# K-clustering algorithm using Kruskal MST with Disjoint Set in place to check for cycles

here below a working implementation that finds the minimal distance between k(set =4 below) clusters in a graph. I have doubts mainly on the implementation of the Disjoint Set structure: instead of using STL containers, I went along with an array of pointers as data structure hosting the leader nodes, in order to have some practice with C-style pointers.

Here is the Graph structure: Graph.h

// creates and manages a graph data structure
#ifndef GRAPH_H_INCLUDED
#define GRAPH_H_INCLUDED

#include <array>
#include <vector>
#include <list>
#include <fstream>
#include <string>
#include <iostream>

class Edge
{
public:
// constructors
Edge();
Edge(int, int, int);

std::array<int, 2> getNodes() const;
void setNodes(const int, const int);
int getCost() const;
void setCost(const int);
void swapNodes();
// a get() that allows writing:
int operator[](const int);

bool operator==(const Edge&) const;
bool operator!=(const Edge&) const;

private:
int node1, node2;       // nodes are just indices of the nodes in the graph
int cost;
};

class Node
{
friend class Graph;       // friendship needed by printing routine in Graph
public:
// constructors
Node();
Node(const Edge&);

int getLabel() const {return label;}
Edge getEdge(const int) const;
void setLabel(const int in) {label = in;}
void addEdge(const Edge in) {edges.push_back(in);}
int getScore() const {return score;}
void setScore(const int in) {score = in;}

std::list<Edge>::size_type size() const {return edges.size();}  // size of list 'edges'

// iterators
typedef std::list<Edge>::iterator iterator;
typedef std::list<Edge>::const_iterator const_iterator;
std::list<Edge>::iterator begin() {return edges.begin();}
std::list<Edge>::iterator end() {return edges.end();}
std::list<Edge>::const_iterator cbegin() const {return edges.begin();}
std::list<Edge>::const_iterator begin() const {return edges.begin();}
std::list<Edge>::const_iterator cend() const {return edges.end();}
std::list<Edge>::const_iterator end() const {return edges.end();}

// inserts group of edges to the end of a edges vector
std::list<Edge>::iterator insertEdges(const std::list<Edge>::iterator beg_in,
const std::list<Edge>::iterator end_in)
{
return edges.insert(end(), beg_in, end_in);
}
// erase node
std::list<Edge>::iterator erase(int);

bool operator==(const Node&) const;
bool operator!=(const Node&) const;

private:
int label;
std::list<Edge> edges;
int score;          // new, starts at 10000000, is equal to lowest cost Edge in 'edges'
};

class Graph
{
public:
Graph();
// constructor from txt file
Graph(const std::string&);

Node getNode(const int index) const {return nodes[index];}
void addNode(const Node in) {nodes.push_back(in);}

std::vector<Node>::size_type size() {return nodes.size();}  // size of vector 'nodes'
std::vector<Node>::size_type size() const {return nodes.size();}  // size of vector 'nodes'
void output() const;      // prints graph
void output(const int) const;

// iterators
typedef std::vector<Node>::iterator iterator;
typedef std::vector<Node>::const_iterator const_iterator;
std::vector<Node>::iterator begin() {return nodes.begin();}
std::vector<Node>::iterator end() {return nodes.end();}
std::vector<Node>::const_iterator begin() const {return nodes.begin();}
std::vector<Node>::const_iterator end() const {return nodes.end();}

Node& operator[](const int index)
{
return nodes[index];
}

std::vector<Node>::iterator erase(const int index)
{
return nodes.erase(nodes.begin() + index);
}

private:
std::vector<Node> nodes;
};

bool compareCosts(const Edge&, const Edge&);
bool compareScores(const Node&, const Node&);
bool compareLabels(const Node&, const Node&);

#endif // GRAPH_H_INCLUDED


Graph.cpp:

#include <iostream>
#include <fstream>
#include <array>
#include <list>
#include <string>
#include <algorithm>
#include "Graph.h"

using std::array;       using std::ifstream;
using std::string;      using std::endl;
using std::cout;        using std::list;
using std::equal;

