|
|
Do you have the wave computation logs? |
|
|
Checking. Note: 716 bombers warped in on my home planet (last worm hole), flew half way across the system, killed a FFgen, then flew back to insta-pop the Home FF, then the home system. (What's the log file name? And location?) |
|
|
They're only generated if you have advanced logging on at the time the wave is initially announced, but they're in the RuntimeData folder named "MainThreadWaveComputationLog.txt" and "AIThreadWaveComputationLog.txt" (both are important). |
|
|
We did not have adv logging on. How do you do that? And then we'll load up an autosave. |
|
|
It's on the advanced tab of the settings window :) |
|
|
Yeah, we found it. Also: congrats on finding a multi-autosave functionality that works. :D I missed that along the way. ^..^ |
|
|
Yea, I remember the days of players pleading for keep-autosaves, and of developers pleading for you to keep them for bug reports ;) |
|
|
Yus. Anyway, here's what it has. 1/19/2011 10:53:33 PM (4.072) ----------------------------------- Performing first CheckWave with size factor of 1.3 on wave at Game Time: 1:04:55 CheckWave: populating count of Bomber with base magnitude of 146 numberUnits = kv.Value * this.WaveSize : 189.52 after applying the ship-type-specific cap multiplier (which includes the unit-cap-scale multiplier), numberUnits : 47.38 after applying CanUseNeinzulRegenerator if any, numberUnits : 47.38 after applying Mark-based multiplier if any, numberUnits : 71.07 after applying at-least-one rule, numberUnits : 71.07 after applying difficulty-based (if >= 8 then 4, if >= 7 then 3, if >= 6 then 2, else 1) multiplier if any, numberUnits : 284.28 numberUnitsInt = numberUnits.IntValue : 284 CheckWave: populating count of Dreadnought with base magnitude of 1 numberUnits = kv.Value * this.WaveSize : 1.3 after applying the ship-type-specific cap multiplier (which includes the unit-cap-scale multiplier), numberUnits : 1.3 after applying CanUseNeinzulRegenerator if any, numberUnits : 1.3 after applying Mark-based multiplier if any, numberUnits : 1.95 after applying at-least-one rule, numberUnits : 1.95 after applying difficulty-based (if >= 8 then 4, if >= 7 then 3, if >= 6 then 2, else 1) multiplier if any, numberUnits : 7.79 numberUnitsInt = numberUnits.IntValue : 7 it's a starship so only adding one rule : 1 (had you going there, didn't we) CheckWave: populating count of LeechStarship with base magnitude of 1 numberUnits = kv.Value * this.WaveSize : 1.3 after applying the ship-type-specific cap multiplier (which includes the unit-cap-scale multiplier), numberUnits : 1.3 after applying CanUseNeinzulRegenerator if any, numberUnits : 1.3 after applying Mark-based multiplier if any, numberUnits : 1.95 after applying at-least-one rule, numberUnits : 1.95 after applying difficulty-based (if >= 8 then 4, if >= 7 then 3, if >= 6 then 2, else 1) multiplier if any, numberUnits : 7.79 numberUnitsInt = numberUnits.IntValue : 7 it's a starship so only adding one rule : 1 (had you going there, didn't we) Wave total ships: 286 TypesForDirectAdd count by type: Bomber => 284 Dreadnought => 1 LeechStarship => 1 TypesForCarrierAdd count by type: 1/19/2011 10:53:33 PM (4.072) ----------------------------------- Performing first CheckWave with size factor of 1.3 on wave at Game Time: 1:04:55 CheckWave: populating count of MissileShip with base magnitude of 123 numberUnits = kv.Value * this.WaveSize : 159.67 after applying the ship-type-specific cap multiplier (which includes the unit-cap-scale multiplier), numberUnits : 39.92 after applying CanUseNeinzulRegenerator if any, numberUnits : 39.92 after applying Mark-based multiplier if any, numberUnits : 59.87 after applying at-least-one rule, numberUnits : 59.87 after applying difficulty-based (if >= 8 then 4, if >= 7 then 3, if >= 6 then 2, else 1) multiplier if any, numberUnits : 239.5 numberUnitsInt = numberUnits.IntValue : 239 CheckWave: populating count of Dreadnought with base magnitude of 1 numberUnits = kv.Value * this.WaveSize : 1.3 after applying the ship-type-specific cap multiplier (which includes the unit-cap-scale multiplier), numberUnits : 1.3 after applying CanUseNeinzulRegenerator if any, numberUnits : 1.3 after applying Mark-based multiplier if any, numberUnits : 1.95 after applying at-least-one rule, numberUnits : 1.95 after applying difficulty-based (if >= 8 then 4, if >= 7 then 3, if >= 6 then 2, else 1) multiplier if any, numberUnits : 7.79 numberUnitsInt = numberUnits.IntValue : 7 it's a starship so only adding one rule : 1 (had you going there, didn't we) CheckWave: populating count of LeechStarship with base magnitude of 1 numberUnits = kv.Value * this.WaveSize : 1.3 after applying the ship-type-specific cap multiplier (which includes the unit-cap-scale multiplier), numberUnits : 1.3 after applying CanUseNeinzulRegenerator if any, numberUnits : 1.3 after applying Mark-based multiplier if any, numberUnits : 1.95 after applying at-least-one rule, numberUnits : 1.95 after applying difficulty-based (if >= 8 then 4, if >= 7 then 3, if >= 6 then 2, else 1) multiplier if any, numberUnits : 7.79 numberUnitsInt = numberUnits.IntValue : 7 it's a starship so only adding one rule : 1 (had you going there, didn't we) Wave total ships: 241 TypesForDirectAdd count by type: MissileShip => 239 Dreadnought => 1 LeechStarship => 1 TypesForCarrierAdd count by type: So its the "71" being multiplied by 4 for difficulty 8. This is way excessive for Bombers (both AIs sent Bombers last time around) as they MURDER forcefields like no one's buisiness, while most other units (while doing reasonable combined damage) can be taken out before they take it all the way down. |
|
|
Thanks for hunting those up :) That's not all the data, though; was there any more in either of those files? And yes, diff 8 is murder. Just want to make sure the other numbers are at least obeying the formulas, etc. |
|
|
AIThreadWaveComputationLog.txt (3,038 bytes) |
|
|
I think one of the things he is asking is whether the ship-type-specific multiplier for bomber fleet ships be reduced just a little bit. These are FAR more dangerous against players than the other two triangle ships after all. (Or maybe it is already low, its just that the diff 8 is "hiding" that with its brutality) |
|
|
Other file attached directly Keith. |
|
|
Dang, a warp jumper and an attritioner. You are in for a hard game there. X_X |
|
|
Nothing wrong with either of them. It was BOTH of them sending a bomber wave. |
|
|
Yea, if this is the same scenario that Chris and I were trying to replicate the desync in, I think these various AI tricks like desyncs and massive bomber waves are actually attempts by the AI to be _merciful_ to you. Because they know the hybrids won't be. Anyway, collating the data for the bomber wave: Starting CreateHomogenousWaveToPlanet at Game Time: 1:04:54 ; Player.AIType: WarpJumper ; Player.AIDifficulty: 8 ; AIProgressionLevel: 105 ; AITechLevel: 1 ; WaveSize: 1.3 aiTypeBasedAIPIncrement : 0 workingShips = ( ( ( AIProgressionLevel * Player.AIDifficulty ) / 2 ) / ( 11 - Player.AIDifficulty ) ) : 140 workingShips *= FInt.FromParts( 0, AILoop.Instance.AIRandom.Next( 800, 1100 ) ) : 146.29 workingShips = Min(workingShips,Player.AIDifficulty * 10) :146.29 numberShips = workingShips.IntValue :146 Inside AdjustNumberShipsFromAIType, multiplier: 1 after AdjustNumberShipsFromAIType call, numberShips :146 numberTech123 = numberShips - numberExplosive - numberCore :146 CheckWave: populating count of Bomber with base magnitude of 146 numberUnits = kv.Value * this.WaveSize : 189.52 after applying the ship-type-specific cap multiplier (which includes the unit-cap-scale multiplier), numberUnits : 47.38 after applying CanUseNeinzulRegenerator if any, numberUnits : 47.38 after applying Mark-based multiplier if any, numberUnits : 71.07 after applying at-least-one rule, numberUnits : 71.07 after applying difficulty-based (if >= 8 then 4, if >= 7 then 3, if >= 6 then 2, else 1) multiplier if any, numberUnits : 284.28 numberUnitsInt = numberUnits.IntValue : 284 (omitting starship and summary info) Feel free (anyone) to comment on any of the steps in that calculation. What I'm interested in isn't so much "that's too many bombers!". As a matter of fact I agree with you; even for diff 8 that's just waaaaay too many bombers on Low caps. One way or another, we're nerfing that. My question is more "where is the calculation going off the rails?". Keep in mind: the only difference if this was normal or high caps is that this line: after applying the ship-type-specific cap multiplier (which includes the unit-cap-scale multiplier), numberUnits : 47.38 Would have a value of roughly 94 or 188, respectively. I need to get some sleep, but will check back later :) |
|
|
I think the real problem is in this line: workingShips = ( ( ( AIProgressionLevel * Player.AIDifficulty ) / 2 ) / ( 11 - Player.AIDifficulty ) ) : 140 Just doing the math that the rest follows, if the AI was Difficulty 10, the end result is 1025 units. I think that the Player.AIDifficulty should only be on one side of the divisor, it's causing a quadratic increase based on the difficulty. Rather than 9 being "50%"* harder than 8, it's actually 250% harder. And Difficulty 10 is another 250% harder than 9. *Made up number. |
