If I take a from-scratch set of weights like Pawn string or Noxxic’s recommendations and want to put them in to create a strategy to compare with AMR’s machine learning strategy…
How do I choose Offset?
I have seen AMR suggest somewhere in a help tip that you can play around with the offset and do different sims.
But do you start at 0? Do you change by 100,000 each way? by 50,000?
What does offset even influence and mean?
Do you use suggest starting with a similar machine-learning gearing strategy and then toggle it to stat weights and then change only the entries you care about to build a new one starting with offset 0?
Suggesting we just play with different offsets is a pretty wide trial and error excercise.
The “offset” is essentially the intercept were you to treat the stat weights as a linear function – that is essentially how people use stat weights.
For DPS specs, usually a negative number will work best. If you press that “benchmark” button, it will test it. Just keep tweaking it until the benchmark gets as low as you can get it.
I wouldn’t sweat it too much though… the important thing is that your custom weights rank gear the way that you want them to. Once you go down the route of hand-entering stat weights, you are saying “I don’t really care how well these weights predict DPS or match a simulator… I just want them to do something specific that I want it to do.”
Hmm thanks. I was hitting the Benchmark button and nothing was happening.
I don’t want to have to use stat weights. But I am feeling like in hybrid fights that have mostly single boss target and some AoE there might not be a decent machine learning model if the single target and the mythic+ gearing strats lead you to something too far in either direction of what we need.
You could try spellblade – that’s sort of in between the more AoE-heavy Mythic+ and pure single target like Krosus.
1 Like