Search the Community
Showing results for tags 'regressing from datasets'.
@admin I have the following suggestion to introduce a way to generate data for modeling a meaningful Battle Rating. Taking data from the logs you have now will result in a bad starting position for any iterative process, whether it is applied by hand or automated. It will require many iterations to get to a stable solution. The reason for this is, the fact that you have (if I understand it correctly) data that encompasses different phases of the game with several buffs and nerfs along the way. This would at least explain why the Wasa has the same BR as the Bellona, despite being worse in every aspect except for a light advantage in turning rate and a negligibly better sailing profile. Also ships not being used does not mean they are weak (e.g. the Agamemnon,) there are just better alternatives that are as easy to obtain (why sail an Aggy if you can afford and crew a Bellona?) My suggestion is as follows: Create a build of a battle instance. Load one AI ship on both sides, with equivalent starting postions (eg. facing each other at a beam reach). Pair each ship with every other ship. Run the battle as often as you can, considering resources, for every pairing. This will create a set of data that shows each ships relative performance without taking into account different skill levels and knowledge stuff or fleet composition etc.. Use this ranking as a base for a BR order, which can be adjusted on data that will be accumulated from then on. You will get a starting point for the proess that resembles the current situation without the many tunings of the past. EDIT: This post is solely about the way that data is acquired, not about how it is used to calculate BR!