|
|
wavesize.png (11,358 bytes) |
|
|
Figured I'd make a graph. The increase is much scarier than quadratic in the higher levels: it's hyperbolic. (this is also one place where we're actualy limited to 10 - 11 would give /literally infinite ships/) I'd probably nerf the bomber wave count (they do tend to be scary in the hands of the AI because human structures has relatively low health, never mind the fewer targets that they need to aim at). I'd also look into replacing the hyperbolic growth with exponential, but that's very much a feel issue and you don't want people complaining about you making Doom too easy. |
|
|
Before we can suggest a new exponential equation, we have to ask what difficulties seem balanced with that one part of the calculation with the current logic, and then fit the new exponential equation to that subset of the values. Of course, figuring out what difficulties are balanced is no easy task. (Note, it can be a disjoint subset. That is, for example, you could say that 6, 7, 7.6, and 8.3 are balanced, but 7.3 and 8 are not. The curve can be fit around those "omitted" values. These are only hypothetical numbers; I personally am fine with the current 7.3 and 8). Maybe even all the values above some difficulty are skewed upwards, and that will require some more rethinking of constant values. Of course, any big change of this sort will probably have to wait until after 5.0, and lots of thinking, math, examples, and play-testing, so don't expect a revisiting anytime soon. |
|
|
Well, no, 11 would give you a division by zero error :) A few things to keep in mind: 1) Difficulty 10 is intended to be _ludicrously_ hard. Not even remotely fair. If people actually start beating diff 10 (unless it was some really specific setup they could exploit) we probably need to rebalance it. 2) Difficulty 9 is intended to be somewhat less totally-out-there, but still not a setting anyone wanting a remotely "normal" game should pick. 3) Difficulty 8 is intended to be a very stiff challenge even to players who really know the game, but distinctly winnable if you know what you're doing. 4) Difficulty 7 is intended to be a challenge but winnable to someone who understands the game and is decently skilled at it. In short: the hyperbolic growth is intentional, particularly at diff 9 and 10. Another thing to keep in mind: 5) Bombers are triangle ships, we're not likely to change their ship cap to something different than fighters or missile frigates. And we're not very likely to give the AI one "cap" for multiplication purposes and a different cap to the humans. If Bombers are that scary then we probably just need to nerf them for everyone. |
|
|
Now that I think about it, implementing that suggestion of warning for force-fields under attack (0002561) will take care of most of the complaints with bombers in AI hands. Regarding the bombers as a triangle ship, I think they are a little too good at their jobs currently for a triangle ship, for both the humans and the AI (their hull bonus multipliers are too high). Maybe make them a little more general purpose (increase attack, decrease multipliers, so they get more DPS in general, but less against structures, ultra-heavy, heavy, and command hull in particular), and re-factor a bonus ship types (or maybe make a new one) that has the effectiveness of the current bombers against those hull types I listed. These bombers would still do well against structures, but not "a wave of them can take out force-fields in 2-3 volleys" good) Furthermore, I would like to suggest making all the triangle ships be very general purpose with slight to moderate advantages in certain situations, and only the bonus ships can have super-effectiveness against. Of course, this "general purposing", if you decide to take it up, would be post-5.0. |
|
|
> Well, no, 11 would give you a division by zero error :) And if you take calculus you'd know that this particular formula approaches infinity as AIDifficulty -> 11. ;) > Bombers are triangle ships, we're not likely to change their ship cap to something different than fighters or missile frigates. I had Mk1, 2, and 3 Fighters (that's roughly 130 fighters) and only managed to kill about 250 of the 700 before they bombed my command station. That's [b]including[/b] the detour they took to go blow up one of my FFgens on the other side of the system. > I think they are a little too good at their jobs currently for a triangle ship They kind of are. They one-hit Frigates, but Frigates take 3 shots to kill a fighter and a fighter takes about 3 to kill a bomber. |
