JCC-PSTN

An implementation of the content from the paper "Joint Chance Constrained Probabilistic Temporal Networks via Column Generation" submitted to SOCS 2022. Included is a python implementation of the column generation method for solving Strong Controllability of Joint Chance-Constrained Probabilistic Simple Temporal Networks, along with some test cases and results.

Quick Start

To run the experiments, Gurobi optimiser is required along with the Gurobi Python package GurobiPy. For details see:

https://www.gurobi.com/documentation/9.5/quickstart_linux/index.html

In order to solve a particular instance it is necessary to:

• Load the problem using PSTN = pickle.load(filename)
• To solve the JCC-PSTN problem via column generation convex.solveJCCP(PSTN, risk, gap, log, logfile, max_iterations, cg_tol)
  • This should output a tuple (model, JCCP) where model is a Gurobi Model object (for information on how to query parameters see https://www.gurobi.com/documentation/9.5/refman/py_model.html#pythonclass:Model) and 'JCCP' is an instance of JCCP (for information on how to query parameters see JCCP_class.py)
• To solve the problem using the Strong Controllability LP with Boole's inequality do LP.solveLP(problem, name, risk).
  • This returns a tuple (model, result) where model is a Gurobi Model object as above. Result is a dictionary {"Runtime": float, "Objective": float, "Variable": value}

In order to return the schedule from the Gurobi model, you can use getSchedule(PSTN, model) and to get the relaxations use getRelaxations(PSTN, model).

Examples

In the folder pstns/problems there are a number of problem cases generated from the IPC planning domains. PDDL domain, problem and plan files are stored in the folder pstns/domains/tfd-benchmarks.