Promi 2.11

Upload your latest creations here.
Please note that by posting comments to posts in this forum you forfeit your personal copyright on whatever you post. AI Scripters reserves the right to re-use objective comments in script reviews without requiring the posting author's permission. Credit, where appropriate will be given in such cases.
Locked
Aynshtaynn
Forumer
Posts: 49
Joined: Thu Aug 01, 2019 12:49 am

Re: Promi 2.08

Post by Aynshtaynn »

Felipe_EL_Guapo wrote:
Mon Aug 03, 2020 5:29 pm
I noticed that when the games start, Promi firstly prioritises sheep instead of houses, so there is a little delay there. If the sheep are seen, then the villagers immediately go for them and then realise they need houses. If the sheep aren't seen, then, they wait for a few seconds and then they build a house.
In addition to that, when starting the game as Mayans, Promi idles the extra villager Mayan bonus gives for a second. I don't know about the Chineese yet though. Nothing big but still inefficient :huh: :lol:

kunyi
Guru
Posts: 758
Joined: Tue Mar 21, 2006 7:48 am

Re: Promi 2.08

Post by kunyi »

ty
(you got Houses bug at start on PIDM DE, extreme ai too, start with military buildings than after houses, lost many time and poped ) ( target-focus player bug maybe idk, i realized this on DE extreme AI around 2-4weeks ago, but still)

User avatar
Promiskuitiv
Waheguru
Posts: 1364
Joined: Thu Nov 10, 2011 1:07 pm

Re: Promi 2.09

Post by Promiskuitiv »

Thank you Kunyi, i will fix it in the next version!

Very busy days, but i've just updated Promi to v2.09.
Mainly improving TC placement on the LandNomad map as well as the first lumbercamp timing on the Dustbowl map.

Aynshtaynn
Forumer
Posts: 49
Joined: Thu Aug 01, 2019 12:49 am

Re: Promi 2.09

Post by Aynshtaynn »

Promiskuitiv wrote:
Fri Aug 14, 2020 12:57 pm
Mainly improving TC placement on the LandNomad map as well as the first lumbercamp timing on the Dustbowl map.
So no quality of life improvements for more standard game modes?

User avatar
Aleph
Yogi
Posts: 487
Joined: Sat Jan 17, 2015 10:07 am
Location: Germany

Re: Promi 2.09

Post by Aleph »

Aynshtaynn wrote:
Sun Aug 16, 2020 4:07 pm
Promiskuitiv wrote:
Fri Aug 14, 2020 12:57 pm
Mainly improving TC placement on the LandNomad map as well as the first lumbercamp timing on the Dustbowl map.
So no quality of life improvements for more standard game modes?
Promi is literally the DE default AI, it has far more quality of life shit than any other AI in existence. That's probably the main thing he's worked on, as his day job, for the past 2 years. Personally I'm glad he can finally focus on the competitive side of things again!

Aynshtaynn
Forumer
Posts: 49
Joined: Thu Aug 01, 2019 12:49 am

Re: Promi 2.09

Post by Aynshtaynn »

Aleph wrote:
Sun Aug 16, 2020 6:31 pm
Promi is literally the DE default AI, it has far more quality of life shit than any other AI in existence. That's probably the main thing he's worked on, as his day job, for the past 2 years. Personally I'm glad he can finally focus on the competitive side of things again!
Yes yes definitely, Promi is an awesome AI already by itself! I just wanted to ask if there was a difference between 2.08 and 2.09 versions besides LN and Dustbowl. I definitely agree with you and personally, I'm so happy that he is able to work on Promi AI as well as DE. :jump

User avatar
Promiskuitiv
Waheguru
Posts: 1364
Joined: Thu Nov 10, 2011 1:07 pm

Re: Promi 2.10

Post by Promiskuitiv »

There are usually at least QoL improvements contained in each update as well. :)
I often get ideas for small tweaks and implement them, it's the same for this update where i'm testing certain smaller changes (to specific BOs).
Submitted Promi 2.10 for the next games of the AIScripters Championship. :)

Felipe_EL_Guapo
Yogi
Posts: 250
Joined: Wed Jun 11, 2014 8:54 pm

Re: Promi 2.10

Post by Felipe_EL_Guapo »

Thanks for the new Promi. Will you be able to help me with the query I asked to you by PM?

