glandfried
commented
2 years ago Hi @zass30,
The equation 11 of the documentation [1] models the forgetting rate.
The parameter History(... gamma = g) is the forgetting rate, as used in code 7, 11. This rate will govern all players without prior. We can set different rate initializing the parameter priors, History(..., priors), which is a dictionary Id - Player(Gaussian(mu, sigma), beta, gamma) class.
Note that the class History uses the temporal distance between events to determine the amount of dynamic uncertainty to be added between games. So the equation 11 is implemented as:
The amount of time between events is set out by the parameter times History(..., times).
At the moment this is the only forgetting function implemented in the three packages.
[1] https://github.com/glandfried/TrueSkillThroughTime/releases/download/doc/landfried-learning.pdf
Let a user specify a manual uncertainty decay function. This could either be run at set intervals in a game's history for all players (for example as a daily job), or else called ad-hoc when querying a given player's ranking, or else called as a callback when parsing new games. As an example, a basic decay function could be that sigma is multiplied by some factor, like 1.1, for every day a player has not played.