// Edge

// constructors
Edge::Edge():node1(0), node2(0), cost(0) {}
Edge::Edge(int node_1, int node_2, int len): node1(node_1), node2(node_2), cost(len) {}

array<int, 2> Edge::getNodes() const
{
array<int, 2> ret = {node1, node2};

return ret;
}

void Edge::setNodes(const int in1, const int in2)
{
node1 = in1;
node2 = in2;
}

int Edge::getCost() const
{
return cost;
}

void Edge::setCost(const int len)
{
cost = len;
}

void Edge::swapNodes()
{
node1 = node1 - node2;
node2 += node1;
node1 = node2 - node1;
}

// same as getNodes() above
int Edge::operator[](const int index)
{
if (index == 0) return node1;
else if (index == 1) return node2;
else {
try {throw;}
catch(...) {cout << "edge index must be either 0 or 1" << endl;}
return 1;
}
}

bool Edge::operator==(const Edge& rhs) const
{
if ( (node1 == rhs.getNodes()[0]) && (node2 == rhs.getNodes()[1]) && cost == rhs.getCost() )
return true;
else
return false;
}

bool Edge::operator!=(const Edge& rhs) const
{
if ( !(*this == rhs) )
return true;
else
return false;
}

// Node

//constructors
Node::Node(): label(0), edges(0), score(10000000) {}
Node::Node(const Edge& edg): label(edg.getNodes()[0]), edges(0), score(10000000)
{
edges.push_back(edg);
}

Edge Node::getEdge(const int index) const
{
Edge ret;
list<Edge>::const_iterator iter = edges.begin();

return *iter;
}

void Node::addManyEdges(const vector<Edge> input)
{
for (size_t i = 0; i != input.size(); i++)
{
edges.push_back(input[i]);
}
}

list<Edge>::iterator Node::erase(int index)
{
list<Edge>::iterator iter = edges.begin();

return edges.erase(iter);
}

bool Node::operator==(const Node& rhs) const
{
return label == rhs.getLabel() && equal( edges.begin(), edges.end(), rhs.begin() );  // no need to equate scores
}

bool Node::operator!=(const Node& rhs) const
{
return !(*this == rhs);
}

// Graph

// constructors
Graph::Graph(): nodes(0) {}
Graph::Graph(const string& file_input): nodes(0)    // constructor from file
{
string filename(file_input + ".txt");
string line;
ifstream is;
is.open(filename);

int number_nodes; //, number_edges;
is >> number_nodes; // >> number_edges;
nodes.resize(number_nodes);  // reserve the Node vector of size 'number_nodes'

int node1, node2, cost;
while (is >> node1 >> node2 >> cost)
{
int nodes_array[2] = {node1, node2};
for (int& node_i : nodes_array) {
if (nodes[node_i - 1].size() == 1)
nodes[node_i - 1].setLabel(node_i);
}

Edge current_edge(node1, node2, cost);

if (node1 != node2) {
current_edge.swapNodes();
}
}

is.close();
}

// prints all input nodes
void Graph::output() const
{
for (size_t i = 0; i != nodes.size(); ++i)
{
cout << "Node " << nodes[i].getLabel() << ", size = " << nodes[i].edges.size() << "  with edges: ";
for (size_t j = 0; j != nodes[i].edges.size(); ++j)
{
int node_left = nodes[i].getEdge(j).getNodes()[0];
int node_right = nodes[i].getEdge(j).getNodes()[1];
int cost = nodes[i].getEdge(j).getCost();
cout << "[" << node_left << "-" << node_right << ", " << cost << "]   ";
}
cout << endl;
}
}

