View Single Post
Old 08-13-2019, 12:14 AM   #25
Newly Registered User
Join Date: Aug 2009
Posts: 32
Thanks: 5
Thanked 0 Times in 0 Posts
Jossy is on a distinguished road
Re: Lookup the value from a previous record

Thanks for picking up the mantle CJ.

Do you really need a query that presents all rows for all players? or are you only interested (at any one time) in the players in an upcoming match?
There will be two main outputs needed:

1. The ability to see the latest ELO rating for a given player. In this case it wouldn't be necessary to calculate all rows for all players. However, the way the calculation works it will probably be close to all records as you will need the scores of each of their opponents for all their previous matches. In order to get those you will need the scores of all the opponents' opponents and so on and so on.

2. A comparison of the accuracy of the forecast results to the actual results, e.g. did the model predict a winner with 75% accuracy when the player was calculated to have a 75% probability of winning? I will then add in a range of additional metrics and calculations to try and improve this accuracy, e.g. the number of points won in a match.

Why is the P2 CurrentELO hardcoded? Why is the P2 K overall value hard coded?
CurrentELO and K are only hardcoded for Djokovic's opponent. Otherwise you need all their matches in the dataset to calculate cELO and K, and then all their opponents' matches and so on. The file size would have been above the 2Mb limit.

Is there any relevance of the location - are you saying if player A plays player B the 'projected outcome' would be the same regardless on where they play
I assume by this you mean the name of the tournament? No - just a bad habit of always seeing it so adding it to the query! At a later point I will want to create a query to calculate ELO for a specific surface, e.g. grass or clay, but once I figure out the 'overall' method then I'm sure I can adapt it.
Jossy is offline   Reply With Quote