Guardian Druid Leech oddities


Came in here to ask about unusual prioritization of leech for my druid, but I think I solved it. Along the way, however, I found some aspects of the interface I think could be improved.

The gearing strategy that initially produced the question is this one. In the stat weights box, it does clearly place a high value on leech, but that box is apparently disabled because I’ve selected machine learning. If those values are still relevant, even when machine learning is chosen, they should be shown as such. Leech is also not included in the machine learning examples, which is defensible (how would you farm for it?), but it furthers the impression that the leech weight below is not being used.

Selecting the current default Ursoc script and converting it to machine learning produced results more in line with my expectations. The only difference between these two strategies are the version number and the stat weights. I doubt that the minor version difference produced the difference I see, so I’m left thinking that the apparently disabled stat weights do, counter to my intuition, matter very much. I would have thought that machine learning disregards all user input except for the rotation, boss, and so on at the very top of the page.

That leads to my next concern. Why can’t those values be chosen on this page? The current design forces the choice between the default Mythic+ and Ursoc scripts on the character page, and then requires each to be opened on a separate gearing strategy page to enable machine learning. I feel like the choices of script and gearing strategy are part of the same workflow and should occur in same place.


The machine learning option factors in all stats (primary, stamina, leech, avoidance, secondaries, etc.). The display is just showing you the secondary stat balance to give you an idea of what it’s going for. In the future we plan to tweak the UI to visualize what it is doing better, but for now think of it like this: Machine Learning takes all of your stats and tries to predict your NPS as a function of all of them. This predictive model can be complex and describe a multi-dimensional “surface” that is hard to visualize, but the important thing is that it predicts well. Better visualizations are in the works so you can “see” what it’s trying to do a bit better… but it is hard to do.

The version number could have a big impact, depending on what was updated between the two. Maybe a spell was tweaked, fixed, changed in-game.

The gearing strategy picker on the optimizer is just a list of available strategies, grouped by boss for ease. When you pick one of those strategies and then edit it to enable machine learning, you are creating a separate, custom strategy saved to your account, which then shows up in the gearing strategy picker, grouped under the boss it was made for.


Sure, I understand the gist of the optimization process. My big question is whether the selection of machine learning is still, in any way, using the numbers from the stat weight box, or whether they’re completely ignored. If those values are being used, they should not be grayed out. If those values are not being used, then the version number is the entire difference between the two strategies.


The stat weights are not used if you are using Machine Learning.


Okay, that is what I would have hoped, which tells me my guess about version number was wrong.

Thank you.