// prints 10 neighbours around picked node
void Graph::output(const int picked_node) const
{
for (int i = picked_node - 5; i != picked_node + 5; ++i)
{
cout << "Node " << nodes[i].getLabel() << ", with edges: ";
for (size_t j = 0; j != nodes[i].edges.size(); ++j)
{
int node_left = nodes[i].getEdge(j).getNodes()[0];
int node_right = nodes[i].getEdge(j).getNodes()[1];
int cost = nodes[i].getEdge(j).getCost();
int score = nodes[node_right - 1].getScore();
cout << "[" << node_left << "-" << node_right << ", " << cost << ", " << score << "]   ";
}
cout << endl;
}
}

bool compareCosts(const Edge& a, const Edge& b)
{
return a.getCost() < b.getCost();
}

bool compareScores(const Node& a, const Node& b)
{
return a.getScore() > b.getScore();
}

bool compareLabels(const Node& a, const Node& b)
{
return a.getLabel() > b.getLabel();
}


BFS implementation for Kruskal (alternative to Disjoint Set Kruskal implementation) BreadthFirstSearch.h

#ifndef BREADTHFIRSTSEARCH_H_INCLUDED

#include <limits>
#include "Graph.h"

int const infinity = std::numeric_limits<int>::infinity();

bool breadthFirstSearch(const Graph&, const int, const int);



#include <iostream>
#include <queue>
#include <vector>
#include <map>
#include <algorithm>

using std::cout;    using std::endl;
using std::cin;     using std::find_if;
using std::vector;  using std::queue;
using std::map;

bool breadthFirstSearch(const Graph& G, const int start_node_label, const int target_node_label)
{
// define type for explored/unexplored Nodes
enum is_visited {not_visited, visited};
map<int, is_visited> node_is_visited;
for (Graph::const_iterator iter = G.begin(); iter != G.end(); ++iter)
node_is_visited[iter->getLabel()] = not_visited;

Graph::const_iterator start_node_iter =
find_if(G.begin(), G.end(), [=](const Node& i){return i.getLabel() == start_node_label;});
if ( start_node_iter == G.end() )
return false;
node_is_visited[start_node_label] = visited;

Node next_node;
next_node = *start_node_iter;

// breadth-first algorithm runs based on queue structure
queue<Node> Q;
Q.push(next_node);

// BFS algorithm
while (Q.size() != 0) {     // out of main loop if all nodes searched->means no path is present
Node current_node = Q.front();
Node linked_node;       // variable hosting node on other end
for (size_t i = 0; i != current_node.size(); ++i) {     // explore nodes linked to current_node by an edge
int linked_node_label = current_node.getEdge(i).getNodes()[1];
find_if(G.begin(), G.end(), [=](const Node& a){return a.getLabel() == linked_node_label;});
if ( is_linked_node_in_G != G.end() ) {    // check linked_node is in G
if (node_is_visited[linked_node_label] == not_visited) {    // check if linked_node is already in the queue
Q.push(linked_node);                                    // if not, add it to the queue
//                    cout << "current " << current_node.getLabel()       // for debug
//                         << " | linked = " << linked_node_label + 1
//                         << " | path length = " << dist[linked_node_label] << endl;
if (linked_node_label == target_node_label)  // end search once target node is explored
return true;
}
} else {
if (linked_node_label == target_node_label)  // end search once target node is explored
return false;
}
}
Q.pop();
}

return false;
}


DisjointSet.h

#ifndef DISJOINTSET_H_INCLUDED
#define DISJOINTSET_H_INCLUDED

#include "Graph.h"

class DisjointSet
{
public:
DisjointSet(const size_t);
~DisjointSet();
DisjointSet& operator= (const DisjointSet&);

int find(const Node&);
void unionNodes(const Node&, const Node&);

int get(int index) {return *leaders[index];}

private:
size_t size;        // graph size needed for allocation of pointer data members below
int* base;          // array of int each Node of the graph has its leader
int** leaders;      // array of pointers to int, allows to reassign leaders to Nodes after unions

int find_int(int);  // auxiliary to 'find' above

DisjointSet(const DisjointSet&);    // copy constructor forbidden
};

#endif // DISJOINTSET_H_INCLUDED


DisjointSet.cpp (here is where advice would be most appreciated)

// Union-find structure (lazy unions)
#include "DisjointSet.h"

DisjointSet::DisjointSet(size_t in): size(in), base(new int[in]), leaders(new int*[in])
{
for (size_t i = 1; i != in + 1; ++i) {
base[i - 1] = i;
leaders[i - 1] = &base[i - 1];
}
}

