Page 48 of 49

Re: Promi 2.10

Posted: Thu Jul 01, 2021 1:55 pm
by TheNightPanda
Arpheus wrote:
Thu Jul 01, 2021 11:19 am
It's so great to see your enthusiasm about our ai's :) One thing I'm curious about. Which settings do you use for 8xspeed and how does it affect the number of passes per second? You may not know this but maybe someone else do. With the new micro capabilities of ai's like odette this has huge impact on performance on DE.

Re: Promi 2.10

Posted: Thu Jul 01, 2021 4:15 pm
by Arpheus
TheNightPanda wrote:
Thu Jul 01, 2021 1:55 pm
It's so great to see your enthusiasm about our ai's :) One thing I'm curious about. Which settings do you use for 8xspeed and how does it affect the number of passes per second? You may not know this but maybe someone else do. With the new micro capabilities of ai's like odette this has huge impact on performance on DE.
I had a feeling sometimes in the past that 8x might make micro less precise but I was not really sure about that and lately I haven't seen issues with delayed movements or attacks or things like that and even deer pushing worked fine. Is it a thing to consider? Doing tests at 2x speed or less would take considerably longer. And with settings you mean the game settings?

For the tests I had the following settings:
Arabia Random Map
Map Size: Tiny (2 players)
AI Difficulty: Extreme
Resources: standard
Population: 200
Game speed: fast -> in game I click + twice and it goes up to x8
Reveal Map: normal
Starting Age and ending Age: standard
No Treaty and Victory standard.
None of the advanced settings is checked and I always did mirror civ. One time Teutons rest of games were Spanish

I have high hopes that some AIs will reach Voobly Barbarian's level one day for DE which motivates me to give feedback to help them improve. I am sure a large part of the playerbase would enjoy even stronger AIs.
Still hoping that one day Barbarian will be officially released for DE to also include the new DE civs and the diverse changes. Promis DE AI is not too far away from the unofficial DE Barbarian AI. Odette is making progress with frequent updates but is not yet at the level of Promis DE AI at least from my testing at 8x speed.

I just did a test with 2x speed with odette vs barbarian. Odette went for scout rush (haven't seen that before) and the micro was very impressive. It is possible that the deer pushing was more precise at 2x - it was almost flawless.
The scout rush killed quite a few villagers and cause lots of chaos and idle time. Barbarian then defended with a lot of spearmen and Odette had to retreat most scouts.
Odette made a good skirm army with a few spears to complement the scouts but lost all the momentum and advantage in that game to a defensive tower filled with 5 vills and some skirms + spearmen. The Odette skirms had problems with the spearmen in melee and the tower again had a big impact on the outcome of the fight.
I think the best way to deal with the towers in feudal age is really to attack from another direction where no tower blocks the way or men at arms but that is an investment.
After that loss barbarian literally destroyed Odette since it had no more army to defend.

Re: Promi 2.10

Posted: Sat Jul 03, 2021 4:06 am
by Promiskuitiv
Thank you so much for the extensive feedback Arpheus!

I'm very sick right now so I won't be able to create a lengthy response now but I'll update this/create a new post regarding this later and hopefully soon you'll see some changes in Promi soon that are resolving several of these items. :D

Re: Promi 2.10

Posted: Sat Jul 03, 2021 8:22 am
by Leif Ericson
Hope you get better soon Promi!

Re: Promi 2.10

Posted: Sat Jul 03, 2021 9:28 am
by Arpheus
I hope its nothing severe. Best wishes for a fast recovery - get well soon

Probably a lot of players and me of course too are really excited for new AI updates. We really need an official AI that plays at a very high level on extreme difficulty in the Definitive Edition. I already really enjoy playing against your AI since it is the strongest one which can play all civs at the moment. I am also really glad that there is this unofficial barbarian AI to train against - this is a reliable training/testing partner for your AI. Thanks to TheMax for making such a great AI and I heard on voobly it is stronger even!

This time I want to share some observations from Player vs AI play.

