Time Series Ratings (ISSUE)
Closed this issue · 1 comments
I have implemented your rating system but there seems to be a problem with the way that it rates the players and Im yet to figure it out exactly. Let me give you an example: Let's assume player x has been involved in 10 competitions, it seems that the 9th competitions rating INCLUDES the results of the 10th competition (even though the 10th competition has occurred after the 9th). This means that the rating in the first competition includes some data of all results of future competitions which means the rating is not time series - it contains data that otherwise should not be available as of that time. Do you have any solution so that the ratings that are produced for each competitor in the csv file output do not include future results, and only past results - therefore making it time-series, and useful?
More specifically, competitor x has a rating_mu of 1675 as at competition n and time t, if i add a new .json file for a later competition n+1 at time t+1 which competitor x is involved in and wins, the rating_mu for competition n changes to 1815 (+140), even though competition n+1 is at time t+1 and should not change the rating at competition n as data at time t + 1 should not be visible at time t as it is not time series. Does this make sense? Apologies if i have not explained it well
All results are 0-indexed - not likely the problem. Any solutions would be greatly appreciated. Thanks