DisjointSet::~DisjointSet()
{
delete[] base;
}

DisjointSet& DisjointSet::operator= (const DisjointSet& rhs)
{
if (this == &rhs)
return *this;       // make sure you aren't self-assigning
if (base != NULL) {
delete[] leaders;     // get rid of the old data
delete[] base;
}

// "copy constructor" from here
size = rhs.size;
base = new int[size];
leaders = new int*[size];
base = rhs.base;
for (size_t i = 0; i != size; ++i)

return *this;
}

// auxiliary to find: implements the recursion
{

return parent;
}

// returns leader to input Node
int DisjointSet::find(const Node& input)
{
int parent( input.getLabel() );
if (input.getLabel() != *leaders[input.getLabel() - 1])
parent = find_int(*leaders[input.getLabel() - 1]);

return parent;
}

// merges sets by assigning same leader (the lesser of the two Nodes)
void DisjointSet::unionNodes(const Node& a, const Node& b)
{
if (find(a) != find(b)) {
if (find(a) < find(b))
leaders[find(b) - 1] = &base[find(a) - 1];
else
leaders[find(a) - 1] = &base[find(b) - 1];
}
}


KruskalClustering.h

#ifndef KRUSKALCLUSTERING_H_INCLUDED
#define KRUSKALCLUSTERING_H_INCLUDED

#include <vector>
#include "Graph.h"
#include "DisjointSet.h"

int clusteringKruskalNaive(const Graph&, const std::vector<Edge>&, int);
int clusteringKruskalDisjointSet(const Graph& graph0, const std::vector<Edge>& edges, int);

#endif // KRUSKALCLUSTERING_H_INCLUDED


KruskalClustering.cpp

// Kruskal MST algorithm. Implementation specific to  (k=4)-clustering
// -naive (with breadth-first-search to check whether new edge creates a cycle); cost: O(#_edges * #_nodes)
// -and union-find implementations.  Cost: O(#_edges*log2(#_nodes))
#include <iostream>
#include <string>
#include <vector>
#include <algorithm>        //std::find_if
#include "KruskalClustering.h"

using std::cout;            using std::endl;
using std::string;          using std::vector;
using std::find_if;

int clusteringKruskalNaive(const Graph& graph0, const vector<Edge>& edges, int k)
{
Graph T;    // Minimum Spanning Tree
vector<Edge>::const_iterator edges_iter = edges.begin();
int sum_costs(0), number_of_clusters( graph0.size() );  // keep track of overall cost of edges in T, and of clusters
while (number_of_clusters >= k) {

// find out if first node of edge is already in T
Graph::iterator is1_in_T = find_if(T.begin(), T.end(),
[=] (Node& a) {return a.getLabel() == graph0.getNode(edges_iter->getNodes()[0] - 1).getLabel();});
bool is1_in_T_flag;         // needed because T gets increased and thus invalidates iterator is1_in_T
Node* node_1 = new Node(*edges_iter); // no use of pointer here, it creates a new Node anyway, can't move Nodes to T
if ( is1_in_T == T.end() ) {      // node_1 not in T so we add it
number_of_clusters--;         // node_1 is not its own cluster any more
delete node_1;                // node_1 copied to T so ...
node_1 = &(T[T.size() - 1]);  // ... it now points there
sum_costs += (*edges_iter).getCost();
is1_in_T_flag = false;
} else {                          // node_1 is in T already
delete node_1;                // if so, just update the pointer
node_1 = &(*is1_in_T);
is1_in_T_flag = true;
}

// perform BFS to check for cycles
bool check_cycles = breadthFirstSearch(T, edges_iter->getNodes()[0], edges_iter->getNodes()[1]);

