Greasy Fork

Please Increase winRate for Auto Pack Selection in Brawls

If all battles have a winRate of 0.85, then there is a 20% chance of losing all 5 lives; 0.90 gives a 5% chance; 0.95 gives a <1% chance. With Eden appearing in over half the brawls, the winRate is too generous. Actually, the problem is not the test of winRate, but premature termination of testing each pack.

The auto pack selection makes two go decisions on the high end:

First, 10 wins in a row ( i >= 10 && stat.winRate == 1.0) allows packs with too low a winRate. At p=0.85, there is a 20% chance of 10 wins accepting a battle that should be rejected (i.e. winRate < 0.85). To statistically match the winRate, the number of wins in a row should be:

So while accepting after 10 wins in a row seems like a good way to avoid wasting CPU time, it lets packs through with less than the acceptable winRate.

The second go point is (stat.winRate > 0.85) [18+ wins of 20, since 17/20 = 0.85]; effectively a winRate of >= 0.90. This should give a less than 5% chance of losing all 5 lives, but it requires 22 wins in a row to statistically ensure.

That's not the problem, but rather that there are no packs with new Titans among the packs.
I've added three full Earth, Fire, Water elemental packs to the list.
If there are any other packs that should be added, please suggest them.

A win rate of 85% is quite sufficient for this event, but I adjusted the calculations slightly to make the search for a pack with a suitable win rate more accurate.
In general, most Titan battles aren't random, so this doesn't matter much.

Just FYI, the changes you made to Brawls have resulted in a DRASTIC reduction in speed with no noticeable improvement in the results. I was able to win only 21 battles before running out of lives, and it took about 5 minutes. I used to run out of lives previously too, but at least it was over in 30 seconds. :-P

Not complaining, just informing you of changes to the way the system works.

It now tests up to 22 battles instead of 10, aiming for 90% winRate. It also aborts after 5 losses instead of 2, so it should take about twice as long to test packs. Probably aborting after 12 wins/0 losses or 3 losses for 85% was good enough, but the problem may be that there were no good packs. I used Avalon/Asherona/Verdoc/Araji/Eden without automatic selection and did well, although it seems that some defenses have already been moved to defeat that attacking pack.

Maybe auto generation of packs? It can get a little complicated if done efficiently. Packs can have up to 3 super titans, so if the mandatoryID is a super titan, there are 7242 raw combinations, and 7310 if the mandatoryID is not a super titan. But I don't think packs without super titans are meaningful, so down to 4930. And then filter by requiring at least one totem.

The easiest method would be to iterate through all 33649 packs and then filter those without the mandatoryID, those with >=4 super titans, those with 0 super titans, and those without a totem.

Interesting, thanks for the explanation.

I don't think it's necessary.
The current combinations worked well before.
I think adding a few effective combinations with the new Titans for each of the three elements would be enough.

It seems like it's already lasting much longer; adding more combinations would make things even worse.