The last days I have been trying to defeat 2 Extreme AIs at the same time but so far I have only been able to do so a few times with Slow speed setting (I am not one of the super fast players with crazy reaction speed) and a few reloads during the game when things went too bad. My goal is to beat 2 Extreme AI more and more reliable without loads and at casual and then normal speed but that is still a way to go.

If you can prevent or stop their trading in 1v2 you have a chance to win. If not and they get unlimited gold from trade its over very soon in lategame. You just can't defend from unlimited treb/siege ram/massive unit pushes from 2 different angles. And your AI knows how to push with trebs - it dives forward with the army to make it really hard for you to reach these trebs - even using bombarb cannons to counter the trebs is difficult that way which is great play from the AI. An Improvement here for the AI would be if it keeps some halbs close to the Trebs to make it even more difficult to take down the trebs.

Also early scout rushing is not so easy against your AI. You can get 3-4 kills maybe but the spears soon are everywhere chasing you around. If you divide your scouts into more than one group you might be able to distract the spears a bit but the AI already defends well here. The villagers running away causes idle but in this case also makes it harder to get fast kills and allows the spears to catch up fast.

You mainly get these few kills because the AI doesn't prepare for the scout rush. If you change that according to scout intel early feudal it will be hard to get value out of a scout rush which is good for Extreme difficulty - it should be as hard as possible.
If you prepare spears and place them in resource locations you could think about making sure that the defending spearmen never go too far away from the ressource location so that you can't lure them away too far. Splitting up the spearmen is viable in this case because spearmen are such a hard counter to scouts (be careful against sicilian scouts which take out spears 1v1 and in castle age pikes with light cav). That would also allow the AI to fight the scouts with vills + the spears and not have to retreat all vills from a resource location and cause lots of idle time when some enemy scouts show up.


For the future I could imagine that you could do advanced tactics like baiting the enemy into a strong defense with Towers or Castles where you can replenish fast and once you win that engagement you push forward and use the momentum you won from that engagement. Of course there has to be reasoning behind it. If you have a huge power spike it is worth using it but you also might loose the army if that power spike was not large enough and you fight under enemy TC/Tower/Castles.
Like TheViper says - never take a fight that you can't win. Even only slightly higher power on one side can lead to a snowball. But if you can't run away fight to the death to cause as much damage as possible.

Also making production buildings on other sides of the enemy base if that is possible in later stages of the game and flood from there too could be a really strong move.
And if you want to go to the extreme put the low HP vills around the TC instead of far outside at wood lines (only on extreme difficulty please). That will increase the challenge even more to get vill kills.

The one thing that will be hard to defend from in the early game for the AI is archer and archer/skirm rushes. You can't split your army or it will be easily defeated (snowball) so you have to react to these and follow them and keep your units together - but don't follow too far out because they might lure you away from the base to attack from another side.
Atm TheViper and other Pros can beat 3 of your AI because of very early feudal uptimes + archers only with godly micro. They dodge non-ballistics shots (and sometimes even ballistics) and that way destroy the enemy army which snowballs them to a win.

That might also be one of the advanced things for the future to get the Extreme DE AI to high elo levels. Start dodging more shots to make it harder to outmicro and split the own skirm/archers group in 2 commanded groups in a way that they don't all shoot at the same time to make dodging all shots from pro players a lot harder.
If they dodge the first shots they will probably eat the shots from the 2nd group and take at least some losses. You could make more than 2 groups but that could be unfair AI advantage and at a certain point impossible to win against.
Then you could even make the AI adapt to the enemy. If his units don't take damage from the first volley they probably dodge shots so start dodging shots too and make commanded groups for staggered shooting.
But this dynamic adaptation of micro would make it harder to measure how good you are if the AI plays better against better players than against not so good players. Maybe just add this staggered shooting and dodging to Extreme and only one of the 2 to Hardest.
If Pros can't abuse micro in the very early game they will have a really hard time and probably not be able to handle 3 AIs at the same time any more and will probably have to go for 1v2

Another thing that prevents the AI to win against good players is the lost momentum because it seems statisfied to attack some buildings when there is so much potential to deal lasting damage to a players economy by just diving into the base and harassing and idling them.

