In this article, I take a look at 3 different methods for finding your best gear:
- Best in bags
- Sim every combination (batch simulation)
- Hand pick some gear combos to sim (batch simulation)
Hi Cat. Grats on the tool and the approach. Love to see this kind of innovations in this space. But (you know there was a ‘but’ coming, didn’t you), as I understand ‘Best in Bags’ relies on the computed ‘Stat Weights’. Now, the problem I have with this is that ‘stat weights’, while being accurate in at a very small distance form the current gear point, in many cases do not predict very well the outcomes when extrapolated to distances typical of even replacing one item such as a ring or neck (I main Fire Mage, so this might be class specific). I guess the same reasoning behind AMR’s ML approach applies here as well. The ‘fitness landscape’ is rugged, so extrapolating from a local gradient might with significant probability land you somewhere far from optimal, and even worse than where you started from. Could utulizing data from ML be a way to go to improve ‘Best in Bags’?
Best in Bags uses the machine learning data, not stat weights, and we’re also rolling out “version 3” over this week and next that uses a much larger data set that can adapt to your talents, choose legendaries more reliably, etc.
Thx for the clarifications, I misread the article then.
The article is a little bit out of date. The new Machine Learning Data is vastly superior (at least in terms of technique, if not outcome).
The practical how to use remains the same.