|
|
Bombers deserve some rebalancing, yes, but do you really think their balance is so far out of wack that it deserves postponing 5.0 so they can take some time rebalancing it? (they can always be worked on post-5.0) |
|
|
I have taken calculus; he didn't say approaching 11, he said 11 ;) But yes, if the game actually had a difficulty 10.3 or 10.6 or whatever the results of this computation would become asymptotically more ridiculous. The reason the formula says 11 - diff is that the max diff is 10. It's not a coincidence ;) Anyway, bombers have almost exactly the same cap-dps-vs-non-bonus-targets and cap-dps-vs-bonus-targets as missile frigates. The "bombers one-hit frigates" vs the other triangle-counters taking 3 hits is not necessarily what you want to look at, but rather: how many seconds does it take? Bombers have a long reload. All that said, the fact is that the bonuses that bombers have are more important than the bonuses that frigates have when it comes to taking down heavy defenses. I could just nerf bomber bonuses from 10x to 8x or 7x, though I might need to update my will first. |
|
|
TechSY: I've already redownloaded 3.12 in an attempt to play what I enjoyed, and the friend who got me to buy this game has moved on entirely. The problem lies not in "how far out of whack is the bomber" but in the size and scope of the recent changes: they're all HUGELY significant, global-balance altering, changes. Bombers need the DPS they have (change # 2) because AI Force Fields are so beefy (change # 1), but in so doing the bombers become overpowered in the AI's hands (prompting change # 3) which reduces their usefulness for the player (prompting change # 4). I.E.: If it isn't broke, don't fix it (by which I mean, stop giving the AI awesome candy which requires the entire rest of the game needing a rebalance). |
|
|
Are AI force-fields still too tough compared to human ones? I know spire shield posts are a bit insane right now, but that sort of just needs nerfing to be consistent with the recent nerf to AI ff. |
|
|
I haven't actually seen one recently. In the 1 hour I played last night (with some three or four reloads) we took 5 systems. I never saw an AI FFgen in the three systems I killed. |
|
|
Yea, I don't think nerfing bombers is a great idea right now. But I also don't want to do something fiddly like say that they get some special multiplier to reduce their counts in waves. I guess my main question is: does difficulty 7.6 not give you what you want? Diff 8 is generally going to be "I could die, and I could die really fast". I can nerf wave sizes more but I don't want to make the difficulties below 8 too easy. "The problem lies not in "how far out of whack is the bomber" but in the size and scope of the recent changes: they're all HUGELY significant, global-balance altering, changes." Yes, the game is already quite different than it was in 4.0. And 4.0 was really different from 3.120. But it's not because we enjoy just totally mixing things up as a recreational thing: - From 3.120 to 4.0 we scrapped the "shield" mechanic and replaced it with "armor". This was in response to a general feeling of discontent with shields, and it was a good move on a few other counts. - More importantly, and _massively_ supported by previous customer feedback, we scrapped the type-specific bonuses and replaced them with the hull-types-and-bonuses system. Those two changes totally upended the balance of the game, so 4.0 is frankly not balanced. That's what we've tried to fix in this dev cycle leading up to 5.0. So every single fleet ship has been rebalanced (more than once, in some cases), and a good number of the turrets and starships too. Hopefully we have a foundation of balance now that can simply be built upon rather than torn up. All that said, if you enjoy 3.120 more, I'm glad it's still around :) |
|
|