Aynshtaynn
Forumer
Posts: 49
Joined: Thu Aug 01, 2019 12:49 am

Re: Promi 2.10

Post by Aynshtaynn »

There's a problem about taunt 250. Even in AI only game, the players apart from the one you're supposed to control don't report the strategy they'll use. Even the ally players, they just send blank message in the chat

Arpheus
Forumer
Posts: 40
Joined: Wed Mar 10, 2021 1:11 pm

Re: Promi 2.10

Post by Arpheus »

Hey Promi, the DE AI has become really strong putting up a great fight and it seems it is not too far away from having realistic chances to win against the unofficial DE barbarian AI. It already won a game in a rare occasion.
I hope this is the right place to post some feedback on the DE AI and I am excited to see more updates on your AI.

Here some observations and thoughts about how to improve your AI based mainly on AI vs AI play:
All the games were played on 8x speed standard settings against the unofficial DE Barbarian AI on extreme difficulty (no cheating) on Arabia with Mirror civ Spanish mainly and Teutons for one game

What is great (in general not only AI vs AI)
  • Lategame pushes are very strong with tons of siege and strong unit compositions. Also the retreat when a fight looks bad is great to conserve some army for later and for defense
  • Team games against DE AI very efficient trade
  • Defensive tower at a forward ressource location is great for defensive purpose and helps to have a point to retreat to for military units when risking to be overwhelmed
  • Great economy buildup
  • Sheep scouting seems fine
  • Deer Pushing is really good
  • Drush is strong when it happens - without drush the chances to win for DE AI against Barbarian seem lower
  • Instant counter production like pikemen to beat scout rush is great and very effictive vs barbarian AI BUT even better would have been if the AI found out about the scout rush earlier by scouting for archery ranges or stables.
What could be improved
  • Early scouting is very chaotic - it runs through the explored areas a lot and with that behaviour wastes so much time and often doesn't find the sheep because it doesn't run in circles. These sheep might be found by the enemy which is really bad.
    Sometimes it seems to do well for a short amount of time by running 1/5th of a circle or so but then it stops that and runs through explored areas again. Such early bad scouting efficiency has a lasting effect on the entire game by not finding important ressources and sheep/deer and taking longer to finally go search the enemy base and not having time to scout the enemy base again in early feudal to find out if it makes archers/skirms or scouts.

    I think it should try to circle the base early with sheep scouting and with the scout itself to find a good spot for a near lumber camp which is adjacent to the TC to make sure it won't place it on the other side facing away from the TC which it does sometimes maybe because of bad scouting. Then keep running some circles around the base to find the sheep and don't move through the explored area unless there is a reason to do so.
    To make it perfect if that is possible it should also not try to explore areas which the early sheep scouting already explored
    A good example is the barbarian AI which does it not perfectly but far better. Barbarian AI early scouting with making the circle a bit larger to cover more unrevealed area would be a great scouting.
  • Lumber camps facing away from the townceter and facing towards the enemy base is such a dangerous thing to do.
    These vills are easy prey as they will have a hard time to retreat. Not sure if this can be fixed but that would definitely be worth it. At least it usually builds a tower there a few minutes later to protect the wood line which is a good thing since that wood line often is the closest thing to the enemy base!
  • Sometimes it places the 2nd lumber camp far in front of the TC (there was another woodline forward between the TC and the place it chose) which is really dangerous when there are backward woodlines that could be used to have a stable wood income in the most important early game phases. Maybe it didn't select these because they were only partially scouted (bad early scouting)
  • It pulls back the drush for the single enemy scout which is wasted momentum - if a real threat is incoming it's a good idea but it should not move back half the map just for the main scout of the enemy - villagers can handle that
  • During a drush the DE AI retreated 1-2 low hp militia. I wonder if that is worth it or if harassing villagers more might not be more valuable. Maybe there is a reasoning behind that to bring them in with the next attack wave but I think low HP militia don't have much value later on
  • Different attacking directions and moving around castles/towers/tc into enemy resource locations would be very strong. I wrote quite a few suggestions in the Odette AI topic a while ago how to improve AI's in general
  • Units not dying to tower + castles is important - that has a strong negative snowball effect and is one of the most easy ways to cheese the AI and not having to really face and destroy enemy army with your own army which is also very bad practice for real games vs humans which will not suicide their army into towers and castles.
    The AI should not move in to attack towers unless it can bring them down or has a great benefit by doing so like killing vills. But most of the time the vills will enter the tower and kill the skirms/army.

