Sir,
Please tell me about the variables input you have used for the program like 'G', 'F', 'param'.
If I already have the graph structure why do i need another signal f and what's this param. I will be highly obliged if you could help me with my doubts.

In some of the functions for creating graph signals G = gsp_erdos_renyi(N,p)
G = gsp_random_regular(100,3) , G.coords is taking default (0,0) value.

The current code throws "GSP_GRAPH_SPARSIFY: Epsilon out of required range" if gsp_graph_multiresolution() comes to the point that the current graph has three nodes. I think L175 in gsp_graph_multiresolution.m should be if Gs{lev+1}.N>3 as 2/sqrt(3)>1, a condition in gsp_graph_sparsify(). Another useful condition would be sum(find(Gs{lev+1}.A,1))>0 as otherwise there is nothing to sparsify and this later leads to error in results=gendist(Pe',q,1);.
In addition a piggy-backed comment: gsp_graph_sparsify() should be documented that it is non-deterministic.

I think the behavior on L141 in gsp_graph_multiresolution.m is wrong in the case when largest_eigenvector(1)==0, which now leads to filling largest_eigenvector with 0s as sign(0)=0. I suggest L141 being executed only if largest_eigenvector(1)~=0.
Also, as eig works up to eps precision, one should IMHO have nonnegative_logicals=(largest_eigenvector >= -eps); instead of nonnegative_logicals=(largest_eigenvector >= 0);, as the current may lead to problems with graphs with multiple connected components, which are then numerically not properly identified to be kept, which may lead to NaNs in Kron reduction.

Hello everyone,
In the following function: "gsp_isomap" you are using the "dijkstra" function hence it is not implemented in matlab! I tried to download one from MathWorks File Exchange but I found lots of functions!
Can you send me the right one?
Best regards and thanks in advance
Jennifer

I found a bug that makes an incorrect gft result with a ring graph.

Example

Gfted one-hot vector's amplitude is constant.
And its phase should be a line.
However, this example shows gft makes an incorrect result.

N = 64;
G = gsp_ring(N);
G = gsp_compute_fourier_basis(G);

f = zeros(N);
f(10) = 1;

plot(1:N,angle(gsp_gft(G,f)))
xlim([1 N])
ylim([-pi pi])
ylabel("phase [rad]")
yticks([-pi,0,pi])

The cause

This is because the sort of eigenvalues and eigenvectors in

However, this sort is important when we talk about graph frequency...

The current implementation of the Erdos-Renyi graph seems to have an issue related to the generation of the adjacency matrix. In particular, the edge probability does not follow the input probability p but it is roughly half. This is due to the following lines:

G.W = sprandsym(N,p)>0;
G.W(1:N+1:end) = 0;

where sprandsym(N,p) generates a sparse matrix with nonzero probability p but the nonzero entries are normally distributed, so taking >0 only selects half of those.

A quick fix, albeit inefficient, is

G.W = tril(rand(N,N)<p);
G.W = G.W - eye(N);
G.W = G.W + G.W';

In line 86 of the gsp_nn_hypergraph.m file, this line of code should be G.W (edge, ii)=sqrt(sum(w((1:k)+(ii-1)*k))), instead of G.W (edge, ii)=sqrt (sum (w (edge)).

Hiya,

I'm loving the toolbox! I was wondering if you had plans to add in different wavelet families like the Haar or daubechies wavelets?

When I run gsp_demo_learn_graph_parameters.m, I am prompted some functions have not been found such as squareform_sp(), sum_squareform(), lin_map() and so on.

function appears to have many dependencies not included in the repo, including:
normalize_data
gabwin