Jump to content

Suggestion


rich12545

Recommended Posts

I believe I might have found a solution to the problem.

Suppose a change was made to the AI ROE. Something like this.

If [enemy unit] is more than [sight radius (or maybe 1.1-1.2 x)] then ignore

AI sighting of cities would not be affected.

This would eliminate the bulk of the problems associated with the all-seeing AI. It looks like it wouldn't be very time consuming to implement. Not perfect of course, but good enough. It would make the game playable so far as I'm concerned.

Link to comment
Share on other sites

From one of my earlier posts:

In fairness, I also appreciate an AI that uses FOW. The Strategic Command series does this and it works well, but as stated this is really something that should be coded from the start and not backfitted later. Perhaps a fix could be implemented to restrict specific unit spottings, such as for subs and paratroops, etc, in response to some of the harsher criticisms. But this would have to be something Brit works on as a hobby development effort from now on. Far better to help encourage him to continue doing what he can when he can...

Maybe Brit can do something along these lines or adjust the ROEs, or maybe not. Or if he releases the AI dll as open source, maybe others could do something. It is likely not trivial. Whatever changes could be made would also have to be assessed to see if they actually improve the computer opponent performance or not. Again, not trivial.

We'll have to see what Brit says, since he knows best.

Link to comment
Share on other sites

I believe I might have found a solution to the problem.

Suppose a change was made to the AI ROE. Something like this.

If [enemy unit] is more than [sight radius (or maybe 1.1-1.2 x)] then ignore

AI sighting of cities would not be affected.

This would eliminate the bulk of the problems associated with the all-seeing AI. It looks like it wouldn't be very time consuming to implement. Not perfect of course, but good enough. It would make the game playable so far as I'm concerned.

Well, the AI needs to do a little more than that. I'm afraid that he might be too docile if he did that. For example, if he sighted one of your ships, but then you moved out of view range, then according to this algorithm, he'd act like your ship didn't exist anymore. A human player, on the other hand, might send a few ships or aircraft to the area to scout you out again because he knows you're still around.

Link to comment
Share on other sites

Well, the AI needs to do a little more than that. I'm afraid that he might be too docile if he did that. For example, if he sighted one of your ships, but then you moved out of view range, then according to this algorithm, he'd act like your ship didn't exist anymore. A human player, on the other hand, might send a few ships or aircraft to the area to scout you out again because he knows you're still around.

Good point. Suppose, in addition to the op, we added something like this:

If [enemy unit] was sighted normally then AI would remain aware of sighted unit for x number of turns, perhaps two or three.

Also, if all those enemy units were ignored by the AI, would that hamper general AI movement or would the AI move, expand, etc in the same way?

I think if something that didn't take much time could be implemented, it would make the game better.

Another tactical consideration that I've considered is that the AI from maybe 2-3 turns away would spot a transport and would easily be able to intercept it and that this would make sending transports very dangerous probably necessitating escorts every time. However, based on your answers in the other thread it seems I've been overestimating AI capability. That made the all-seeing AI much worse in theory than it is in practice.

Link to comment
Share on other sites

Good point. Suppose, in addition to the op, we added something like this:

If [enemy unit] was sighted normally then AI would remain aware of sighted unit for x number of turns, perhaps two or three.

You mean not using FOW for two or three turns??! This is totally unacceptable!!1!

Link to comment
Share on other sites

I've been watching some game playbacks, at here are my observations:

1. The AI will "post" ships around it's ocean territory.

2. If a weaker or equal ship group comes within ~2.5 turns movement of these ships, they will set course to attack.

3. It can see your subs, so if they get near destroyers they will be sunk.

4. The AI will repair damaged ships

5. Elite AIs will build a LOT of ships, mostly destroyers & transports. A fair amount of subs, fewer cruisers and BBs. They do not seem to be limited by resources.

6. AI does not escort the transports at all.

Overall given the above the AI is quite effective at sea combat. There isn't a great strategy (yet) to really beat it other than raw production. Of course, you can say this is a good thing since if there was the AI would just be a whipping boy :)

Given the way it is, I don't use subs against the AI at all. They don't build too many capital ships anyway, so there isn't a big need for them. You can pick off his transports fairly easily since they don't escort them. However, the AI compensates for this by building tons of them (elite AI must have a build it production bonus).