    One nice move could be to move under the tower and stay there to deny a gold camp for a while for example. That might force the enemy player to make another lumber-/mining-camp and waste 100 wood for that + possibly waste villager time of the villagers trapped inside the tower which would be a great benefit until army shows up to clear the skirms/archers under the tower.
  • Barbarian loves monks for healing/converting which is worth a lot and super annoying for the enemy. Not sure if this is necessary but it would be a powerful addition for sure
  • Standing under an enemy tower with 3 skirms and then more and more skirms - it might work maybe after many minutes to bring down the tower but the efficiency is horrible
  • In a game the AI couldn't find the enemy base and then started to run in circles for a while before it went back to continue scouting (not having found the AI). It could have continued it's circle around the map to find the enemy
  • Deer pushing when the deer are at the absolute border of the map doesn't seem to work too well. It should move next to the deer at the border to make it move away from there to get behind the deer to push it. This is probably a rather rare issue having to do with weird map generation (2 ostrich were at the absolute border of the map)
  • DE AI moved forward with 5 or 6 skirms to attack but they were not in one single stack but rather spread from walking there so the DE AI lost them very fast to barbarians stacked 6-7 skirms without dealing damage.
    Then the enemy skirms snowballed and destroyed all DE AI skirms + denied the mill + 7 farms and killed a few vills. Here it would have been great to first not move out with only skirms, keep them all stacked and to add some scouts if the enemy has not a single archer to get rid of these skirms rather fast.
    Since it didn't the skirms got more and more - having 20+ and the game was decided already at around 20 min - DE couldn't get ressources efficiently with skirms around and soon barbarian reached castle age with knights + rams and ended the game at 32:30
  • Overall it seems the DE AI has trouble ending games in Castle Age. It often tries one or two attacks which don't work and then waits until imperial - which is fine I guess. Barbarian e.g. often ends games in Castle Age with massive knight/ram pushes once it gets a lead.

    When the DE AI gets a lead it often cannot use that lead efficiently at least against the barbarian AI and most of the times gets pushed back by the barbarian which can turn the entire game. That is often caused by barbarian fighting under TCs / Towers and adding lots of spears + monks to their own knights. Also on the defense they can refill their units a lot faster.

    This is probably also a reason why it might benefit the DE AI to not waste too many units in castle age but use an Imperial age power spike with rams + trebs.

    The DE AI often gets to imperial age before the barbarian AI (if it survives until this point) and then tries to push fast to make use of that power spike. It then instantly researches the last cav armor BUT waits too long for the Cavalier upgrade. It should instantly research that with a high priority since it lost some games doing that too late and because of that loosing the entire army. That upgrade is cruicial if it goes for a knight push

    Also it won't add Trebs instantly - maybe because of it delayed the castle to go up earlier - not sure about that. But adding 1-2 trebs to the first Imperial push would add so much powerful momentum to take down the enemy castle. And it should not suicide the army into the enemy castle but keep it out of range and protect the trebs.
  • Another way to make use of that first imperial age attack and power spike would be to just move around that castle and attack the base from another direction which could be so powerful.

