Flintlock patch's perfume values

[Predator145]

Flintlock's perfume ability opens up so many doors to having the AI prioritize improvements and train different types of units.

We know that by having the right stats and thus values could have the AI build a mix of different units of the same strategy flags (like it does with "light tanks" (8/3/3) and "combat engineer" (11/6/1) in the WW2 pacific scenario). Before Flintlock's perfume patch, there were organic ways to perfume a unit such as giving land units inconsequent flags like "stealth ability, lethal land/sea bombard, airdrop without operational range, stealth attack vs dummy targets).

Some abilities like "all the terrain as roads" or "detect invisible" are worth so much that when checked the AI will build nothing but these regardless of unit stats. When I gave explorers the defense AI strat in the hope of seeing them pillage the AI ended up fortifying their cities with a lot of them. It was a failure.

Negative/positive perfume could help with that. The question is? How much are the flags and abilities in the editor worth in perfume points? Only time consuming trial and error will discover that.

So if you've been experimenting, please share your knowledge here.

 

[Predator145]

I've found out through debug testing that by adding "Lethal Land Bombard"+"Lethal Sea Bombard"+"Stealth"+"Stealth Attack" with no targets and "Enslave" with no unit to the stock game horseman results in the AI building a mix of horsemen, swordsmen and longbows. Before it would never build horses if it had the other 2 available.

I've also found out that the stock game AI will continue building a small number of weaker offensive units even though a far superior one is available, contrary to the notion that the AI does not perform a cost to stats analysis. Low shield cities close to the front during war time have been observed to build longbows and immortals even though knights were available. The AI does perceive that it needs to get some units out ASAP.

 

[justanick]

I've also found out that the stock game AI will continue building a small number of weaker offensive units even though a far superior one is available, contrary to the notion that the AI does not perform a cost to stats analysis. Low shield cities close to the front during war time have been observed to build longbows and immortals even though knights were available.

All 3 have an attack value of 4. This does not seem far superior.

It seems to me that in order to disprove the idea that the AI does not perform a cost to stats analysis one should prove that it builds longbows or immortals when cavalry is available.

Else it could be something like this: Compare attack. If same compare cost. If same compare defence...

 

[Predator145]

All 3 have an attack value of 4. This does not seem far superior.

It seems to me that in order to disprove the idea that the AI does not perform a cost to stats analysis one should prove that it builds longbows or immortals when cavalry is available.

Else it could be something like this: Compare attack. If same compare cost. If same compare defence...

Unfortunately in the debug game I'm observing Cavalry itself got perfumed with the same settings as the Horseman in order for the AI to not stop building it as the main offensive unit once they get to Replaceable Parts. With that setting no more longbows were built. In that same game, the resourceless Greeks still build some longbows when they have rifles, which are flagged as offensive as well.

 

[Ozymandias]

Flintlock's perfume ability opens up so many doors to having the AI prioritize improvements and train different types of units.

We know that by having the right stats and thus values could have the AI build a mix of different units of the same strategy flags (like it does with "light tanks" (8/3/3) and "combat engineer" (11/6/1) in the WW2 pacific scenario). Before Flintlock's perfume patch, there were organic ways to perfume a unit such as giving land units inconsequent flags like "stealth ability, lethal land/sea bombard, airdrop without operational range, stealth attack vs dummy targets).

Some abilities like "all the terrain as roads" or "detect invisible" are worth so much that when checked the AI will build nothing but these regardless of unit stats. When I gave explorers the defense AI strat in the hope of seeing them pillage the AI ended up fortifying their cities with a lot of them. It was a failure.

Negative/positive perfume could help with that. The question is? How much are the flags and abilities in the editor worth in perfume points? Only time consuming trial and error will discover that.

So if you've been experimenting, please share your knowledge here.

You can start with: "The BIG Q Continued - AI Flag Choices For Building Unit" and Post #4 in, "An Intro To & An Encyclopedia Of Modding." Of further note is that the AI grossly undervalues the "Blitz" Flag; I recommend throwing something else in, to help the AI along.

 

[Ozymandias]

My question would remain: How much does Perfume override the AI's "innate" preferences. If I have the time, I'll pull those notes together, consolidated.

 

[Predator145]

The AI has this tendency to want to build slow attackers even there are enemies close to the city. Here the AI has access to to Knights but prefers building Longbows because of the enemy units nearby. This is evidence of the AI performing a cost-benefit calculation. Yes, the Aztecs had Chivalry.

