Question

I have these informations to save in a variable for my basic neural network simulation.

• Node (NodeId, State)
• Relationship (SourceNodeId, TargetNodeId, Weight, State)

State is the activation level which is the only value which changes during simulation. It is an unsigned float.

I want to easily get all incoming and all outgoing relationships for the current node. By easily I mean very great performance. (I have around 1,000,000 nodes with each 50 relationships on average.)

The main part of my program looks like this (pseudo code).

foreach(Node in Nodes)
{
     Inputs[] = all incomeing relationships;
     Node.State = sum of all Inputs[] elements;

     Outputs[] = all outgoing relationships;
     normalize all Outputs[] elements temporarly; // so that the sum of their weights is 1

     foreach(Output in Outputs[])
     {
          Output.State = Node.State * Output.Weight;
     }
}

I hope you understand what I want to do. If not I will try to explain better.

What type of Object would be best to have quick access to the nodes by their SourceNodeId and by their TargetNodeId?

PS: Programming in C# using Visual Studio.

Answer 1

I think list with nodes + two syncronized Dictionaries will be fast.

List<Node> allNodes;
Dictionary<Node,List<Node>> sourceTarget;
Dictionary<Node,List<Node>> targetSource;

I recommend you to encapsulate them into single object… Two-Way-Dictionary:

class TwoWayDictionary<T1,T2>
{
    private Dictionary<T1,List<T2>> sourceTarget;
    private Dictionary<T2,List<T1>> targetSource;

    // Here shold be public methods and accessors ...
}