2 problem with minspantree.m in Matlab 2018a

Hi dear all; I want to find a min-span-tree of a matrix. I figured that MATLAB's own minspantree.m may be the most efficient algorithm. So i use

open minspantree.m

And here comes Question:

1. The code in minspantree.m used G.EdgeProperties.Weight and G.Underlying. G is a graph object. But when I use G.EdgeProperties.Weight or G.Underlying in Command window, both returns error: Error using graph/subsref (line 15) No public property 'EdgeProperties' for class 'graph'. Why?
2. minspantree.m line 62：

    [pred, edgeind] = primMinSpanningTree(G.Underlying, w, rootNode, restart);

Is primMinSpanningTree a function? But I can not find any: function [ ] = primMinSpanningTree() in minspantree.m, neither can I find primMinSpanningTree.m file in my whole disk. So what is primMinSpanningTree? What is its code? How can I find it and open it?

Thanks all very much.

Answer 1

[Christine Tobler]

The properties Underlying and EdgeProperties are private properties of the graph object, which can only be accessed in methods of the graph object. The function primMinSpanningTree is a builtin method of the internal helper object that is stored in G.Underlying.

It is not recommended to use either of these directly - the recommended way is to construct a graph object and call the method minspantree for this object.

[Tiger Zhao]

Thank you very much. Now I understood.

However, It is really Inconvenient for users to first convert matrix to graph object, to do some calculate, and then to re-convert graph object back to matrix. When the quantity of matrix is extremely large, this method is very inefficient.

I hope Matlab could launch new graph toolbox that both IN an OUT are matrix.

[Christine Tobler]

Hi Tiger,

I just came back from vacation and saw your reply. Sorry I did not realize that your motivation for looking at minspantree.m was performance.

If you have the Bioinformatics toolbox, you could try their function graphminspantree, which takes and returns a sparse matrix.

There is also the File Exchange package MatlabBGL, which has a function mst that takes and returns a graph represented as a matrix.

If you don't mind, I would like to know more about your workflow. Why do you need to convert a graph to a matrix and back again, instead of using the graph format throughout? Are there some things you can't do with a graph, where you need to have the matrix representation available instead?

[Tiger Zhao]

Hi Christine Tobler,

Thank you for your enthusiasm.

For example, There is a dynamic adjacent matrix, which means the graph is changing over time. The dimension of this dynamic adjacent matrix is P*P*N, where N could be very large.

So at every time point i, N=i, find the MST, calculate the MST's adjacent matrix, save them.

Of couse, I could do this using graph object. But it's a little bit slow. I tried both method. The both IN an OUT are matrix method is faster.