Even if the AI was modified so it couldn't see subs, I still don't think it would change much. They use so many destroyers that (the player) building a bunch of subs just doesn't make sense anyway.

Overall it seems to me that when the AI wins, it isn't because of the lack of FOW. They just can outproduce you. Since there isn't a clear tactic to always get the best of each engagement, you lose by attrition (no defense/terrain bonus on the sea, etc...).

I think we would get a better improvement to the game by added some more tactical elements to sea battles. This we could probably do with the editor. One idea I saw before was sea mines to protect resources. I have tried increasing the engagement ranges of CR & BB beyond visual to give a benefit to using spotters and such.

If we did a non FOW algorithm, it would be something like:

1. Send all units on patrol to find enemies

2. Remember locations of sighted units, with uncertainty circle based on movement

3. Vector attackers toward center of uncertainty circle

Looks great, right? Here is the problem. I could easily let the enemy see my weak units, draw them in, and then pounce with the strong ones. Repeat 100x times, game over. Same thing every game. Now to fix that problem we need to add some learning to the AI algorithm, which is a much bigger project. The current AI is quite good at sending the right units to its battles (too good you might complain). Remember the the opposite is a pushover AI.

Anyway, the point is that maybe with some more tactical elements to sea combat you can get around the FOWless issue in a satisfying manner.

Link to comment
Share on other sites

Suggestion?

A rule that you set a unit to return to nearest repair base and it it fights on the way it still returns to be repaired?

Also while repairing if it comes out to fight it returns to base for repair.

When repairs are done this rule goes away.

I suggested that a long time ago. And that a unit has a minimum number of hit points left before it seeks a city for repairs.

Link to comment
Share on other sites

I've been watching some game playbacks, at here are my observations:

1. The AI will "post" ships around it's ocean territory.

2. If a weaker or equal ship group comes within ~2.5 turns movement of these ships, they will set course to attack.

3. It can see your subs, so if they get near destroyers they will be sunk.

According to Brit answering a couple of questions I asked in another thread, the AI doesn't have the capability to do 2 & 3. But according to your experience it does. This is what I've been talking about the last week or so.

4. The AI will repair damaged ships

5. Elite AIs will build a LOT of ships, mostly destroyers & transports. A fair amount of subs, fewer cruisers and BBs. They do not seem to be limited by resources.

6. AI does not escort the transports at all.

Escorts have been discussed at length but never implemented. In the Empire Enhanced I was able to set escorts for everything. If the AI has those units available, it will use them.

Overall given the above the AI is quite effective at sea combat. There isn't a great strategy (yet) to really beat it other than raw production. Of course, you can say this is a good thing since if there was the AI would just be a whipping boy :)

Given the way it is, I don't use subs against the AI at all. They don't build too many capital ships anyway, so there isn't a big need for them. You can pick off his transports fairly easily since they don't escort them. However, the AI compensates for this by building tons of them (elite AI must have a build it production bonus).

I didn't know the AI had a huge production bonus. I don't mind a small one, maybe 5-10%, but the game is not fun is you're overwhelmed all the time. In Empire Enhanced you can set the combat or production bonuses from zero all the way up for each AI separately.

Even if the AI was modified so it couldn't see subs, I still don't think it would change much. They use so many destroyers that (the player) building a bunch of subs just doesn't make sense anyway.

Apparently the AI still doesn't produce ships in a good percentage. Last year I reported that it was producing almost all battleships. Looks like it's too far going the other way.

One nice thing about Empire Enhanced is that this can be adjusted in the script. But it's the most difficult thing there was at least for me. And it was the last thing I was able to get right.

Overall it seems to me that when the AI wins, it isn't because of the lack of FOW. They just can outproduce you. Since there isn't a clear tactic to always get the best of each engagement, you lose by attrition (no defense/terrain bonus on the sea, etc...).

I think we would get a better improvement to the game by added some more tactical elements to sea battles. This we could probably do with the editor. One idea I saw before was sea mines to protect resources. I have tried increasing the engagement ranges of CR & BB beyond visual to give a benefit to using spotters and such.

If we did a non FOW algorithm, it would be something like:

1. Send all units on patrol to find enemies

2. Remember locations of sighted units, with uncertainty circle based on movement

3. Vector attackers toward center of uncertainty circle

Looks great, right? Here is the problem. I could easily let the enemy see my weak units, draw them in, and then pounce with the strong ones. Repeat 100x times, game over. Same thing every game. Now to fix that problem we need to add some learning to the AI algorithm, which is a much bigger project. The current AI is quite good at sending the right units to its battles (too good you might complain). Remember the the opposite is a pushover AI.