Here is a good example of a great game by a player doing 1v2 against your Extreme AI. You can see that it is already hard to beat the AIs and from this game you can learn a lot for AI improvements.
https://www.youtube.com/watch?v=2kd1ec_H2O0
At around 14 minutes you can see the AI breaks a wooden gate and has a free way into the economy but since it doesn't immediately see villagers it seems to be satisfied to stay there and shoot at a mill when it could do damage to the economy.
This behaviour is one of the main things next to wasting units to Towers/TC/Castles that holds the AI back a lot.
At around 14:25 or so the archers move in but shoot at a lumbercamp instead of denoying a TC to go up. AI should never shoot at buildings with archers unless maybe as saracens or in rare lategame situations or if they want to create a gap in a wall- but for that priorize wall --> house --> other buildings - never resource drop offs.
The moment the TC goes up they retreat to outside the enemy base and chase a single knight for no reason - he is faster so he gets away anyways. Not sure if there is an easy way to change this behaviour since it might be the same code as for moving to a unit to engage.

What they could have done: move into the base and when villagers are pulled back into a TC place archers just outside of the TC range to idle the farms (be careful for Lithuanian TC +range tech). If possible place the archers near the side where the other archers move to so they are as close to each other as possible to reunite faster if needed. Send a 2nd group around in the base and deny villagers from taking wood and gold and maybe get some vill kills. But keep the pikes close to the archers in case that the enemy brings army like the knights actually harassing the AI teammate.

But the way the AI actually handled it it was not even a threat for the player when the tons of archers were outside shooting a lumber camp and he could continue harassing the teammate and had time to mass knights to clean up the army out there.
Moving in there could have won the game in the long run - releasing the pressure from the teammate and idling the enemy. In a 1v2 game you can't idle your vills or your 2 enemies will get ahead.

At 15:48 2 archers see a knight and shoot him - following into the TC range which causes the full army to retreat because the one archer got shot at. Also room for improvement.

At 16:30 another massively wasted opportunity. 8 Knights push out against 20 crossbows + 6 skirms + 2 pike. The archers are not stacked and keep shooting at the tc while the other archers not too far away are getting killed. Also no move backward shoot move shoot micro from the archers also no focus firing down 1 knight after another.
At 16:41 already half the army runs away from the knights dealing zero damage for a while to the knights. 2 Knights at 60% HP and 2 other at around 80-90% - all others still full. Soon 2 more knights join the fight and later even more.
At 16:50 it finally fights back - seems to focus fire a bit but switches target too often. If it did the move shoot move shoot micro focus fire it would have been far stronger. At least it kills a few knights and fights back now.

If a Pro had played this engagement probably all 8 knights would be dead and a part of the army would probably still be there.
The AI has to first stay stacked - move stop micro and focus fire knights - only if the overkill is more than like 50% for a single shot volley it might start thinking about how to fire.
Then it has to take a decision. Is the way home short enough that it can save units and is that saving of units worth it. If not just fight to the death and kill as many enemy units as possible instead of offering free kills. Otherwise it will play into the snowball for the enemy.

That was again a wall of text but I think these observations can help in the long run to make the AI better and eliminate weaknesses.

You should really watch the game - great play by SteelYr and the AI did not bad with room for improvement :)

I am really looking forward to these changes! Just make sure to place these improvements carefully into the diverse difficulties from easy to extreme so that newer players don't get overwhelmed. Extreme in my opinion should not hold back and use every bit of advanced tactics like pros do but it should not use absolutely unfair 2.000 apm AI crazyness when microing units - which none of the AI has done so far.

Again best wishes for a good and fast recovery!

Re: Promi 2.10

Posted: Sat Jul 03, 2021 12:05 pm
by Arpheus
Here another thing that I experienced a few times which let me cheese a win. The AI often wastes far too many units to enemy castles instead of sitting back in front of the own trebs protecting them and waiting until the treb has done its work to then push in.

