Belief Propagation Example #6
Conversation
| @@ -116,15 +116,15 @@ function ITensors.apply(o⃗::Vector{ITensor}, ψ::ITensorNetwork; cutoff, maxdi | |||
| end | |||
|
|
|||
| function flattened_inner_network(ϕ::ITensorNetwork, ψ::ITensorNetwork) | |||
| tn = inner(ϕ, ψ) | |||
There was a problem hiding this comment.
I think inner should already take the proper dag (it should already dag the first tensor network ϕ).
There was a problem hiding this comment.
If the dag isn't be handled correctly by inner than it should be fixed there.
| function construct_initial_mts(psi::ITensorNetwork, g::NamedDimGraph, s::IndsNetwork) | ||
| #Make empty lists | ||
| forwardmts = ITensor[] | ||
| backwardmts = ITensor[] |
There was a problem hiding this comment.
What about storing the message tensors in a Dictionary that maps the edges the message tensors are on to the message tensor? Then it is easy to look up the message tensors associated with a vertex by searching through the incident edges of the vertex.
|
|
||
| #Normalise it | ||
| Mdat = matrix(M) | ||
| M = M / sqrt(tr(Mdat*Mdat')) |
There was a problem hiding this comment.
You should be able to use normalize!(M) here.
|
Glad to see how simple it is! We could probably just move |
|
|
||
| println("One Dimensional Chain") | ||
| g = NamedDimGraph(Graphs.SimpleGraphs.grid([n]), [i for i = 1:n]) | ||
| s = siteinds("S=1/2", g) |
There was a problem hiding this comment.
I guess we should allow passing a Graphs.SimpleGraph to siteinds if that isn't supported right now. Also we could make a default constructor NamedDimGraph(g::SimpleGraph) = NamedDimGraph(g, 1:nv(g)).
Request to merge some changes to main branch.
Changes:
i) BeliefPropagationFunctions.jl - general purpose functions for implementing belief propagation
ii) ExampleBeliefPropagation.jl - Code for running a small example of using belief propagation to calculate local expectation
values for a random ItensorNetwork