = false;
}