// create an identical edge, but with nodes positions swapped
Edge swapped_edge = *edges_iter;
swapped_edge.swapNodes();
// find out if second node of edge is already in T
Graph::iterator is2_in_T = find_if( T.begin(), T.end(),
[=] (Node& a) {return a.getLabel() == graph0.getNode(edges_iter->getNodes()[1] - 1).getLabel();});
// (either node1 or 2 not in T, or both, or both present but in different clusters of T)
if (!check_cycles) {         // if edges_iter creates no cycle when added to T
if (is1_in_T_flag){          // if node_1 was already present in T
(*node_1).addEdge(*edges_iter);       // just add new edge to node_1 list of edges
sum_costs += (*edges_iter).getCost();
} else
number_of_clusters++;   // if node_1 not present, it means number_of_cl was decreased above ...

number_of_clusters--;       // ... and number_of_cl can decrease just by one, if adding an edge to T
if ( is2_in_T != T.end() )  // node_2 already in T: just update its list of edges
else {                      // node_2 not in T, so add it
Node node_2(swapped_edge);
}

} else {                        // cycle created by *edges_iter
if (!is1_in_T_flag)         // in case the cycle happened just after adding node_1:
(*is2_in_T).addEdge(swapped_edge);   // add edge to node_2, num_clusters already updated by node_1
}
if (number_of_clusters >= k)    // advance to next Edge if num_clusters > k
edges_iter++;

// debug
//        T.output();
//        cout << "next edge: (" << (*edges_iter).getNodes()[0] << "-"
//             << (*edges_iter).getNodes()[1] << ") " << endl;
//        cout << "clustering: " << number_of_clusters << endl;
}
cout << "Sum of MST lengths is: " << sum_costs << endl;

return (*edges_iter).getCost();
}

// same algorithm, implemented with Union-find structure
int clusteringKruskalDisjointSet(const Graph& graph0, const vector<Edge>& edges, int k)
{
DisjointSet disjoint_set_graph0( graph0.size() );       // create Union-find structure

Graph T;
vector<Edge>::const_iterator edges_iter = edges.begin();
int sum_costs(0), number_of_clusters( graph0.size() );

while ( number_of_clusters >= k ) {
// if nodes in Edge have not the same leader in the disjoint set, then no loop is created, and T can add the edge
if ( disjoint_set_graph0.find(graph0.getNode(edges_iter->getNodes()[0] - 1)) !=
disjoint_set_graph0.find(graph0.getNode(edges_iter->getNodes()[1] - 1)) ) {
sum_costs += (*edges_iter).getCost();
number_of_clusters--;               // no cycle created so the edge will be added to T

// look for node_1 in T
Graph::iterator is1_in_T = find_if(T.begin(), T.end(),
[=] (Node& a) {return a.getLabel() == graph0.getNode(edges_iter->getNodes()[0] - 1).getLabel();});
if ( is1_in_T == T.end() ) {   // if node_1 not in T add it
Node node1(*edges_iter);
} else                         // if node_1 already in T only add to it this edge

Edge swapped_edge = *edges_iter;
swapped_edge.swapNodes();
// look for node_2 in T
Graph::iterator is2_in_T = find_if(T.begin(), T.end(),
[=] (Node& a) {return a.getLabel() == graph0.getNode(edges_iter->getNodes()[1] - 1).getLabel();});
if ( is2_in_T == T.end() ) {     // same as for node_1
Node node2(swapped_edge);
} else

// merge the 2 nodes' sets: update their disjointed set leaders
disjoint_set_graph0.unionNodes( graph0.getNode(edges_iter->getNodes()[0] - 1),
graph0.getNode(edges_iter->getNodes()[1] - 1) );
}

if (number_of_clusters >= 4)
edges_iter++;
//debug
//        T.output();
//        cout << "next edge: (" << (*edges_iter).getNodes()[0] << "-"
//             << (*edges_iter).getNodes()[1] << ") " << endl;
//        cout << "clustering: " << number_of_clusters << endl;
//        for (size_t i = 0; i != graph0.size(); ++i)
//            cout << disjoint_set_graph0.get(i) << " ";
//        cout << endl;
}
cout << "Sum of MST lengths is: " << sum_costs << endl;

return (*edges_iter).getCost();
}


Lastly, main.cpp :