I have problems with 3.12, but at least I understood where I stood: The AI out-numbered me, had "infinite" resources, had some candy-tech (capturable and non-capturable buildings), but was restricted to the same tools that I had available (i.e. ships). It got some I didn't, I got some it didn't. Post 4.0 that relative equality is gone. The entire Armor mechanic (which I like) is meaningless as far as I'm concerned, because of the special AI units (cough: guardians) who's damage is so significant as to make armor pointless (what's 600 damage lost when each shot does 5000?) and who's armor is so beefy as to make player damage values similarly pointless (what's 5000 damage when each shot gets cut to 20%?). And now AI fortresses are 5x as beefy, AI force fields are beefier, hybrid-hives are Riot Control on crack (and single-target DPS spec'd), etc. etc. etc. The AI now holds the "bigger ships, bigger damage, more units, better units, more diversity, and infinite resources" high ground. The player no longer holds any advantage at all. [b]THAT'S[/b] what I don't like about 4.0 |
|
|
Well, you get warheads ;) But I see what you're getting at. Guardians are due for a rebalance, by the way, their armor values particularly. And yes, armor importance diminishes rapidly as base shot strength increases. Whether a 10-20% increase in survivability (in the case of taking 600 damage off a 5000 damage shot) is meaningless is a bit less clear to me. Hybrids are intended to be a flat-out advantage to the AI, by the way, so no surprise there. Their main design goal was to provide a more "traditional-AI" extra opponent for players who wanted that kind of thing. If you want a scenario where you get major advantages that the AI doesn't, the Fallen-Spire campaign in LotS certainly has some of those. The AI responds by throwing some rather dangerous extra attacks at you, of course. There's also spirecraft and golems, if you're wanting human-only advantages (specific AI types get those, and the hard variants of those minor factions can let the AI get some in the exogalactic-strikeforces, etc, but you get the idea). |
|
|
In short: I'm happy to try to address your concerns, they're just a bit of a blindsiding so I'm trying to figure out what's going on ;) (edit: which I suppose is what 4.0 felt like to you) |
|
|
Bomber numeric stats themselves are fine, the problem is that bomber bonuses apply to 95% or more of the things you really need to have high effectiveness for. Bombers don't need a nerf, they just need to not be the answer almost every time you ask "Ok, what do I need to blow up this hard to take down planet feature X". It's pretty much ALWAYS bombers. Bombers are fine for AI vs human in my opinion since they need at least ONE thing they can send in waves that can kill players (and for the most part barring a huge strategic error on human side, bombers are the only guaranteed dangerous AI wave ship), but for human vs AI it's out of whack. In response to Draco, humans hold both the initiative and the ability to attack with far greater power concentration and aren't anywhere near as stupid as the AI is about attacking (I hope). Difficulty 8+ could probably have that additional multiplier that changes from 3 to 4 just be a straight 3 so there's not such a gigantic jump from 7.6 to 8, and then smoothe out the top level difficulties so it stops increasing at a greater than linear rate because due to the ship cap limitation on players, each additional ship the AI gets in concentration is ALREADY more effective than the last without having to add an increasingly large number per bump up in settings. |
|
|
Yea, from the human-vs-ai perspective in a normal game it doesn't so much feel like I have 3 triangle types: I have bombers to blow up objectives, and 2 supporting types to destroy enemy bombers and destroy the things that destroy my bombers. I don't see that as the end of the world, and honestly I tend to not see it at all because I tend to play with Fallen Spire on (and thus don't really need bombers to take down Core FFs), but it is something I would like to at least make more nuanced later on. |
|
|
To clarify my last statement, the AI going from 10 bombers to 20 is a FAR FAR FAR easier change than from 100 to 200. 100 to 200 is far more than twice as diffult, and is also a much bigger change than 10 to 20. 10 to 20 is a much smaller true difficulty change per mathmatics than 100 to 110 even. This is entirely due to the absolute cap on power for player ships through the ship caps. |
|
|
To add a bit more, it seems like somewhere in the 9s this starts to overflow what the player can mathematically muster without rigging games against the AI through secondary options to shortchange the AI types. No wave mad bombers and stuff like that. The wave multipliers per mk being out of line such that they make higher mk waves stronger per AIP on top of that probably doesn't help either. |
|
|