In some 1v1 games when I play the defensive approach I easily win vs a Extreme AI in lategame because it sends its full army to attack the fortified stonewall in front of my castle while its trebs shoot my castle from behind. While the treb takes its time to destroy my castle which is repaired and filled with ranged units my onagers or siege onagers MELT enemy units/rams which try to create a breach in the wall uselessly and my castle fire MELTS these units too with all the firepower from the ranged units inside the Castle.
So the AI basically completely wasted half up to 3/4th of its army because it was not patient enough to wait for the trebs to do their work before attacking. Since in 1v1 games gold is a finite ressource this means that this one big defense pretty much won me the game instantly because I am also on the neutral golds usually which the AI doesn't defend and care about until late game and now the AI just lost pretty much all its gold investment in one single battle.
After that the counterattack is pretty much a walk in the park since I lost almost not a single unit - maybe my BBC even destroyed the trebs since I usually build my castles at least 5 tiles behind the wall so that my Siege engineers BBCs can shoot trebs from behind the wall and I focus fire any ranged unit down first before shooting the enemy trebs with my BBCs so that the BBCs don't get sniped by ranged units.

This defense is very strong but if the AI at least didn't waste all units at the wall I would only have maybe stopped his treb push but the AI would still have it's army. Thiese suggestions are about how to handle fully stonewalled bases.

What it could do:
  • use 4+ trebs and the castle will fall eventually (AI already does that from time to time). But keep the own units in front of the trebs - don't bring them into enemy castle and onager range without a reason. Use bombard cannons or 13 range monks if possible to snipe/convert enemy onagers and enemy bombard cannons but move the army with the BBC to protect them for the duration. If enemy bombard cannons are too effective at taking down the own trebs move forward from time to time to shoot the bbc with ranged units - retreat back to safety once bbc threat is dealt with. Also snipe enemy trebs with own bbc. Repair the own trebs if the enemy tries to counter treb (important)
  • Just breach the walls at many places at once with siege rams / trebs / BBC / 3 petard (2 with siege engineers) or whatever and flood the enemy base and ignore the castle
  • Shoot enemy castle with trebs. Also make lots of petards (if population space is not a prob) - like 15 to 20 or so and hide them in the castle. Then petards leave the own castle and instantly the army moves forward with the petards (hard to focus fire them for the enemy - that would need insane micro) - blow up 2 wall segments at the same time with 3 Petards each (2 with siege engineers) and then flood the enemy base / army / castle depending on the defense build up.
    All surviving petards go instantly for the castle (takes 12 for a totally upgraded castle with all techs or 9 with siege engineers - less if the castle doesn't have the upgrades). The trebs from behind should have weakened it already and keep shooting it. If the petards reach that castle its gone and walls are open - the flood can start. With this push there will probably also be Rams to help take down the Castle or walls.
    If population space for the petards is missing just make more for the castle when the walls are breached and the units are flooding the enemy base.

    If the enemy has a huge army behind the wall to hold the wall it might be more useful to just treb down the castle and keep the units safe behind - blow up enough wall segments with trebs or siege rams and go for a fight once the castle is down - or break walls at other places and flood from there - Raid eco.
  • If the enemy is a crazy defender and has siege onagers pretty much everywhere in reach of his walls to stop any siege ram / unit push the only ways I see to get through these walls without heavy cost-ineffective losses is either treb down a few wall segments at once with trebs or BBC from far behind the wall but that might take some time if he builds more stonewall behind the destroyed segments.
    Or just use petards at many places at once to instantly break through walls and use the loose formation or send a few units in front to take out the siege onagers and to enter the base without giving siege onagers less value from each shot.

Re: Promi 2.10

Posted: Sun Jul 04, 2021 1:27 pm
by TheNightPanda
Arpheus wrote:
Thu Jul 01, 2021 4:15 pm
TheNightPanda wrote:
Thu Jul 01, 2021 1:55 pm
It's so great to see your enthusiasm about our ai's :) One thing I'm curious about. Which settings do you use for 8xspeed and how does it affect the number of passes per second? You may not know this but maybe someone else do. With the new micro capabilities of ai's like odette this has huge impact on performance on DE.
I had a feeling sometimes in the past that 8x might make micro less precise but I was not really sure about that and lately I haven't seen issues with delayed movements or attacks or things like that and even deer pushing worked fine. Is it a thing to consider? Doing tests at 2x speed or less would take considerably longer. And with settings you mean the game settings?