The Mongols made a mixture of Keshiks (modded to 4/1/3 with 2 def bombard) and given Mongol knights. At close distance, it still builds Man at Arms (Med Inf) and Longbows.

The Inca border cities warring with Aztecs produced Longbows instead of Knights due to the distance. The knights were already editor perfumed by being given lethal land and sea bombard+stealth+enslave with no units+stealth attack with no targets and yet it will build longbows/med inf. I've yet to see this when cavalry is available though. Also, those editor perfumed Knights were largely preferred over offensively flagged riflemen and are buit in a mixture with guerillas:

The Spanish AI also went crazy with Conquistadors instead of making any knights. So I tried -40 perfume. No effect. They've made so many conqs that I've had enough and slapped on a -10000 perfume. That stopped the conq production. The ATR ability has extremely high value. I'm still not sure what C3X negative perfume is required to get them to build a reasonable mix. My guess right now is at least -100. Conqs don't have the atk strength to be the backbone of offensive units.

Artillery should have a perfume to get the AI to use it more once they have RP. The AI is best using a stock game regiment of bombard unit production before that. These small number helps them mostly with removing walls. Esp with overpowered cavalry and underpowered muskets, there is little need for an SOD of cannons.

Also, here's evidence of the AI making the occasional Explorer:

Courthouses got a +40 perfume and give 1 happy face. The AI builds it everywhere. The net effect is positive though they will build it in their capital is you can see.

 

[Ozymandias]

1. The AI, without taking into account Flags etc., will basically overall value a Unit as "A" +"D".
2. I don't believe that anyone has figured out where "M" kicks in, despite considerable research somewhat sorting out Flags.
3. The AI will also re-flag Units from "A" to "D" (and vice-versa) as it sees fit (usually highest "A" from "D" if "A" value is higher; etc.)
4. As I noted above, if you're Flagging any Unit, then the entire "World Of AI Unit Production" changes.

 

[timerover51]

I apologize for my ignorance, but what exactly is "perfume"?

 

[Predator145]

I apologize for my ignorance, but what exactly is "perfume"?

Modifications to increase the AI's likeliness to build something but not alter gameplay stats. So let's say I want to the AI to focus more on fast attackers. I could use the editor to give the horseman inconsequential things like the stealth ability, lethal land and sea bombard, stealth attack with no targets, enslave with no resulting units. This tricks the AI into thinking the unit is worth more than it really is and will select to build it more often. That I call it editor perfuming.

A C3X perfume requires C3X to be installed and is modified on the default.c3x_config sheet. This is what we mean when we mention "perfume". This is very useful for the AI to build time sensitive things like the Forbidden Palace or Auto Production SWs as soon as they're available. Other time sensitive improvements like walls, barracks, harbors, aqueducts and courthouses also benefit. Overall, smart perfuming increases the AI's performance. Conversely, bad perfuming makes it perform even worse than in stock game.

 

[timerover51]

Modifications to increase the AI's likeliness to build something but not alter gameplay stats. So let's say I want to the AI to focus more on fast attackers. I could use the editor to give the horseman inconsequential things like the stealth ability, lethal land and sea bombard, stealth attack with no targets, enslave with no resulting units. This tricks the AI into thinking the unit is worth more than it really is and will select to build it more often. That I call it editor perfuming.

A C3X perfume requires C3X to be installed and is modified on the default.c3x_config sheet. This is what we mean when we mention "perfume". This is very useful for the AI to build time sensitive things like the Forbidden Palace or Auto Production SWs as soon as they're available. Other time sensitive improvements like walls, barracks, harbors, aqueducts and courthouses also benefit. Overall, smart perfuming increases the AI's performance. Conversely, bad perfuming makes it perform even worse than in stock game.

Which editor would you use to do this?

 

[Predator145]

Which editor would you use to do this?

Any editor would do.

 

[Predator145]

The AI will value the occasional fast attacker even if its best attacker is slow and much, much better. In this case the Incas have access to no Oil and Saltpeter so their best attackers are Marines. But they still have access to Horsemen. Their east coast city ended up training a Horseman to fight the Mongols because the AI takes distance into the equation.