I just lost a whole giant thing. Fck. Gd dmn security tokens, I swear. "Did you double-submit?" NO. I waited 10 minutes between starting and posting! Basically: The AIP + AIDiff causes an exponential rise in power, whereas players are stuck on a linear path. > they're just a bit of a blindsiding so I'm trying to figure out what's going on ;) Straw that broke the camel's back. Its a bunch of things that have added up over time to make the game unenjoyable. |
|
|
Yea, the security tokens nail me quite often too, extremely frustrating (and why doesn't the browser still have the formdata with my text!!!), I just type my longer notes in my text editor first now. Anyway, it sounds like you're pretty fed-up in general. That's going to happen with just about any game. Sometimes it happens because the game doesn't change (and bugs or balance issues never get fixed, or it works fine but just gets stale with time) and sometimes it happens because the game _does_ change. We've made changes we thought were good for the game, and many like it, but it's not going to be perfect or even good for some folks. Sometimes after a break a game feels worth returning to. If you want to see exponential growth of human power, I again suggest the fallen-spire campaign, but I also understand if you're just tired of the whole thing ;) And Suzera, yes, Diff 9+ is simply... hazardous. Intended for people to whom losing is fun ;) |
|
|
Keith, the other problem is that the AI has so many advantages now the game is near-linear (and the Fallen Spire even more so). There are no longer any choices of strategy for the player. "You must take this planet. Ok, now this one. That one's completely useless to you, but it has that Core Generator Thing you [b]have[/b] to pop. Alright, next planet, then this one...." Fallen Spire? Don't even get me started. You need to take somethink like 3 systems before you even get any rewards, and from there on out the entire thing is hyperlinear. "Take this named system, fly this piece of junk back to your homeworld, rinse, repeat." |
|
|
Well, you can turn the core shield generators off, due to that exact complaint. And the FS is more of a guided thing, that's partly in response to several players feeling like the normal game doesn't give enough direction. Btw, you don't actually have to take the systems with the shards (depending on difficulty and what ships you have it may be tactically necessary, of course). Anyway, I'm not trying to persuade you to play the game against your will, just seeing if there's some existing solution. You sound sufficiently weary that the best service I can provide is to shut up ;) |
|
|
It's not that 9+ is "hazardous", it's just that going from 9 to 9.3 is like the jump in mathemematical difficulty going from 1 all the way to 8 in one step. There's no granularity for people to try something a little harder. It starts jumping from "pretty hard but doable" to "impossible" with a .3 shift instead of "pretty hard" to "a little harder than that" like the .3 shifts in the 7s. |
|
|
Also, chart in the OP is missing the base difficulty wave size multipliers and 8+ all needs to be increased 33%, and something through 6 needs to be decreased 33%. If I am not mistaken: 1-? is 1x ?-6 is 2x 7-7.6 is 3x 8-10 is 4x |
|
|
> Anyway, I'm not trying to persuade you to play the game against your will, just seeing if there's some existing solution. Its more than just that, even (can I use an HTML/BBCode list...?). [list] [*]Parasites are useless (I remember you congratulating me on coming up with a new strategy when I would poke a bees nest in order to replenish my own forces because of leeching). [*]Transports are neigh-worthless now, being unable to even cross a single enemy system (making skipping or "deep"-raiding impossible). [*]Cloaking is worthless, as the AI has anti-clocking on every god be damned wormhole in the game. [*]Neinzul are worthless because you need 3 different ones of them to make any impact at all (the AI counters all ship types all the time and neinzul are hyperspecialized to the point of ineffectiveness, whereas the AI gets 3, 5, even 7 different Neinzul ships every time it uses them). [*]Bullshit guardians. Flak guardian has the armor that makes it "weak" to fighters, but at the same time uses an attack/damage/aoe that's particularly potent against fighters making the illogical choice of Frigates the best way to kill them (note: I realize this data is about 30 versions old, but I have had little chance to observe singular guardians in action since then). [/list] And that's just what I can come up with in 5 minutes, while at work. |
