question about code

[KAI-SUWer]

it is a nice work, but i have some questions to ask you, wish you can help me to solve.
code from dlpthresCurve file:

          adj = adj(:,:,2:end);  
          predMat = predMat(:,:,2:end);
          if strcmp(links,'new')
              nodeNew = ~nodeExisting;  
              activeMask = nodeNew & activeMask;

now we suppose t=1，so 'nodeNew' is edge that doesn't exist at time t=1；'ActiveMask' is the fully connected condition of activated nodes at t=1. Is ’nodeNew&activeMask‘ limiting new connections at t+1=2 to active nodes at t=1 ?

[KAI-SUWer]

activeMask = false(n,n,tMax-1);
for t = 1:tMax-1
activeMaskCurr = nodeActive(:,t)*nodeActive(:,t)'; % Should be changed to nodeActive(:,t+1)*nodeActive(:,t+1)'
activeMaskCurr(diag(true(n,1))) = 0;
activeMask(:,:,t) = activeMaskCurr;
if directed == false
activeMask(:,:,t) = tril(activeMask(:,:,t));
end
end
adj = adj(:,:,2:end);
predMat = predMat(:,:,2:end);
if strcmp(links,'new')
nodeNew = ~nodeExisting;
activeMask = nodeNew & activeMask; % If it is changed, it corresponds to all possible new edges of t+1
elseif strcmp(links,'existing')
activeMask = nodeExisting & activeMask;
end
adjVect = adj(activeMask); % Extract ground_trueth links of all possible new edges at the corresponding time
predVect = predMat(activeMask) % So you can calculate PRAUC in terms of adjVect and preVect

Please see my code comments.

[kevin-s-xu]

now we suppose t=1，so 'nodeNew' is edge that doesn't exist at time t=1；'ActiveMask' is the fully connected condition of activated nodes at t=1. Is ’nodeNew&activeMask‘ limiting new connections at t+1=2 to active nodes at t=1 ?

This is correct. We do not try to predict links to nodes that have just become active at time t+1 (new nodes). Without any information about those nodes, any predictions we make would be to a "generic" new node since we don't have any information regarding the new node. See Section 5 of Yang et al. (2015) for discussion on this.

Yang, Y., Lichtenwalter, R. N., & Chawla, N. V. (2015). Evaluating link prediction methods. Knowledge and Information Systems, 45, 751-782. https://www3.nd.edu/~dial/publications/yang2015evaluating.pdf

[KAI-SUWer]

Thank you very much for your answer.
I have one last question, how to calculate P and N in the indicators? Do you tally up edges and non-edges in all historical snapshots used for prediction, or otherwise? Thank you again.

Instead of calculating the P and N values in your MATLAB code, you use a full one matrix to get a base_PRAUC.