The AI was further trweaked to reduce (hopefully eliminate, but we'll see) its bad switching a couple of days ago.
AI suggestions
- Thread starter Koshling
- Start date
The AI was further trweaked to reduce (hopefully eliminate, but we'll see) its bad switching a couple of days ago.
JosEPh_II
TBS WarLord
But Koshling what if the switchings are not bad? But just super players perceptions?
You have players that think the AI should Never use, say Despotism or other "lower Civics" after Ancient/Classical era. That just isn't right and lays out a predetermined path that everyone (player and AI alike) Must use. Or be chastised for and penalized for doing so.
JosEPh
You have players that think the AI should Never use, say Despotism or other "lower Civics" after Ancient/Classical era. That just isn't right and lays out a predetermined path that everyone (player and AI alike) Must use. Or be chastised for and penalized for doing so.
JosEPh
Yes, but I don't think that. As I said in response to Blue Genie the reasons it was switching to despotism were IMO valid ones (though over-weighted somewhat, so not necssarily good in the situation of his save game). There is NO CODE WHATSOEVER that knows ANYTHING about specific civics. It is ALL driven by the effects of the civics, so it is quite impossible to tell it 'despotism is bad' per se. All I do is adjust how much it values different effects. The thing I think Blue Genie was more complaining about was the NUMBER of civic switches, and the going back-and-forth between two repeatedly, which is mostly waht has been fixed by taking better account of the anarchy penalty, and future growth needs.
JosEPh_II
TBS WarLord
Okay understood.
JosEPh
JosEPh
@ Koshling, sorry for posting this issue in multiple threads, it was clearly annoying me
i posted some saves here http://forums.civfanatics.com/showpost.php?p=11419505&postcount=4851
Oops!
Mea culpa!
I totally broke AI city founding in a performance optimization last week
Working on a fix now, but I'm timed out for a few hours. Expect a fix to hit the SVN later today (I'll post on the SVN thread when it does)
You're right, I should have posted a save.
In the first (Kor Two) the Phoenician AI has a tribe waiting in its capital doing. I suspect that this is because it already has 3 cities and wants to wait until Tribalism to found the 4th. However if you play on and wait until their 3rd city is destroyed (they are at war with the Navaro), you will see the tribe wait there and do nothing for a couple of hundred turns. They also have several hunters who are just sitting in the capital doing nothing, this also continues for a long while.
I have another save from later on in the same, which I think illustrates some other problems but it's over the size limit for the forum, so I'm unsure what I can do with that.
Found the bug that causes this. It's an error that occurs when calculating expected losses and kills for large stack combat. Fix will be on the SVN later today.
In my current game the AI is now expanding, building hunters and cities and fighting (instead of having massive non-moving stacks). Some great work from Koshling. 
However I've noticed that AI is very inept at defending its cities outside its capital. In the save below the Persians (to my left) have three cities. The Tibetans are about to attack one of them (Susa) with a moderatly sized army, the city being only guarded by a slinger and stone axeman. However just 3 tiles away there is an army of Persian stone axemen which could easily defeat the attacking force. Yet they just sit there doing nothing while the enemies close in and destroy Susa.
Suggested fix: Some kind of alert from a city in danger. This would call in nearby units to aid in its defence. So when the Persian AI saw the Tibetan army approaching, a 'call' would be sent to nearby units to rush in and defend the city. Maybe this could override the existing instructions a unit had.
I admit that I have no idea how practical this idea is, so feel free to rip it apart.
However I've noticed that AI is very inept at defending its cities outside its capital. In the save below the Persians (to my left) have three cities. The Tibetans are about to attack one of them (Susa) with a moderatly sized army, the city being only guarded by a slinger and stone axeman. However just 3 tiles away there is an army of Persian stone axemen which could easily defeat the attacking force. Yet they just sit there doing nothing while the enemies close in and destroy Susa.
Suggested fix: Some kind of alert from a city in danger. This would call in nearby units to aid in its defence. So when the Persian AI saw the Tibetan army approaching, a 'call' would be sent to nearby units to rush in and defend the city. Maybe this could override the existing instructions a unit had.
I admit that I have no idea how practical this idea is, so feel free to rip it apart.
Thunderbrd
C2C War Dog
I've noticed the AI desperately needs some tuning where its selection of promotions is concerned. As a human player, I choose them to fulfill a number of various roles in their stacks and make sure my stacks have every base covered - but, and this is most prevalent in say... Barbs selecting the promo that allows them to go beyond minimum exp for barbarian kills... the ai seems to select on a far more random basis. This is a very big hindrance for their effectiveness.
JosEPh_II
TBS WarLord
@T-bird,
Really? Is this early game your posting about?
I'm in the ren era and the AI is very adept at it's promo selection. I came across some Rebauldequins that were nearly impossible to take out. They had 4 1st strike plus 40% Str and %'s against mounted, firearm, and melee. I still can't figure out how they got that combo. Bombard elephant and Hvy cannon was the only way to stop them. They chewed up my musketmen like a cow eating straw. Even my own Rebaulds failed 70% of the time against them, cause I had selected collat damage instead of 1st strike.
Now this particular game has been going on since v22 was introduced so the promos have been adapting/changing during it's course. So I don't know about the promos and early eras.
JosEPh
Really? Is this early game your posting about?