Anyway, the point is that maybe with some more tactical elements to sea combat you can get around the FOWless issue in a satisfying manner.

Well, the AI won't be changed. Only small bugs will be fixed in the future.

I'd been thinking about this hoping I could come up with something to improve the game enough to play and have fun. EoS is a great game with a lot of very nice features. All I could think of is explained above. It looks like it's not going to happen, I'm going to give it up and stick with Empire Enhanced. The AI gives an excellent game without cheating. It doesn't have all the bells and whistles EoS has but it's a lot of fun.

Link to comment
Share on other sites

I never get to the point when nukes are around. In some of my scenarios I have given the AI nukes, but they don't use them very well. I figured it wasn't a big deal since most games don't get there.

I find that 1 on 1 vrs the AI you will pretty much always win. It isn't real good about using all it's units to the best of it's ability. When you setup AI teams against you is when it actually gets hard.

I'm putting together a rule set mod with a lot of changes. Heavy air transports, sea mines, changes to air combat, and only transports can capture sea resources. I'll have to see how it plays....

I'm thinking it would be real nice to have a check box on each unit in the editor that doesn't allow the AI to use it. I really doubt it will be able to use the sea mines :)

I guess I could use a unit subset for that?

Link to comment
Share on other sites

You're running into what was one problem with the AI when Empire Enhanced first came out. It included a lot of different units. The AI could produce them and use them but not well. For example, the engineer. The AI.dll would have it build roads but it did so randomly so it was pretty useless and was an expensive unit that the AI would produce.

What I figured out was in order to get the AI to work properly you simply needed to eliminate those units from play. After experimentation the first few years I settled on the same units that Empire Deluxe uses. The AI.dll uses them very well. In order to get a better game you might have to do the same with EoS. Simply don't put any units the AI doesn't use well into the game.

The last few months I've been thinking about making another mod for Empire Enhanced with 3-4 additional ground units. This would improve the land aspects of the game. Probably will need to make one or two of them new. But they would have to be in line with the games AI.dll to work well. The two games are similar that way. The nice thing about Empire Enhanced though is more of the AI is easily available in the script like production choices. And some in the config like production handicaps. In EoS it's all hard coded.

Link to comment
Share on other sites

Actually I was able to create a unit subset called AI units in the same rules. I just disabled the new units in that unitset. When I start the game, I can use the standard unit set and use the "AI unit" set for the AI. Pretty slick, took about 5 minutes :)

If the rules editor had a section for AI scripting to go in, wouldn't that be sweeeet :)

Link to comment
Share on other sites

AI seems to do fine when using the AI units set. This changes games play a lot.

I suspect there's more that could be done using the editor to create mods optimizing AI gameplay. If specific suggestions are made for improving or enhancing the editor and/or scripting AI behavior, Brit may be able to accomodate if the code adjustments are easy enough. That may be the way to proceed...

Link to comment
Share on other sites

One thing that gave me more hope for this game was Brit's answers to my questions in the other thread. According to him, the AI doesn't go after units that are a couple of turns away. It just isn't good enough to respond to units that way partially because although it knows where the unit is, it doesn't know what direction it's moving. Assuming this is true, and Brit should know, then negates to a large degree any problems to tactical situations caused by lack of FOW.

Yet according to Jeep's post he has seen it. And other things.

So for Jeep.

1. Are you sure this is happening consistently? That the AI units go after enemy units 2-3 turns away? Because Brit said it couldn't be. Have you studied this?

2. You have talked about the AI's overwhelming force. Are you sure the AI is getting a huge production bonus? Have you actually studied this or is it ad hoc?

3. Re the AI's production choices, have you studied this or does it just seem there are too many of some rather than others ad hoc?

For Brit.

You made a good point re my op. I countered with a possible workaround to that point. Do you think it would work without causing other problems? Would it be too difficult or too time consuming to implement?

One problem putting all units outside sight range on ignore is it might affect general expansion and movement and slow it down. Would it?

One great thing about EoS is the ruleset. A lot can be done with it.

