Backgammon Rollout Settings in GNU Backgammon

Backgammon rollouts are used in backgammon analysis, particularly in computer-based analysis using programs like GNU Backgammon. Backgammon rollouts would ideally simulate all possible subsequent moves to come with the option that has the highest probability of winning. This would take an infinitely long time. Therefore GNU Backgammon has backgammon rollout settings that terminate the rollouts according to specified conditions, achieving a balance between accuracy and time.

The backgammon rollout settings in GNU Backgammon consist of three parameters, which are minimum number of trials, maximum number of trials and joint standard deviation (JSD). The backgammon rollouts will stop once the moves are 'x' JSDs apart, where 'x' has to be specified. The larger is 'x', the lower will be the error, for the same level of the other settings. The most common value for 'x' is 3. In no case should 'x' be set at less than 2.

In some simulations the set value of 'x' is achieved fairly quickly. Experts have found that there is a high chance of error in such cases because a sufficient number of options have not been explored. Hence there is the inclusion of a minimum number of trials in the backgammon rollout settings. The larger is the minimum number of trials, the lower will be the error. In some simulations the set value of 'x' is not reached for a very large number of rolls. Hence in order to ensure that the backgammon analysis does not take an unduly long period of time the backgammon rollout settings include a maximum number of trials.

For novice backgammon players the typical backgammon rollout settings in GNU Backgammon should be minimum 300 trials, maximum 10000 trials with 3 JSDs. As the backgammon player gains expertise he can make adjustments to suit his needs.