Public test data and competition focus on constant constraint matrices

[ambros-gleixner]

Dear MIPcc23 participants,

Let me share an update on 1. the public data provided for the competition and 2. an updated focus of the computational evaluation.

1. We have now extended the public test data to eight instance series. An overview can be found in datasets/README.md.

2. The committee has now decided to focus the competition on instance series with constant constraint matrix.

While we are aware that changes in constraint coefficients do occur in practically relevant applications of MIP reoptimization, this is certainly the most challenging scenario and provides the least structure in order to learn from previous solving process.
Hence we decided for the purpose of the competition to narrow the focus for the computational evaluation to the instance series where the constraint coefficients are constant. Only right-hand sides may vary, and of course objective coefficients and bounds of the variables.

Naturally, we still hope to see innovation also for the more difficult scenario in the future. This motivated us to provide two public instance series built from formulations where the matrix coefficients may change. We are certainly curious to see, whether approaches exist that do not rely on a constant constraint matrix and are still successful in improving performance as the series progresses, whether now or in future research. The data provided will hopefully serve this purpose.

Finally, note that we have also published a simple evaluation script at examples/simple_evaluation that may serve as a starting point in Python for the minimally required checks to be performed.

Register until January 31st!

Happy MIPing,
the MIPcc23 organizers:
Ambros Gleixner, Timo Berthold, Mathieu Besançon, Suresh Bolusani, Claudia D'Ambrosio, Gonzalo Muñoz, Joseph Paat, and Felipe Serrano