So if you don't want the AI to waste its shields in cases like this, give Knights and Horsemen a big negative perfume. The occasional Cavalry IMO is good for supplementing the AI's slow moving army to threaten the human's thinly spread front.

 

[Ozymandias]

I began what became a community wide analysis - ultimately spanning multiple threads of 800+ Posts and thousands of iterations of scores of ;test cases - regarding the AI criteria for Unit build choices. Motived by Perfume & Whatnot, I'm finally pulling it all together. ETA: 1–2 weeks.

 

[Vuldacon]

regarding the AI criteria for Unit build choices. Motived by Perfume & Whatnot, I'm finally pulling it all together. ETA: 1–2 weeks.

Ozy... Yes, there Many Factors Governing and Influencing AI Unit Build Decisions. Looking Forward to Your Analysis

Everyone has their opinions based on what they have observed through the years of Game Play.
Over All I believe the AI does a decent job however, adding extra adjustments/programing to "tweak" AI Unit build decisions is beneficial to Game Play. That said, this is usually based on personal preferences more than anything else.

Certainly speed counts highly in decisions for example but what Force would you prefer to deal with, 100 Zulu Warriors or 10 Rifleman Attacking your City? So what is Power and what strategy provides it? Are Numbers of Units stronger then fewer stronger Units? Many factors to consider
... Nuclear Weapons trump numbers but do you prefer Army Ants or the Elephant and Tiger.
All living things strive to gain and expend their power. Life itself is the Will to Power.

 

[Predator145]

The AI can't really make good use of human wave assaults because it's hardcoded to avoid very low odds of success fights. When you reach the cavalry vs infantry stage, it would try to by pass your infantry defended cities to reach the empty ones behind them. Of course the cavalry would never make it as rail mobile artillery cuts them into ribbons and the human player goes on to fish for leaders. So even a 6 atk to 10 def ratio would break the AI.

 

[timerover51]

Ozy... Yes, there Many Factors Governing and Influencing AI Unit Build Decisions. Looking Forward to Your Analysis

Everyone has their opinions based on what they have observed through the years of Game Play.
Over All I believe the AI does a decent job however, adding extra adjustments/programing to "tweak" AI Unit build decisions is beneficial to Game Play. That said, this is usually based on personal preferences more than anything else.

Certainly speed counts highly in decisions for example but what Force would you prefer to deal with, 100 Zulu Warriors or 10 Rifleman Attacking your City? So what is Power and what strategy provides it? Are Numbers of Units stronger then fewer stronger Units? Many factors to consider
... Nuclear Weapons trump numbers but do you prefer Army Ants or the Elephant and Tiger.
All living things strive to gain and expend their power. Life itself is the Will to Power.

I have one game where I am playing my mod of Test of Time where the Spanish built one army and while using Espionage I discovered that the Mongols were building one.

As for numbers of units stronger than fewer stronger units, especially in defense value, my conclusion is that they are not. However, that is based on my modded games. A straight standard game may be different.

 

[timerover51]

The AI can't really make good use of human wave assaults because it's hardcoded to avoid very low odds of success fights. When you reach the cavalry vs infantry stage, it would try to by pass your infantry defended cities to reach the empty ones behind them. Of course the cavalry would never make it as rail mobile artillery cuts them into ribbons and the human player goes on to fish for leaders. So even a 6 atk to 10 def ratio would break the AI.

Again, in my modded games, I have had the AI make some very low likelihood of attacks. A couple of Barbarian Cossacks (I have boosted the Barbarian Horseman quite a bit) will attack an army of Swiss Mercenaries, 4/6 with a hit point boost of 8 per Swiss Mercenary, plus a boost of 8 hit points for the army. Those do occur in my mods. I have not played an unmodded game in maybe 15 years.

 

[Ozymandias]

I have not played an unmodded game in maybe 15 years.

Ditto.

 

[Predator145]

Modded games retain stock game's AI behavior when it comes to target selection since C3X does not have any mechanisms to alter that. The AI will prefer softer targets if they are available. This is well documented with tactics like the "funnel of doom" using armies. Should you however leave no softer targets available, it will use human wave attacks. Landing forces on an island/continent will have everything thrown at them. And with C3X enabling the AI to use artillery, it will make sense to amph. assault a place and have the rest of the spear shaft dig in if there are no mountain tiles to land on. Having 50 rail mobile artillery descend upon your beach head is no joke.