afknst / gxb2_battle_simulator

Battle simulator for GxB2
GNU General Public License v3.0
0 stars 0 forks source link

EA and tier list testing #5

Closed afknst closed 2 years ago

afknst commented 2 years ago

So Sala and I (I didn't push my part yet) wrote some functions to create and change a team randomly (for OCE), which would be helpful to implement EA.

(Btw it didn't work well in OCE since they added more debuffs and it becomes harder to generate a successful configuration.)

I plan to use EA for tier list testing (i.e. to find good teams and good setups for each girl), but I need your advice.

The design (draft):

  1. Pool (population): including all new girls and some strong girls, full lab, all pink antiques and gear, nc3 crystal, full excursion, and maxed servants. No outfits/skins.
  2. Generation: including a% top teams from the previous generation, b% those teams with mutations, c% newly generated teams. I think a% = b% = c% = 1/3 would be OK.
  3. Mutation: same girls with random setups.
  4. Score: I think we can simulate JV. It should be faster than an all-play-all tournament (which we used to do for the previous tier lists).

What do you guys think?

Salamandinay commented 2 years ago

In my last update in OCE, I use it by waiting for 2/3 wins, then see which one can do the most damage in the 3rd battle and remove the hardest buff(s) and that usually clears.

I've never implemented something like that, so I don't really have a good idea. But I don't like having all pink antiques, not even the largest whales do that yet so the results won't be as useful.

Also, I can get the setups of all teams in UL if that can be used in any way. (and is there a UL battle simulator or is it too complicated with the following battles?)

afknst commented 2 years ago

For pink antiques, yeah I think we can limit the sum of pink stars for each team. E.g. if a team has two P3 EDs and four normal EDs, then its sum is 3 + 3 = 6, something like that. Which number do you think is appropriate here?

Also for the servants, I think UC100 might be too high for ordinary end-game people. Maybe UC80 is better?

To simulate UL, we need to:

  1. Change the battle type to ARENA_ALL_SERVER
  2. Return die_info and hp when generating a battle report
  3. Implement the substitution mechanism for girls and servants
  4. Pass outfits status to report creator if you also want to simulate the outfits system:
    dressAttrsA = data.dressAttrsA, -- { Courage, Charisma, Knowledge }
    dressAttrsB = data.dressAttrsB
Salamandinay commented 2 years ago

If it's too hard to decide, we could just run it twice, with pink antiques and then without, and have two lists.

UC80 is fine.