/* A max-spacing k-clustering program based on Kruskal MST algorithm.
The input file lists a complete graph with edge costs.
clusters k = 4 assumed.*/
#include <iostream>
#include <fstream>
#include <string>
#include <vector>
#include <algorithm>        // std::sort;
#include "Graph.h"
#include "KruskalClustering.h"

using std::cout;                using std::endl;
using std::string;              using std::ifstream;
using std::vector;              using std::sort;

int main(int argc, char** argv)
{
cout << "Reading list of edges from input file ...\n" << endl;

// read graph0 from input file
string filename(argv[1]);

Graph graph0(filename);
//    graph0.output();   // debug
cout << endl;

// re-read input file and create a list of all edges in graph0
ifstream is(filename + ".txt");
int nodes_size;
is >> nodes_size;
vector<Edge> edges;
int node1, node2, cost;

while (is >> node1 >> node2 >> cost) {
Edge current_edge(node1, node2, cost);
edges.push_back(current_edge);
}
is.close();

// sort the edge list by increasing cost
cout << "Sorting edges ...\n" << endl;
sort(edges.begin(), edges.end(), compareCosts);
//    for (vector<Edge>::iterator iter = edges.begin(); iter != edges.end(); ++iter)       // debug
//        cout << (*iter).getNodes()[0] << " " << (*iter).getNodes()[1] << "  " << (*iter).getCost() << endl;

cout << "Kruskal algorithm: Computing minimal distance between clusters when they are reduced to 4 ...\n" << endl;

int k = 4;   // number of clusters desired

// pick implementation, comment the other
//int clustering_min_dist = clusteringKruskalNaive(graph0, edges, k);
int clustering_min_dist = clusteringKruskalDisjointSet(graph0, edges, k);

cout << "k = " << k << " clusters minimal distance is: " << clustering_min_dist << endl;

return 0;
}


## Edit: added input .txt file

For completeness, here below an input file in the format accepted by the program. The file contains an undirected Graph, first line is the number of Nodes, the others are Edges(node1, node2, distance). The file should be a .txt .

8
1 2 50
1 3 5
1 4 8
1 5 47
1 6 3
1 7 42
1 8 36
2 3 60
2 4 34
2 5 6
2 6 27
2 7 62
2 8 61
3 4 58
3 5 53
3 6 37
3 7 54
3 8 12
4 5 63
4 6 29
4 7 52
4 8 44
5 6 1
5 7 16
5 8 6
6 7 45
6 8 52
7 8 60

• ...perhaps looking a bit messy, too many comments and debugging, commented-out "print" statements scattered around the files? Sep 13, 2020 at 19:28

# Use your own typedefs

I see this code:

typedef std::list<Edge>::iterator iterator;
...
std::list<Edge>::iterator begin() {return edges.begin();}


You should use your own typedefs, so the second line becomes:

iterator begin() {return edges.begin();}


Apart for less typing for yourself, it also avoids possible mistakes where you return a different type than the one you typedef'ed, and if you ever want to change the type you only have to do it in one place.

# Simplify your classes

Looking at class Edge, I see that most member functions allow trivial getting and setting the private member variables. Why not make those member variables public, and allow them to be accessed directly?

Also, if you have a getNodes() that returns an array and an operator[] to access the nodes as an array, maybe you should have store node1 and node2 as a std::array<int, 2> nodes to begin with. Although if you want edges to be directed, I would keep them separate and name them from and to.

Furthermore, use default member initializers to void having to write any constructor:

class Edge {
public:
bool operator==(const Edge &) const;
bool operator!=(const Edge &) const;

std::array<int, 2> nodes{};
int cost{};
};


Now you can do:

Edge e{1, 2, 3};
e.nodes = {4, 5};
e.cost = 6;
e.nodes[0] = 7;


You can still add some convenience functions as needed, such as swapNodes(). Also, note that in C++20, with the above class, you can let the compiler generate default comparison operators for you in one line, like so:

class Edge {
public:
...
auto operator<=>(const Node&) const = default;
};


The same goes for class Node and class Graph: avoid making member variables private to then just write lots of wrapper functions to allow them to be accessed anyway.

