Why is the Maximizer doing this?

[fronobulax]

I am maximizing on just meat drop. I always thought the negative number in parens meant that the change was going to lower something. Indeed, when I did execute the suggestion Meat Drop did get lower.

My knee jerk reaction is to call this another unintended feature but it may be that, with the GUI and other changes, there is some setting that is not to my liking.

This is a general question about why for a simple maximization it is recommending something it knows gives a suboptimal result.

@MCroft

Is there a bug behind the identical scores?

 

[MCroft]

View attachment 10079

View attachment 10080


I am maximizing on just meat drop. I always thought the negative number in parens meant that the change was going to lower something. Indeed, when I did execute the suggestion Meat Drop did get lower.

My knee jerk reaction is to call this another unintended feature but it may be that, with the GUI and other changes, there is some setting that is not to my liking.

This is a general question about why for a simple maximization it is recommending something it knows gives a suboptimal result.

@MCroft

Is there a bug behind the identical scores?

One thing I want to do is put a verbose section that shows all the weights for an item in the maximizer, so we can see what it's comparing against. UI, wise, it would probably be a 3rd item in the context menu. It's on my long list, not my short list.

I assume that you don't have a back item that's worse, like "Cloak of -90 meat drop" (whatever that would be).

Questions I have:
1: What do you have equipped?
2: What is the output of maximize meat drop, 2 dump ?
3: What is the output of meat drop -tie?

We have a test that someone recently marked "Ignore" in the MaximizerTest, specifically for leaving a slot blank.

// Sample test for https://kolmafia.us/showthread.php?23648&p=151903#post151903.
@Test
@Ignore("Currently failing")
public void noTieCanLeaveSlotsEmpty()

What I've been saying to ppl who ask Maximizer questions on Discord is "eventually it will make enough noise that someone will say 'Alright, I'm going to see how it works and then fix it'." That may be now.

I think that if I get time this week, I'm going to bite the bullet and make a damn flowchart of the maximizer maximizing, so I can understand it and then stick that documentation in the wiki for future reference.

 

[fronobulax]

The keep lines in the screen shot show what else was equipped. I think the Bjorn was there but I will have to check.

The single maximization generating a suggestion that lowers the value is a big flashing red light for me. I am fishing for a reason besides a Maximizer bug.

I have handled constrained optimizations before. Seems to me it would be nice to have free, reliable third party code to handle it so we can focus on formulation. That said, multiple targets, tie breakers and outfit bonuses don't fit into an easy formulation.

 

[the dictator]

There's two parts in those screenshots that stand out for me, and neither makes sense for a 'simple' maximization:
A) Somehow, the maximizer did only 624 calculations (mine does over 100k when I'm in aftercore, and if I abort it early, it will lead to such errors).
B) The maximizer claims you would lose 85 score when making that switch, when you lost only 10.
Edit: C) The fuzzy polar bears seem to have a similar thing going on in the second maximization.

 

[fronobulax]

There's two parts in those screenshots that stand out for me, and neither makes sense for a 'simple' maximization:
A) Somehow, the maximizer did only 624 calculations (mine does over 100k when I'm in aftercore, and if I abort it early, it will lead to such errors).
B) The maximizer claims you would lose 85 score when making that switch, when you lost only 10.
Edit: C) The fuzzy polar bears seem to have a similar thing going on in the second maximization.

That character has never ascended and is a display case completionism. So the 624 calculations could be because of a limited inventory.

I did not notice the PB ears but that seems to be a second case.

Maybe I should break down nd run some dumps.