|
|
A linear wave size scaling curve would still lead to a greater than linear amount of difficulty scaling due to the way ship caps limit players. Having it be exponential on top of that just means people have to get A LOT better per difficulty before thinking about moving up by .3 in 8+, and especially 7.6 to 8. There aren't any halfway points to learn to get better in smaller increments past 7.6. |
|
|
Linear increase in ability combined with an exponential growth in difficult reminds me of...this one tower defense game. *Locates* Onslaught 2. I can't find the curves for damage per dollar anymore, but it did work out that the game was basically unplayable after about Wave 400, even at the most optimal turret configurations and combinations. Except one: http://www.pages.drexel.edu/~mmj29/temp/Lyne329b.jpg It was the only tower that had an exponential growth in damage potential, as each tower fed into each other tower doing 125% of its normal damage. Chained together 140 long, each tower only linking to 1 other tower, the damage output was measured in scientific notation (I think it ended up being something like 4 * 10^29th). |
|
|
Player skill gain isn't something you can really measure as linear, or greater/less than linear or whatnot. Mathematically though, even a linear increase in AI ship counts per difficulty leads to a greater than linear mathematical difficulty per AI ship count increase curve, with a vertical asymptote where it becomes mathematically impossible. Evaluating where exactly that asymptote is is of course very difficult in a game like AIW with such varied settings, but even accepting that it is ok that it be there (which means you can challenge the proest of pro AIW players) that still leaves the problem that the granularity of difficultly post-7.6 is very very bad, and the greater than linear ship count for the AI increase exacerbates the problem more than it needs to be. |
|
|
> Player skill gain isn't something you can really measure as linear, or greater/less than linear or whatnot. I meant the units the player has available and the stat increases those units have for teching up. |
|
|
ships.png (34,473 bytes) |
|
|
Added another chart including the AIDiff multipliers (the if >= 6, 2, else if >= 7). I don't know the original math used for the first chart, I based this one off 100 AIP and the debug information from my game. (Also: what's kv.value? It appears to be 1.3 at that line in the debug info, but no info on where it comes from) |
|
|
Correct multipliers (Your second chart looks right regardless though): Difficulty < 6 1 6 <= Difficulty < 7 2 7 <= Difficulty < 8 3 8 <= Difficulty 4 |
|
|
I used the right multipliers I just didn't type them out. Still wondering about this line though: numberUnits = kv.Value * this.WaveSize : 189.52 What's kv.Value? |
|
|
Everything in the logging should match the order of this if I recall correctly: http://arcengames.com/mediawiki/index.php?title=AI_War_-_Why_Do_Enemy_Waves_Get_So_Large%3F#How_Wave_Sizes_Are_Calculated |
|
|
Must be "duration since last wave." Although the debug info doesn't represent that well. |
|
|
kv.Value is in part of the data not included above. A single wave will actually have 3 entries in the main-thread logfile: (this is an example from some days ago) 1/17/2011 3:35:59 PM (4.069) ----------------------------------- Triggering Normal Wave; wave size factor: 1.21; Game Time: 2:34:53 1/17/2011 3:36:00 PM (4.069) ----------------------------------- Receiving AddInboundWave from AI Thread at Game Time: 2:34:53 WaveSize factor: 1.21 Raw Units Dictionary Entries: Bomber => 63 Dreadnought => 1 Fighter => 2 ArmorShip => 1 SpacePlane => 2 1/17/2011 3:36:00 PM (4.069) ----------------------------------- Performing first CheckWave with size factor of 1.21 on wave at Game Time: 2:34:53 CheckWave: populating count of Bomber with base magnitude of 63 numberUnits = kv.Value * this.WaveSize : 75.95 after applying ShipCapMultiplier if any, numberUnits : 37.98 after applying CanUseNeinzulRegenerator if any, numberUnits : 37.98 after applying Mark-based multiplier if any, numberUnits : 56.96 after applying at-least-one rule, numberUnits : 56.96 after applying difficulty-based (if >= 8 then 4, if >= 7 then 3, if >= 6 then 2, else 1) multiplier if any, numberUnits : 170.89 numberUnitsInt = numberUnits.IntValue : 170 (it then goes on to give details on the other types in the wave) |
|
|