# Avoid unnecessary temporary variables

In Edge::getNodes(), there is no need to use a temporary variable, just write:

array<int, 2> Edge::getNodes() const
{
return {node1, node2};
}


Of course this function is not necessary at all if you just store them as a public array to begin with.

# Use std::swap() instead of writing your own

Instead of using the hand-written trick to swap two variables (did you check that you don't run in to undefined behaviour due to signed integer overflow?), there's a function in the standard library to do that for you:

void Edge::swapNodes() {
std::swap(node1, node2);
}


# Don't try {throw;}

If you want to signal an error using an exception, just throw an exception of the proper type that contains the error message, like so:

int Edge::operator[](const int index) {
if (index == 0) return node1;
else if (index == 1) return node2;
else throw std::runtime_error("edge index must be either 0 or 1");
}


What you are doing is too complicated (5 statements on 3 lines of code instead of a single statement), doesn't allow the caller to handle the error, and is illegal: you are only allowed to use throw without an argument if you are already in exception handling code.

# Don't if (x) return true else return false;

If you already have a boolean expression, you don't have to use an if-then-else statement to return exactly the same boolean value, just write:

bool Edge::operator==(const Edge& rhs) const {
return node1 == rhs.node1 && node2 == rhs.node2 && cost == rhs.const;
}


# You can use initializer lists in member value initializers

Ok that sounds like nonsense, but what I mean is that you can construct a std::list with an initializer list to set the initial list members, and do that when doing member variable initialization in a constructor. For example:

Node::Node(const Edge& edg): label(edg.getNodes()[0]),
score(10000000), // up to now, same as before
edges{edg}       // initializer list passed to constructor of edges
{
}


Note that you should combine this with default member initializers, so it becomes:

class Node {
...
int label{};
std::list<Edge> edges;
int score{10000000};
};

...

Node::Node() = default;
Node::Node(const Edge &edg): label(edg.getNodes[0]), edges{edg} {}


# Iterating over the list of edges

A std::list does not have a random access iterator. You have noticed that already, and that is why in Node::getEdge(), you are using std::advance() to iterate to the element at the given index. You should know that this is quite slow, since it has to follow pointers. If you really need random access, it would be much faster to store the list of edges in a std::vector instead. But, in most cases you are iterating over the list of edges, and do some operation on each of them. In that case, don't use getEdge(), as it will get slower and slower the bigger the index is, but use list iterators directly, or even beter use range-for if possible. For example:

void Graph::output() const {
for (auto node: nodes) {
std::cout << "Node " << node.getLabel() << ", size = " << node.edges.size() << "  with edges:";

for (auto edge: node.edges) {
std::cout << "   [" << edge[0] << "-" << edge[1] << ", " << edge.getCost() << "]";
}

std::cout << "\n";
}
}


# Use "\n" instead of std::endl

Prefer using "\n" instead of std::endl; the latter is equivalent to the former, except that it also forces a flush of the output buffer, which is often unnecessary and can slow down your program.

# Writing generic output functions

Don't hard-code the use of std::cout in your output functions, but instead take a parameter that tells it what output stream to use, like so:

void Graph::output(std::ostream &out) const {
for (auto node: nodes) {
out << "Node " << node.getLabel() << ", size = " << node.edges.size() << "  with edges:";
...


Also consider creating an operator<<(), like so:

class Graph {
...
void output(std::ostream &out) const;

friend std::ostream &operator<<(std::ostream &out, const Graph &graph) {
graph.output(out);
}
}:


Now that you have overloaded operator<<() for your class, you can just write:

Graph graph;
...
std::cout << graph;


# Writing generic input functions

The same goes for functions reading input as well. You have a constructor that takes a filename as an argument, but what if I have my graph stored in memory instead of in a file? Use a std::istream parameter, so that anything that can act as an input stream can be used to read the graph from:

class Graph {
...
Graph(std::istream &is);
};

...