Link to comment
Share on other sites

Rich,

On #1, it isn't always consistent. The closer you are the greater the chance. I would say maybe 1/3 of the time this happens at around 2 turns away.

#2 What I meant was that when you lose, it is almost always due to attrition. A single elite AI does seem to have a production bonus (and less resource restrictions). Not sure what it is (20% maybe). However, given the fact that the AI doesn't allocate it perfectly is probably evens out. I usually play against more than one AI on teams against me, hence the overwhelming odds....

#3 Just look at the replays of the games. Each AI seems to build to a different formula. Some will build more air, others none at all. For ships some use more subs than others. In general though, I would say the ratio sits at around

if BBs are 100%

CR at 200%

Subs at 350%

DDs at 500%

TRs at 500%

I think the formula is fine, just observing that you don't really need subs to fight this.

Link to comment
Share on other sites

Assuming this is true, and Brit should know, then negates to a large degree any problems to tactical situations caused by lack of FOW.

This is pretty much what Joel Billings said about the WITE AI; it operates under its own "FOW" and even though it knows where everything is it still isn't smart enough to think ahead too far like a human player.

One problem putting all units outside sight range on ignore is it might affect general expansion and movement and slow it down. Would it?

Yeah it very well might. It might cause a whole lot of unforeseen effects, which require time to playtest and assess. I'm a firm believer in the Law of Unintended Consequences. Rich, it's not like anyone is arguing against having a great non-cheating AI that plays by all the rules a human player does, but the bottom line is what the overall gameplay is like. Is the computer opponent competent and challenging, despite the cheats? That's in the eye of the beholder. ;)

Link to comment
Share on other sites

We've gotten past the possibility of a great non-cheating AI that plays by the same rules. It's not going to happen. So what I'm looking at, is it good enough?

Now, I understand about Jeep saying overpowered and that's not really a problem. Plus on Elite the AI might be getting too many bonuses. Maybe Veteran would be better for general play. I don't know exactly what bonuses the different levels give but it would be nice if Brit would post an explanation.

Looking at Jeep's percentages, I think it's pretty good. Maybe a bit too high in some cases but it can be managed in a couple of different ways. Imo SS should be closer to 300%, DD maybe 300-350 and TR 400. But that's just opinion.

I think SS aren't necessary but can be helpful. Apparently even if a TR or BB "sees" a SS 2-3 turns away it will continue because it doesn't respond well and, again according to Brit, units don't flee very well. So if a TR sees a SS a few turns away it won't turn around and go the other way. Therefore having some SS would be fine. If the AI could see all and respond well they would be worthless but that's not the case.

I was giving the AI more credit than it deserved. I just assumed if a TR saw a SS two turns away it would turn around and go the other way. Or if a BB saw a TR 2-3 turns away it would go after it. But it generally doesn't happen that way. And in spite of my explaining this a few times, nobody corrected it. Not Brit or anyone. That's really too bad, it could have saved a lot of hassle and hard feelings.

Link to comment
Share on other sites

1. Are you sure this is happening consistently? That the AI units go after enemy units 2-3 turns away? Because Brit said it couldn't be. Have you studied this?

Wait, what? It wouldn't surprise me if the AI was sending units after enemy units that are nearby (even if they are outside view range).

2. You have talked about the AI's overwhelming force. Are you sure the AI is getting a huge production bonus? Have you actually studied this or is it ad hoc?

The AI does get a production bonus, but it isn't like a 400% production bonus (like they apparently do sometimes in MOO2). I just looked it up, and the Expert AI gets a 30% production bonus.

You made a good point re my op. I countered with a possible workaround to that point. Do you think it would work without causing other problems? Would it be too difficult or too time consuming to implement?

One problem putting all units outside sight range on ignore is it might affect general expansion and movement and slow it down. Would it?

One great thing about EoS is the ruleset. A lot can be done with it.

Actually, I think Civilization 4 does something like that - it keeps track of units with a slightly larger view range and it can see units for a few turns after they've disappeared into the FOW, so it seems like it might not be a bad solution if they're making it work. One of the other issues involves bringing enough units to a fight. For example, if the AI is going to invade an island, it shouldn't bring too few or too many units. With FOW working, it's necessary for the AI to do more patrols (which wouldn't be a bad thing) and potentially anticipate the number of units that a player normally places on an island.

Link to comment
Share on other sites

×
×
  • Create New...