Train Schedule Solver

Solver for problem instances described on https://github.com/crowdAI/train-schedule-optimisation-challenge-starter-kit .

Building

From the root of the clone repository run:

mvn package

Running

The program takes two JSON files, the first is the input containing the problem instance, the second is the output of the solution.

Additionally, it takes following optional flags:

• max_penalty: Upper bound on the penalty. The solver searches only for solutions which have a penalty less (or equal) to this value.
• max_penalty_per_intention: Maximum penalty for a single service intention.
• connection_badness_factor: Scaling factor between connections and resource conflicts.
• director_type: Back-tracking strategy. One of PRIORITY_CONFLICT, HARD_COLLECTING.

Examples:

java -jar target/solver-0.0.1-SNAPSHOT.jar \
     problem_instances/03_FWA_0.125.json solution_03.json

This will compute a zero-penalty solution and write it to solution_03.json.

java -jar target/solver-0.0.1-SNAPSHOT.jar \
     problem_instances/05_V1.02_FWA_with_obstruction.json solution_05.json \
     --max_penalty_per_intention 13.0 --max_penalty 37.5 --connection_badness_factor 3.5 \
     --director_type PRIORITY_CONFLICT