Graph::Graph(std::istream &is) {
int number_nodes;
is >> number_nodes;
...
}


And then you just use it like so:

std::ifstream is(argv[1]);
Graph graph0(is);


# Do proper error checking when reading and writing

Always be prepared that something can happen while reading and writing files. Maybe the disk is corrupt, you have run out of disk space, the permissions are incorrect, the file is on a network drive and the network is down, and so on. Also be prepared that the file might contain unexpected data.

Your while-loop in the constructor of Graph that reads from a stream will exit either if it reached the end of the stream, or if there was an error. To ensure that it read everything correctly, check if is.eof() returns true. If not, I suggest you throw a std::runtime_error.

When writing to a stream, whether it is a file or std::cout, check at the end of writing everything if out.good() returns true. If not, throw an exception again.

# About using C-style pointer arrays

Yes, you can use those, but now you have to manually allocate and free the arrays. It's fine as an excercise, but an even better excercise is to convert those into std::vectors :)

# Avoid new and delete in general

Manually calling new and delete is hardly ever necessary anymore, and is often a sign something should be managed by an appropriate container, or should not be a pointer in the first place. As an example of the latter, in clusteringKruskalNaive(), there is no need to use new to create the temporary node_1. It is deleted immediately afterwards, possibly without even having been used. Instead write:

bool is1_in_T_flag = is1_in_T != T.end();
Node *node_1;

if (is1_in_T_flag) {
number_of_clusters--;
node_1 = ...
} else {
node_1 = &*is1_in_T;
}


# Try to make the code more readable

Especially in KruskalClustering.cpp, the code is very hard to read. Try to shorten the lines, by using auto where possible, simplifying expressions, giving clear and concise names to things, splitting off complex code into their own functions, and so on. Be consistent in how you initialize variables, I see both int x = 1 and int x(1). Also remove commented out code; you should use a revision control system like Git to store previous versions of you code in. Here is how it might look:

int clusteringKruskalDisjointSet(const Graph& graph, const vector<Edge>& edges, int k)
{
// Start with every node being in its own cluster
auto number_of_clusters = graph.size();
DisjointSet disjoint_set(number_of_clusters);

int min_cost = INT_MAX;
int sum_costs = 0;

// Iterate over the given edges until only k distinct clusters are left
for (auto edge_iter = edges.begin; number_of_clusters >= k; ++edge_iter) {
auto &edge = *edge_iter;

// Skip the edge if it is within a single cluster
if (disjoint_set.find(graph.nodes[edge.nodes[0] - 1]) ==
disjoint_set.find(graph.nodes[edge.nodes[1] - 1]))
{
continue;
}

// This edge joins two clusters
min_cost = std::min(min_cost, edge.cost);
sum_costs += edge.cost;
number_of_clusters--;

disjoint_set.unionNodes(graph.nodes[edge.nodes[0] - 1],
graph.nodes[edge.nodes[1] - 1]);
}

std::cout << "Sum of MST lengths is: " << sum_costs << "\n";
return min_cost;
}


I am left wondering: why keep track of Graph T in this function? It didn't do anything, so I could remove all code related to it. And why subtract 1 from the node indices here? Why loop while number_of_clusters >= k? That will result in k - 1 clusters at the end of the algorithm.

• Brilliant answer, I have learned a good deal from it on how to improve my classes, thank you! Of course I was too lazy to go through the code and modify names after defining my own typedefs, once code works there is a certain tendency in programmers towards inertia ... Also it seems obvious to me now that my Graph.h classes should really be structs, too many awkward getters as they are now. And how you stripped down the Kruskal routine to its essentials really blew my mind! Overall, I have learned a lot about my shortcomings against good practices, this is exactly what I needed. Sep 14, 2020 at 11:28
• your question: I needed the maximum spacing between clusters when they are reduced to k, that is, the next Edge in the sorted vector<Edge>, after the Minimum Spanning Tree T is reduced to k clusters. This is why I get k - 1 clusters on exit. Graph T is the MST; that Kruskal function really should calculate the MST of input Graph. In this context was not really needed and you intuition has greatly streamlined the code. Node indices -1 because in the input file they are labelled starting from 1, which would be vector[0] in C++, I had to litter indices with -1 everywhere Sep 14, 2020 at 11:44