Here observations of how some games went on:

How it won vs Barbarian in a teuton mirror - rare win vs barbarian:
https://easyupload.io/kuq138 (will be online for around 30 days - and is from DE June Patch beta - hope it is compatible)
  • Scouting was surprisingly good this game - instantly found the sheep and enemy base - also did more circle scouting than usual
  • Drush payed off - killed one vill and delayed the enemy mill by a lot
  • barbarian was Feudal at 11 min - DE AI at ~13:30. The last surviving militia then moved around and found the enemy stable - saw a scout coming out of it and moved back to his TC
  • Even though the DE AI had seen the enemy stable and a scout coming it it didn't prepare any spearmen (room for improvement) and by 15:30 5 enemy scouts arrived. Just a second before the vills built a godly placed defensive tower protecting gold, the mill and even stone. Seems like huge luck or great decision making
  • the militia defeated one single scout but the other 4 scouts harassed the 3 vills at a woodline. Only now it starts to make a few spearmen. First spearman dies next 2 manage to kill two scout - if the 2nd spearman wasn't following a scout while being bashed by 2 other scouts for free that would have been more efficient.
  • at the same time barbarian AI comes from another direction with 6 skirms + a spearman - the 2 surviving scouts die to TC fire while trying to snipe vills but the skirms take out the spears easily.
  • But now DE AI reached Castle Age somehow at around 19:30 - more enemy skirms have come but DE AI instantly builds 2 knights and then more knights to easily destroy the barbarians skirms and scouts - from there on an economic advantage for DE AI
  • The 6 knights instantly push forward and can deny a forward tower trying to get up from the barbarian which has to cancel it - that was helpful. That Tower might have messed with DE AI had it gone up
  • the knights get quite some kills - 2 knights unnecessarily died to TC fire while chasing the vills (potential for improvement and snowball if they don't die). The Knights clean up quite some spearmen and distract a lot until 23:45 when barbarian also reaches Castle age --> DE AI retreats the knights and masses knights - huge score lead now with 3 TC boom + military advantage
  • Sends forward a single knight twice for scouting - first gets converted by a monk. Both AIs also collects relics in the meantime.
  • at 32 min it moves forward for a knight push with 24 knights - defeats quite a few enemy pikes, a monk and some enemy knights but most of the fight is under a full TC fire with 16 arrows each dealing 2 damage (+2+2 armor knights, 5+1 attack) !!! (teuton TC with I guess 25 vills inside). Enemy had at least 10 knights a few pikes and at least 6 monks - lucky for the DE AI it plays teutons this game otherwise this would have been a disaster like seen in other games where they converted lots of knights
  • DE AI lost a lot of knights in this fight - a lot to the TCs - but cleaned up most enemy knights and monks. 13 Knights left for DE, 3 injured Knights + 3 Monks for barb
  • DE Reached Imp at 37:30 with 128 vills and 24 knights + 4 pikes while Barbarian has only 83 vills and 7 knights 4 pikes and 6 monks.
  • at ~39 min the DE has last cav armor starts building first castle in forward position and moves out to attack the barbarian with 28 knights, 5 pikes and 4 rams. It is lucky that the Cavalier upgrade completed just in time (ususally not).
    Barbarian has 10 knights, 6 monks and 7 Pikes at that time and still not clicked up to Imp. The fight takes place again under the TC with 18 vills inside but the Army is too powerful and with the rams joining the fight and eating the first TC the barbarian realises that its over. gg wp
How it almost won vs Barbarian
  • It was up castle age first had a stronger knight army but had to fight under a TC and the Barbarian brought a mix of pikes + knights + few monks first. DE AI had to retreat and fight back in the own base - then attacked again under the TC and barbarian had like 5-6 Monks converting tons of knights --> lost battle
  • Then early Imp (before barbarian) it did another theoretically powerful attack but with delayed Cavalier upgrade and faced lots of pikemen + knights - rams from barbarian and barbarian from there on won the game by upgrading to halb and Cavalier + massive ram push. That earlier Cavalier upgrade probably would have changed the outcome of that fight
  • Also the DE AI built the one Castle it had which could have been the saving grace at the border of the map not directly facing the enemy and protecting only a minor part of the eco - central + facing the enemy direction or just in front of the base would both have been good places to build it
