This project contains puzzles that interest me. Currently it has only one puzzle: the bus-driver-planning challenge.

This project uses Gradle. To play with it, follow the steps below:

1. install gradle and git
2. git clone https://github.com/zjxhz/puzzles/
3. cd puzzles
4. to run the tests: gradle test. The reports are under build/reports
5. to run the application: gradle run. The main class is planning.busdriver.Main, as specified in build.gradle. This application reads data from a screenshot file as input, and tries to find a highest score by optimizing a group of randomly generated plans. Check the console output to see the result, it should be self-explanatory.
6. To play it online, visit: http://xuhuanze.me/puzzle/

Since there are many different ways to assign drivers to a single shift, brute forcing all possibilities is not practical, if possible at all. Also, simply making best moves for the current shift, without taking the coverall planning into consideration, does not produce the optimal result.

Inspired by the genetic algorithm, this application generates randomly a group of possible plans, which avoid violating major rules such as assigning shifts on off days, as the first generation. The crossover is done in a way by re-planning(optimizing) assignments on random day ranges for a given candidate(plan). Since optimizing each candidate are disjointed events, this algorithm can generate a "good enough" solution that reaches the 85% goal usually in seconds. Also, it has been observed that it is possible to reach the optimal target(100%), usually in a few minutes.

The robot player reads data from a generated plan, simulates mouse move and clicks on predefined positions. Being offered with different position data, it can play with the online challenge mentioned earlier, or the official quintiq challenge. The delay of each click, however, needs to be increased when playing with the official version as it takes some time for the official application to respond.