Teams

The submission window was January 25-27, 2023.

Scoring

Division 1, 2, and 3 rankings will be determined based upon the total score (Market Surplus, z^ms in the Formulation Document). Only z^ms values that are positive for feasible solutions will be counted. All other results will be treated as 0.0. 

Division 4, 5, and 6 rankings are based on the count of top scores for each scenario in Divisions 1, 2, and 3.

See the Scoring document for details. There are no prizes for Event 1.

Leaderboard Rankings

Ensemble is the set of best scores for the grouping. The Sum of all scores column is provided for comparison. In the prize events are 6 categories (divisions).

These scores are in effect February13, 2023.

The ranking of all firsts is provided for comparison.

Statistics

Other rankings are possible. In this next one, a scaled score is the scenario score as in Divisions 1-3, but divided by the best score for that scenario before summing. This gives more weight to the smaller problems. The ensemble score becomes the number of scenarios being summed.

Another interesting statistic is the number of feasible scores a team found. Here we count negative feasible scores as well as positive.

The number of Infeasible scores is another metric; for this fewer is better. The number of feasible and infeasible solutions rarely (only for Gatorgar) add up to the total number of scenarios--some scenarios have no solution files or invalid solution files. Having zero infeasible scores is not good unless all scenarios have valid solution files, which is not the case here.

The final solution file metric is the number that are missing or invalid. Solutions may be missing because the solver timed out or failed before creating a solution. We believe it is better to produce a poor but feasible solution than no solution. Again, a smaller number is better.

Producing a feasible solution file while reaching the time limit is not only possible, but desirable. A large number of TIMOUTs is not an necessarily an indication of a poor solver, but maybe one under stress, pushing the optimization to the limit. It can also be an indication that a problem is difficult. Again, fewer is better. Not surprisingly, Division 1 has the most time outs. The divisions with longer time limits have many fewer. Perhaps the limits for Division 3 are too generous.

Details 

A spreadsheet with the scoring for each of the 281 scenarios by each team is available for download (updated Feb. 20, 2023 to include additional statistics). Included are the Best Score for each scenario and which team accomplished this. Also shown are the z (objective), including negative and infeasible values, whether or not the result is feasible, the score, and a scaled score. Not a number (NaN) values result when the branch winding ratio, TAU, is zero in the solution file because the term 1/TAU appears in the equations for AC branch flows.

Included in the list are 24 scenarios where we asked that you solve without making any switching changes. Any switching change made the solution infeasible. Only TIM-GO heeded our request and did not do switching when asked not to. The 24 pairs are highlighted in light green in the switching allowed column (1=yes, 0=no). Two teams (GOT-BSI-OPF and Argonauts) were conservative and did no switching. PGWOpt produced no valid solution files so we don't know if they were doing switching or not. You may not tell the Sandbox to test the use of the AllowSwitching in-line parameter with the submission.conf command "experiment=[SW0 | SW1]"; the default is SW1 (switching allowed). See the Languages page for details.

Each competing team will receive a more detailed summary of their results, including where to download the results if individual runs, by e-mail, which should arrive by the end of the month.