For the tests I had the following settings:
Arabia Random Map
Map Size: Tiny (2 players)
AI Difficulty: Extreme
Resources: standard
Population: 200
Game speed: fast -> in game I click + twice and it goes up to x8
Reveal Map: normal
Starting Age and ending Age: standard
No Treaty and Victory standard.
None of the advanced settings is checked and I always did mirror civ. One time Teutons rest of games were Spanish

I have high hopes that some AIs will reach Voobly Barbarian's level one day for DE which motivates me to give feedback to help them improve. I am sure a large part of the playerbase would enjoy even stronger AIs.
Still hoping that one day Barbarian will be officially released for DE to also include the new DE civs and the diverse changes. Promis DE AI is not too far away from the unofficial DE Barbarian AI. Odette is making progress with frequent updates but is not yet at the level of Promis DE AI at least from my testing at 8x speed.

I just did a test with 2x speed with odette vs barbarian. Odette went for scout rush (haven't seen that before) and the micro was very impressive. It is possible that the deer pushing was more precise at 2x - it was almost flawless.
The scout rush killed quite a few villagers and cause lots of chaos and idle time. Barbarian then defended with a lot of spearmen and Odette had to retreat most scouts.
Odette made a good skirm army with a few spears to complement the scouts but lost all the momentum and advantage in that game to a defensive tower filled with 5 vills and some skirms + spearmen. The Odette skirms had problems with the spearmen in melee and the tower again had a big impact on the outcome of the fight.
I think the best way to deal with the towers in feudal age is really to attack from another direction where no tower blocks the way or men at arms but that is an investment.
After that loss barbarian literally destroyed Odette since it had no more army to defend.
I meant wether you used the cheat engine speed hack or the game version of 8x speed :)

Re: Promi 2.10

Posted: Sun Jul 04, 2021 3:51 pm
by Arpheus
TheNightPanda wrote:
Sun Jul 04, 2021 1:27 pm
I meant wether you used the cheat engine speed hack or the game version of 8x speed :)
The game version of 8x speed - I didn't even know there was a cheat engine speed hack

Re: Promi 2.10

Posted: Mon Jul 05, 2021 5:28 pm
by TheNightPanda
Arpheus wrote:
Sun Jul 04, 2021 3:51 pm
TheNightPanda wrote:
Sun Jul 04, 2021 1:27 pm
I meant wether you used the cheat engine speed hack or the game version of 8x speed :)
The game version of 8x speed - I didn't even know there was a cheat engine speed hack
Just testet it and the game version of 8xspeed seems to increase the time between passes just like the cheat engine speed hack meaning that at 8xspeed the ai will have a reaction time 4 times slower than at fast speed (2x). This wont really affect decision making to much but should have a huge impact on any kind of micro that is relying on timing. My test in an hour long 4v4 game with barbie vs Rehoboam showed an average time between passes of 4 game seconds at the beginning of the game gradually increasing to about 8 game seconds at the end meaning that the ai will only be able to issue commands once every 8 seconds of game time. This number is around 0,5 game seconds on UP normal speed (1,7). While the DE number will probably be lower on a more high end PC than mine it will without a doubt still have a big impact on the performance of any ai to use 8xspeed, especially for actions where timing is critical like luring and archer micro.

Re: Promi 2.10

Posted: Mon Jul 12, 2021 4:36 pm
by aoefan
Can you fix the AI resigning below Hardest difficulty?

in Death Match, Ai surrender even there is villagers and military units, no need to destroy their buildings.

average time 3 vs 3 is 50 min (fast speed)

In RM games its a little bit better but the promi 1.55c is far better.

can you fix difficulty scaling pls?