Game Engine Model Overhaul

The tXtFL game engine runs the game models that simulate football action. Over several years and many iterations, we optimized the game engine to provide realistic game scores based on player and team ability, often matching or coming close to scores in real games.

The game engine is not without flaws, however. A quick glance at individual player stats reveals often dismal performance by percentage that only mounts to realistic scores by sheer mounting of play numbers. The goal of this sub-project is to overhaul the game engine so that it performs reliably and realistically both in terms of final game scores and individual player statistics.

The old model

The old model is largely based on randomly drawing numbers, weighting by player skill values and position, summing across players, and comparing against teams. To add the possibility of always getting either extreme performance, sigmoid curves (e.g. arctan) were often employed as additional weights or for calculating scaled values, such as distances. The basic ideas is a sigmoid curve allows both high or low values on either extreme, but these values asymptotically reached predefined values to ensure that no value was astronomically awesome or absolutely abysmal.

One potential flaw, however, is that sigmoid curves make the probability of these extreme values fairly high, even if the extreme values themselves are constrained. Also, many of the formulas were purely empirically based, without necessarily a strong basis in probabilistic human output.

Plans for a new model

• Replace sigmoid functions with Gaussian distributions. Picking randomly from these distributions ensures probabilistically that extreme values are only chosen a small fraction of the time. Player skill sets and statistics could be used to weight these distributions to alter player performance.
• Incorporate refined coaching strategies for working with the play clock. Hurry-up offenses, run-out-the-clock offenses, and other measures could be used to better mimic the flow of the game.
• More to come...