search

WorkingRobot / Craftimizer

The best FFXIV crafting solver via hardware accelerated Genetic MCTS.
MIT License
16 stars 5 forks source link

Unnecessary Meld #6

Open zocke1r opened 9 months ago

zocke1r commented 9 months ago

I really enjoy using the solver but i noticed that it tends to recommend using a meld even though it could finish the craft with the remaining durability. As such i would appreciate if that minor flaw could be solved.

Version: v2.0.2.0

WorkingRobot commented 9 months ago

I'm assuming you mean Master's Mend here? The reason why it recommends that action is because it scores simulations based on the weights you give it. The reason why it thought that's a better idea than to not do it is because the penalty of having a lower durability is worse than the % of your CP it uses and the extra action it needs to take for it. Remember that this isn't a magical all-knowing algorithm that knows exactly what you want. If you don't want to edit your macros after they're suggested, that's still partially on you. Feel free to mess around with the weights and settings until you get the results you want. image

Mystic-W commented 9 months ago

image I messed around with the weights a lot, pretty much tripling the weight of steps and halving that of durability. However, the synthesis helper still does this sometimes. I think an implementation for finishing the craft with the remaining durability when possible would optimize the plugin's efficiency a lot. Having it calculate durability costs of the remaining actions on every step, then checking that against the quality/progress could be a way to do this.

zocke1r commented 9 months ago

Yeah i did something similar with my weights image and now the solver even suggests stuff like using Great Strides after already maxing the quality, just to dump CP. Is this an issue where the solver gets stuck in local maximum, due to cp being so worthless and doesn't find that removing this one step would increase the score? I think having a post solve action that removes all "unnecessary" steps, which dont change the craft outcome (success and quality), would solve the issue and improve the recommended solutions in general.

WorkingRobot commented 8 months ago

That does seem like a good idea, but I think using a heuristic and comparing rlvl to clvl would be better. For example, if the difference is really huge, I can change the weights appropriately like that.

zocke1r commented 8 months ago

I assume rlvl means recipe level and clvl crafter level. So you want to change the weights in case of an easy craft, so that the solver prefers the shorter solution? If yes, I don't know if that would help as i had the issue with unnecessary on non trivial crafts as well.