Lots of hasty attacks by AI

[Stamb]

Road to Leningrad. Axis AI at 100.

AI is making multiple hasty attacks per hex that are useless and only then it tries to do a deliberate one. Same pattern in all turns.






General overview



I would expect maximum of 1 hasty or straight deliberate attack as CV of defenders are visible when units are close to each other. There is no sense to make hasty attack against 10-15-20 CV of the defenders.

Also ai is not clearing isolated divisions and it is possible to cut all ground logistics with them. Only then AI reacts (teleports?, as they appear from the side of Koenigsberg) some units to finally finish this divisions.


Is it acting the same in GC?

Soviet AI is doing deliberate attacks at least in the winter.
But there was no attacks, except one in the other seasons at ALL! Even when there is a huge encirclement and some Axis units have around 3-4 cv in defense, and Soviets could easily reopen the pocket.

[Joel Billings]

Do you have some saves you can attach from the start of an AI turn where I can run it and see what you are seeing? You can attach here or email to 2by3@2by3games.com. Thanks.

[Stamb]

Do you mean with hasty attacks?
Or that i can disrupt logistics with isolated divisions?

[Stamb]

Here you have it:
https://ufile.io/r8mx7x9b

You can run this scenario and you will see it. It is repeatable.

[Joel Billings]

Sorry this got overlooked in the holiday period. In a case like this, feel free to bump it up if you don't hear anything from me about it after you've posted saves. I ran your turn 9 save. In the original run you had 7 deliberate attacks and 15 hasty. When I reran it, I had 11 deliberate and 12 hasty attacks. So it's possible that some AI tweaks have already had some impact on this. I ran turn 8, which originally had 7 deliberate and 14 hasty attacks but now has 10 deliberate and 16 hasty attacks. I will send these to Gary to see if it's doing what he wants, or if he wants to tweak it more to reduce hasty attacks, especially when the AI isn't given any help. Thanks.

[Stamb]

Awesome. Did not want to annoy you/other people from a team with a bump. But I will use this feature if the future.

If it requires too much tweaking then IMHO it would be better if AI uses deliberate attacks all of the time, except for the opening turn (it is customized by a devs, right?). In the worst case deliberate attacks will route/shatter enemies which is much better than hasty attacks that just drains CPP.

Thanks for looking into it.

[Stamb]

Were there any changes in 0.19 beta regarding this issue?

[Joel Billings]

No, but he did make a change that we've been testing since 1.02.19. In your saves it reduced hasty attacks by about 40%. Thanks to the save Gary found a bug in his logic that had the AI planning based on a full stack hasty attack but then going ahead with an attack with just one unit in the stack. I can't claim it's made a huge difference in our initial tests, but it does seem to prevent some of the very bad hasty attacks that were happening. Assuming it continues to check out it will be in the update after 1.02.19. Thanks for the saves.

[Stamb]

Awesome. Really happy that it will improve Axis AI!

[Stamb]

I do not know how AI behaves later on, but it has problems with closing pockets.

It tries to use hasty attacks. Too much of them.

1.png (2.33 MiB) Viewed 262 times

While in some cases it makes deliberate attack while hasty one would be more than enough to do a job.

2.png (625.04 KiB) Viewed 262 times

Is AI "looking" on a CV of the units that it is trying to attack? If it does not see a CV then it might be better to use make it cheat and see it. Or maybe use deliberate attacks in such cases.

[Stamb]

Saves

[Joel Billings]

I do appreciate the saves in this and your other thread re the AI weaknesses. I will ask Gary if he thinks he can improve the issue with doing too many hasty attacks the next time he works on the AI. I think he's more likely to be interested in saves for normal front line situations, than these rear area attacks though. I know he worked a lot on the rear area issues but we had to move on as it just didn't seem like it was a solvable issue as I discussed in the other thread.

[Stamb]

Is it possible to make a script that AI for the first 2-3 turns make deliberate attacks, always. It would fix an issue. Thanks for looking into it.

[Joel Billings]

Probably not as easy as you might think, as Gary has many different attack routines, so he'd have to find all of them and add code to try to avoid hasty attacks before a certain date. Of course that would not be for turn 1, and can you really make a blanket statement that the computer should always only deliberate attack on those turns?

[Stamb]

...can you really make a blanket statement that the computer should always only deliberate attack on those turns?

No. It is just a suggestion. I guess that it is better to make 1 deliberate attack than 2-3 hasty one. With multiple hasty attacks it is bigger CPP loss and MP loss when you sum all of the MP for a hasty attacks it equals or exceeds cost for a deliberate one.
I did not try to offense anybody...

[Joel Billings]

No offense taken. I was just asking if you thought that one could say that it would always be better for the AI to deliberate attack on those first few turns. I don't know the answer to that question, but it sounds like you think it would be better if the AI would always just do deliberate attacks on turns 2 and 3. I don't think the AI is launching the hasty attack with the idea it will do several hasty attacks. Something about the routine is not realizing that the hasty attack is not a good one to make. Although it's also possible that there is enough randomness that maybe sometimes it would work. On the isolated units, Gary is probably counting on the isolation impact to hurt the defender sufficiently to let the AI win the battle, but maybe the ai routines are over valuing the impact of isolation on these battles.

[loki100]

might be related but there is the quirk of under-reporting of cv for formations on the first turn of isolation. Its a bit situational but if the isolated units can draw off a depot (and for T2 in particular this is common in the border battles), then their real cv is often above the shown value (as that factors in the full impact of isolation).

its one of those things as an Axis player you need to keep in mind and its inevitable a few hasty attacks will go bad as a result.

So the AI could well be keying off display/estimate cv to gamble on a hasty attack?

[Stamb]

My guess is that there are some problems with an estimation of a battle outcome.
For example unmodified CV before a battle is 40 for an attacker and 100+ for a defender.
Even in isolation it is not enough for a win.

But AI launches such attack and gets a hold.
And on the next attack ai is doing the same.
And then repeat until defenders surrender.
But it cost 3-4 or even more hasty attacks.

That is why i told that it would be better to make deliberate attacks if it is too hard to tweak hasty one.
(on the first turn hasty attacks with regiments against FZ are fine, but i think it is scripted and is not a typical AI behavior)