I'm in the ren era and the AI is very adept at it's promo selection. I came across some Rebauldequins that were nearly impossible to take out. They had 4 1st strike plus 40% Str and %'s against mounted, firearm, and melee. I still can't figure out how they got that combo. Bombard elephant and Hvy cannon was the only way to stop them. They chewed up my musketmen like a cow eating straw. Even my own Rebaulds failed 70% of the time against them, cause I had selected collat damage instead of 1st strike.
Now this particular game has been going on since v22 was introduced so the promos have been adapting/changing during it's course. So I don't know about the promos and early eras.
JosEPh
Thunderbrd
C2C War Dog
I suppose it could just be the barbs choices are a bit simpler and less well considered.
I've had a quick look, and the barnarians use the same evaluation as everyone else. However, the barb and animal hunter promotions are implemented entirely in Python, so they have ZERO value to the AI evaluatioon routyine (they appear to do nothing at all from their tags, and the actual effect is in Python processing). This makes it starng that the AI ever picks it. I can think of only two explanations:
1) At som point it's the only promotuion available to the unit in question so gets selected by necessity (since it has no tech pre-reqs this seems entirely possible for barbs who don't tch-up the way civs do)
2) The only promotions available all have very low value, so the random element added in the evaluation is sufficient to make this onewin occassionally.
The promotion evaluation routine probably needs some tweaking generally however. For example it doesn't distinguish between different terrain movement promotions, so it will consider tundra and desert movement equally valuable regardless of what part of the wrld it is in.
Antmanbrooks
Prince
Hi, I've noticed recently that the AI builds a couple of stacks early in the game and then does nothing? I have 5 AI's at the moment on the GEM map and none of them have moved their units more than 2-3 squares outside their borders. The effect is twofold, they don't build any buildings and don't kill any animals so fall behind in the tech race very early. I'm about 600 turns into the game and already the scores are me on 27 and the AI's all languishing on 16 or 17 points. My city is size 4 and theirs are all 1 or 2 still. I started in the worst spot too!
However I've noticed that AI is very inept at defending its cities outside its capital. In the save below the Persians (to my left) have three cities. The Tibetans are about to attack one of them (Susa) with a moderatly sized army, the city being only guarded by a slinger and stone axeman. However just 3 tiles away there is an army of Persian stone axemen which could easily defeat the attacking force. Yet they just sit there doing nothing while the enemies close in and destroy Susa.
Suggested fix: Some kind of alert from a city in danger. This would call in nearby units to aid in its defence. So when the Persian AI saw the Tibetan army approaching, a 'call' would be sent to nearby units to rush in and defend the city. Maybe this could override the existing instructions a unit had.
I admit that I have no idea how practical this idea is, so feel free to rip it apart.
Yeh, it's pretty bad at that (and it'll make the top of my list sometime soonish I hope). Part fo teh fix is something like what you suggest. A while ago I added a mechanism for units to request help from other units (that's how workers get escorts when they fel they need them for example), so it can b leveraged here too.
In regard to the specific situation in your save, a few observations:
1) The Persians don't currently have visibility on the tile the Tibetan attack stack is in, so they haven't spotted it yet (by the time they do it'll be too late)
2) The stack to the right of Susa is an attack stack not a defensive one (the AI doesn't switch readily between the two roles, especially from attack to defense) and is targeting your city of Pliska. It does have visibility on Pliska (through espionage I guess?) and knows what you have in it. As such it calculates its attack stack is not yet up to the job so it's holding ground pendign further build up.
Obviously there are a number of things that can be improved here (in principal ayway):
- Better AI deployment of scouts to have early warning visibility on nearby territory
- Easier role switching from attack back to defense when they realise it's needed
- Better deployment of defensive troops between cities, favouring those nearer borders
I plan to address all of these, but it's not low hanging fruit, so it'll be a while.
Antmanbrooks
Prince
Ok on further examination it seems the issue that the AI is having is it isn't exploring with it's units or valuing "Herbalism" enough.
All 5 AI's on my game have poor research rates in comparison to mine. Their research is between +7 and +14 compared to my +31 which is almost exclusively down to my "myth" buildings. But combine that with my early use of herbalism to make wild mushrooms and berry bushes useful, which the AI isn't bothering to do and I am working 3-4 more tiles each turn now.
Each AI power has roughly 12 Clubmen and 10 Slingers and a tracker in 2 stacks, one in their city and 1 outside but within their borders. Only one AI had any units outside it's borders!
All 5 AI's on my game have poor research rates in comparison to mine. Their research is between +7 and +14 compared to my +31 which is almost exclusively down to my "myth" buildings. But combine that with my early use of herbalism to make wild mushrooms and berry bushes useful, which the AI isn't bothering to do and I am working 3-4 more tiles each turn now.
Each AI power has roughly 12 Clubmen and 10 Slingers and a tracker in 2 stacks, one in their city and 1 outside but within their borders. Only one AI had any units outside it's borders!
Is this all with the latest SVN (last couple of days)?
Antmanbrooks
Prince
No, I have v22.1, has it been addressed in the latest SVN?
Yes it has. The AI is significantly imoproved since v22.1. If you have start-as-minors even more so
Dancing Hoskuld
Deity
Yes these are new and probably OP, I have hunters with over 200 XP running around. There are no XML tags for this feature so it needs to be python atm. In realty they should only be considered one or two promotions before you get your last promotion from combat with barbarian animals and units.
In my latest SVN game I noticed that hunters were staying in the AI cities leaving all the animals for me. This probably means I need to tweak the hunter line stats.