  • After being scout and skirm rushed in another game with lots of losses it still got to castle age quite a bit before the Barbarian AI and won a few engagements. Couldn't take down an enemy tower filled with vills because the barbarian defended it also with units - tried two attacks and retreated both times + waited for imp. Once in Imp it instantly does the last Chain barding armor BUT it delays the Cavalier upgrades far too long. It should be one of the very first things to do especially since the DE AI then attacks in imp before the Cavalier Upgrade is through which is so much wasted power spike.
    But still it brought down a forward castle fast with a 5+ Ram + tons of knights push BUT it lost all the 20-30 knights to enemy mass pikes + knights (barbarian still in castle age) especially because that Cavalier Upgrade was far too late. If it was there it might have won that engagement but with the knights it lost the full army after the enemy castle was down.
How it looked like it would win vs Barbarian but then things went bad (very interesting game where lots of great information can be taken from)
https://easyupload.io/ee6nt6 (will be online for around 30 days - and is from DE June Patch beta - hope it is compatible)
  • Barbarian did a scout rush into the back woodline - it completely circled around the base into the back woodline (incredible for an AI to do that) - it took a while and quite a few villager losses to kill a few scouts with new spears and push them back.
    If early scouting was better (less wasted ways through the base) it could find the enemy base earlier and be back earlier to push the deer. After pushing deer the AI could then go back to the enemy base in early feudal to find out if it builds stables or archery ranges.
    That way it could prepare counters accordingly - if the enemy has stables make a few spears and place 2 of them at gold and woodline and maybe another at the mill.
    If 2 archery ranges go for skirms. That way the feudal defense would be very strong and villager losses would be less.
    Against players who fully wall this might not work if they hide their production buildings but in most cases it should help the AI a lot and that might be a good starting point on the way to become the most powerful AI in definitive edition (atm still the unofficial barbarian).
  • Barbarian again circled the base and came from behind with quite some skirms - pushed it back
  • DE AI then did an attack against Barbarian woodline - there was a tower which was filled with vills and lots of scouts and skirms behind. That combination destroyed most of the DE AI army. The problem was it didn't see that army while moving under the tower and once fully under the tower it took the full tower damage + fought against that army for a while and lost most or all of it army. Fighting under a tower filled with vills is a very bad idea
  • It got up to castle age earlier than barbarian which was great and that let it defend from the next attack. DE AI knight production in castle age is very strong and helped it get to imp
  • At around 46 min it lost a huge knight army. Later the DE AI moved back with the full army to take down a single enemy lumber camp close to it's base - not sure if that was worth it.
  • A bombard cannon moved forward completely alone without any other unit protecting it --> instant loss of an expensive unit. Bombard cannons should only move with army
  • Later it moved in for a new push - siege rams went forward to take down a new castle but the army got distracted by another enemy lumbercamp far right and moved there with the full army - except the Siege rams which moved forward to the castle without any protection. The Siege rams were completely wasted that way without army.
    The Siege rams should have tried to follow the army - or even better the army shouldn't have been distracted by a lumber camp which has not much priority to take down. Sending 1 cav unit to there would be enough
  • It also switched the army to hand cannoneers because of a few Pikemen (underpowered unit unless to counter strong inf) which made it loose the game even faster because it now had no more defense against paladins + skirms + trebs + rams.
    If it had instantly researched Cavalier and not attacked until cavalier update is through and after that kept making cavalier and pikes to counter the enemy pike (instead of full HC and no cav) and if it didn't get distracted by the random lumber camps it might have won that game.
How it lost vs Barbarian
  • Archer+Skirm Rush pretty much finished the game as the AI was not prepared for that at all and went for early Castle Age. It could have worked if the AI had built a defensive tower at a good location close the the own base or even better if the scout would have found out that the enemy was massing archers to make a skirm counter army. In Castle Age a mangonel could be an option but the damage was dealt there already

Locked