Teambuilder: Add EV optimization checker

[Karthik99999]

https://www.smogon.com/forums/threads/smart-ev-management-in-the-builder.3736254/

[pyuk-bot]

It’s not necessary to use a brute force approach, or even any kind of iterative approach. The plus, if it moves at all, will always optimally move to the highest raw stat that’s been invested in. The point of moving the plus is to get the highest absolute stat increase from the 10% bonus, regardless of whether that comes from the base stat or EV. The minus, similarly, should only move to the lowest raw stat that’s been invested in to at most what it would be at 252-. Furthermore, the plus can never move from a stat that is boosted further than it could be at 252 neutral (you at least thought of this) while the minus should never move from an uninvested stat as there are no EVs in that stat that can be redistributed for optimization purposes.

The focus of this optimizer should be on saving EVs and not on increasing stats. Sometimes improving the stats happens as a bonus when the stat we’re already moving the plus to either happens to be 1 less than a multiple of 11 or is a high base stat that was originally invested to less than what it would be at 0+, but if we can either provide the exact same stat spread for fewer EVs or a slightly stronger spread for the same EVs, we should be giving the spread that uses fewer EVs every time and leaving it up to the user to determine where the remainder goes. After all, once they reinvest their saved EVs after optimizing their spread, the optimizer will automatically run again anyway.

[Karthik99999]

It’s not necessary to use a brute force approach, or even any kind of iterative approach. The plus, if it moves at all, will always optimally move to the highest raw stat that’s been invested in. The point of moving the plus is to get the highest absolute stat increase from the 10% bonus, regardless of whether that comes from the base stat or EV. The minus, similarly, should only move to the lowest raw stat that’s been invested in to at most what it would be at 252-. Furthermore, the plus can never move from a stat that is boosted further than it could be at 252 neutral (you at least thought of this) while the minus should never move from an uninvested stat as there are no EVs in that stat that can be redistributed for optimization purposes.

Yeah I know brute force is inefficient here. If anyone can implement this without brute force that would be cool, I just did this cause it was easier.

The focus of this optimizer should be on saving EVs and not on increasing stats. Sometimes improving the stats happens as a bonus when the stat we’re already moving the plus to either happens to be 1 less than a multiple of 11 or is a high base stat that was originally invested to less than what it would be at 0+, but if we can either provide the exact same stat spread for fewer EVs or a slightly stronger spread for the same EVs, we should be giving the spread that uses fewer EVs every time and leaving it up to the user to determine where the remainder goes. After all, once they reinvest their saved EVs after optimizing their spread, the optimizer will automatically run again anyway.

Not sure what you mean by this, since I am trying to return the spread that saved the most EVs. Did you find a case where it returned a spread with higher stats instead of more saved EVs?

[pyuk-bot]

Try this set:

Trapinch
Ability: Hyper Cutter
Tera Type: Ground
EVs: 204 HP / 252 Atk / 52 Def
Lax Nature

The correct output would be this, saving 56 EVs:

Trapinch
Ability: Hyper Cutter
Tera Type: Ground
EVs: 204 HP / 144 Atk / 104 Def
Naughty Nature

This code will instead output this, saving only 28:

Trapinch
Ability: Hyper Cutter
Tera Type: Ground
EVs: 204 HP / 144 Atk / 104 Def / 28 Spe
Brave Nature

This is because if the current nature attempt doesn’t beat the current best in EVs saved, it falls back to checking if it beats that spread in stats, and if it does, it saves that spread as the new one to beat, and all future attempts will have to invest extra EVs to match the higher stats. However, if you only fix that and make it simply discard attempts that have fewer leftover EVs than the current best, it will keep outputting the Brave set because the very first set it generates is this:

Trapinch
Ability: Hyper Cutter
Tera Type: Ground
EVs: 204 HP / 144 Atk / 104 Def / 56 SpA
Adamant Nature

The reason these over-invested spreads happen in the first place is that the brute-forcer tries moving the minus off of a stat that was lower than it would be at 0 with a neutral nature, and if that happens to coincide with the plus being moved to the correct stat, it ends up with spare EVs that it can afford to use to shore up the new minus stat to bring it up to 0 neutral level. In this case, the very first nature it tries, Adamant, sets an overly high standard for Special Defense that prevents Naughty, the actual optimal nature, from being selected as it can only match the Adamant spread for exact stats and EV usage.

These problems are best avoided by simply not brute forcing in the first place.

[Karthik99999]

Ah, I see. Honestly could just rid of the stat check, it finds the right set without it

The only reason I had it in the first place was for this Amoonguss set

Amoonguss
Ability: Effect Spore
Level: 50
Tera Type: Grass
EVs: 252 HP / 100 Def / 156 SpD
Bold Nature

Using a +Sp. Def nature on this set leads to no EVs being saved but you have a Sp. Def stat of 121 instead of 120.

These problems are best avoided by simply not brute forcing in the first place.

I'll try to investigate later this if I have the time. Again, you're more than welcome to make a code suggestion here, as I would also love for this to not be brute forced.

[pyuk-bot]

Karthik99999#2

[DaWoblefet]

Good work Karthik and pyuk!