Environments: Python 3.6.10, networkx 1.11
There are four modules, namely node embedding generation ('embedding.py'), training the prediction function on the general prediction problem ('learning.py'), generation of test/validation query pairs with candidate paths ('phase1.py'), and finding the MPSP based on candidate paths ('phase2.py').
input graph format:
each line refers to an edge and has four elements, namely source node, end node, edge length and edge probabilities.
The execution workflow has four steps where each step consists of input -> module -> output (please refer to individual module for hyperparameter description):
(1) input graph -> embedding.py -> node embedding (stored in the folder 'embedding')
Example: python embedding.py --graph BJ --dimensions 128
(2) input graph -> phase1.py -> test/validation query pairs with candidates (stored in the folder 'test' or 'validation')
Example: python phase1.py --graph BJ --output validation --pairs 100 --m 20
(3) input graph, node embedding, validation query pairs -> learning.py -> trained predictive function (stored in the folder 'models')
Example: python learning.py --graph BJ --dimensions 128 --embedder L1 --N 100 --m 20 --ratio 0.1 --length 30 --counter 100
(4) trained predictive function, node embedding, test query pairs, graph for training, graph for deployed -> phase2.py -> results
Example: python phase2.py --graph BJ --dimensions 128 --trainedGraph BJ --targetGraph BJ --counter 500
Note: For your convenience, we have provided examplar outputs for steps (1) to (3) and step (4) is directly runnable.
Format of output of step (4):
Six lines in total.
Line (1): the graph/setting where the prediction function is trained and the graph/setting where the function is deployed.
Line (2): phase 1 average time and the top k values we are testing.
Line (3): performance of method RR - the AAP for each k.
Line (4): performance of FPA - the AAP for each k, ranksum percentiles (i.e., 0,0.25,0.5,0.75,0.8,0.9,0.95 and 1) and average for each k (all of them are stored in a list where the last element refers to the average), and running time.
Line (5): performance of Random - with the same format as in Line (4).
Line (6): performance of LPA - with the same format as in Line (4).
Note that all datasets are avaiable in public and the links are provided in our paper.