Algorithm to make a route from Source Node to End Node using every node exactly once - graph

Question

I'm trying to create a route, given a start node and end node, that will travel to every single node in the graph, and minimize the cost of doing so.
The graph is undirected, and every node is connected to each other directly. The weight of every edge is positive. I think because every node connects to eachother, that there are many "loops" in my graph, but I don't want to generate a route with a loop in it.
So, for a graph with N nodes, I have (N*N-1) directed edges. A graph with nodes A,B,C,D would have edges:
A to B / B to A
A to C / C to A
A to D / D to A
B to C / C to B
C to D / D to C
B to D / D to B
When I implement the Floyd Warshall algorithm from Wikipedia, I only ever get an array of 2 nodes. There's a function in the article that gives you the shortest path from node U to node V, and that's the function that only returns [U,V] (array containing U and V)
I MUST be misunderstanding what it is exactly that Floyd Warshall is meant to solve. I'll attach my code to show how I've implemented it in javascript.
function colorsToEdgeMatrix(colors){
var dist = [];
for(var i = 0; i < colors.length;i++){
dist[i] = [];
var c1 = colors[i];
for(var j = 0; j < colors.length;j++){
if(i == j){continue;}
var c2 = colors[j];
dist[i][j] = colorDistance(c1,c2);
}
}
return dist;
}
function colorsToNextMatrix(colors){
var next = [];
for(var i = 0; i < colors.length;i++){
next[i] = [];
for(var j = 0; j < colors.length;j++){
if(i == j){continue;}
next[i][j] = j;
}
}
return next;
}
//lab colors
function FloydWarshallWithPathReconstruction (colors){
var next = [];
var dist = colorsToEdgeMatrix(colors);
var next = colorsToNextMatrix(colors);
var N = colors.length;
for(var k = 0; k < N; k++){ // standard Floyd-Warshall implementation
for(var i = 0; i < N; i++){
for(var j = 0; j < N; j++){
if(dist[i][k] + dist[k][j] < dist[i][j]){
dist[i][j] = dist[i][k] + dist[k][j]
next[i][j] = next[i][k]
}
}
}
}
return next;
}
function Path(next,u, v) {
var path = [];
if(next[u][v] == null){
return []
}
path = [u]
while(u != v){
u = next[u][v]
path.push(u)
}
return path;
}
var lab = randomLABArray(100); //make an array of LAB color space colors. a LAB element has an array structure [L,a,b]
lab = sortLuminosityLAB(lab); //sorts the LAB colors from light to dark
var next = FloydWarshallWithPathReconstruction(lab); //gets all paths using floyd warshall
var path = Path(next, 0, lab.length-1); //gets the path calculated from lightest to darkest
console.log( path );
Does this algorithm not necessarily return a path that goes through every node? I guess what it does is just spits out the best path for every start and end node, and doesn't guarantee any path goes through every node...
I used the Nearest Neighbor algorithm with decent result, and Brute Force is impossible after 10 elements. I was hoping the Floyd Warshall would give even better results
Huh...so this might actually be a Hamiltonian Path problem and isn't as simple as I thought...

Answer 1

This can be reduced to the Traveling Salesman problem as follows. Pick a large number M which is greater than the sum of all weights in the graph. Add this number to the weights on all edges in the graph except the edge connecting the start node and the end node. That edge should have its cost set to zero. Solve the resulting Traveling Salesman Problem. It is easy to see that the optimal solution of the TSP will include the edge connecting the start and end node. Throw that edge away from the optimal solution and adjust the weights back to their original value -- and you have found the optimal solution to your problem.
Conversely, the regular TSP can be reduced to your problem in an even easier way. Pick a node at random and duplicate it. Let one of these newly duplicated nodes be the start node and the other be the end node. Solve your optimal path problem for this instance. Then -- remerge these two nodes into the original node, which will splice the ends together to form a circuit, which is easily seen to be optimal. This confirms your intuition that the problem is at least as hard as the Hamiltonian circuit problem.