A scalable deep reinforcement learning framework for activation informed influence maximization on Large graphs.
Anonymized GitHub repo.
===================== File specification =========================
- data/ca-CSphd/: Train and val graphs from different families.
-
src/data/config.py: project directory paths
-
src/data/utils.py: utility functions
-
src/data/make_graph.py: preprocess raw graphs
-
src/data/make_dglgraph.py: prepare graphs for dgl liblary
-
src/data/load_graph.py: load graph in dgl liblary
-
src/models/CandidateIMnodes/CandidateIMnodes_NodeClassPth.py : Node Classification model for identifying AIM candiidate nodes.
-
src/models/CandidateIMnodes/CandidateIMnodes_NodeClassPth_test.py: Test trained Node Classification model on new graphs.
-
src/models/traindqn.py: Main function for training deep Q learning agent for identifying AIM seed nodes.
-
src/models/dqnAgent.py: utility functions for Q learning agent.
-
src/models/models.py: Q network model.
-
src/models/testdqn.py: Test trained dqn agent and generate plots of performance.
-
src/models/Baseline_models.py: Functions for getting ground truths from Greedy Hill climbing and Influence capapcity scores.
- Trained node classification (candidiate node identiifcation) models in torch format.
- Trained QLearning agent models.
Run functions in testdqn.py to generate results and plots.
Note: There are other files that are developed on fly but are not needed for the output generation.
==================================== end =========================