To clarify, kv.Value means the value of that key in the "Raw Units Dictionary Entries" list, and just refers to the number after the "=>". "kv.Key" refers to the ship type before the "=>". |
|
|
Still don't understand how "146 units" passed through that line gets to "189 units" (for bombers). |
|
|
Starting CreateHomogenousWaveToPlanet (...) WaveSize: 1.3 CheckWave: populating count of Bomber with base magnitude of 146 numberUnits = kv.Value * this.WaveSize : 189.52 WaveSize = 1.3 kv.Value = 146 result = about 189 Make sense? The WaveSize thing is based on how long it's been since the last wave. I don't actually have a lot of info on how that's computed. |
|
|
Oh, ok. Got it. So essentially I was right, I was just putting things in the wrong place (i.e. juxtaposed the two values to variables) |
|
|
If we're going to talk about how powerful bombers are we do have to look at the rest of the triangle to see how it stacks up. If I recall, bombers actually take a long time to kill compared to the other two, let me see. Mark I ships, cap vs cap, around the triangle, normal caps, epic speed, 4.072: Fighters vs Bombers: 108s Frigates vs Fighters: 60s Bombers vs Frigates: 60s |
|
|
There's a problem with armor not scaling vs damage that evens that out more for normal/blitz speed too. Fighters vs bombers still stands out slower though, just not as much. Should probably take bomberchat to that other issue so this one doesn't have two topics of discussion and this one can be kept more on difficulty/wave size. |
|
|
Ok, these changes are in for 4.073: * Wave sizes reduced further (the previously added divisor changed from 2 to 3) in compensation for the increase in wave size from the fix of the unit-cap-scale-applied-twice bug. Please do give us further feedback (preferably with the wave computation logs obtained via having the advanced logging toggle enabled on the advanced tab of the settings window) on this, whether it's making stuff too easy or if it's still too hard. * The final difficulty-based wave-size multiplier changed from ** "if >= 8 then 4, if >= 7 then 3, if >= 6 then 2, else 1" to ** "if >= 9 then 4, if >= 7 then 3, if >= 6 then 2, else 1" (the change is to the first if). These are very provisional, and perhaps are excessive in the downward direction. I would be surprised if I didn't change this at least once again before the official 5.0 version. But I figure we aren't going to make progress on this unless we go through a few "tweak-and-check" cycles. To summarize: this is a significant nerf to overall wave size and the removal of one of the major "hikes" between 7.6 and 8. Of course that hike is just moved up to 9... but to be honest if you're playing on 9 it's going to stomp you, that's kind of the point. If you want to debate the appropriateness of that we can do that, but let's do it later, I have limited energy :) And to some extent playing on 8 is also supposed to be painfully difficult, but since 7.6 is apparently insufficient challenge for a lot of you (is that correct?) and there is no 7.9 I feel like it's ok to "tame" 8 down a bit so it can basically be what 7.9 would have been. Dunno what 8.3 and 8.6 will be like, they're bound to be pretty brutal. So please let me know what you think. |
|
|
I will agree that 7.6 doesn't seem like a challenge at all (but I will openly admit that I've been playing on 7 or higher since about 2.0, when I first started). |
|
|
Well, I'm guessing that 7.6 in 4.072 would be significantly harder than the 7.6 you remember :) We'll see about 4.073. Anyway, my hope is that players who find that a particular change to wave calculations makes things too hard can simply shift down a difficulty level and get what they were wanting. I think the problem here is that there are a number of folks in the same boat as you who feel like you need more than 7.6, but 8 was just too much of a jump, particularly with the fix to the unit-cap-scale-being-applied-twice bug. I'm tempted to just put it back the way it was before the fix to that... but that would just be wrong ;) I think what was happening is that high-cap waves were _way_ too big and that was one of the reasons (cpu and memory aside) that people just stopped playing on high at all. |
|
|
aip.png (23,003 bytes) |
|
|
Uploaded a new graph with the new multipliers. The jump from 8.6 to 9 isn't actually that large. Only 70% more ships per AIP. Of course, the changes make the number of ships per AIP at Diff 8 be how many we used to have at Diff 7. So...We'll have to see (i.e. is the curve right yet?). (Also note: the Ships per AIP is [i]per AI player[/i], so "a wave" will actually contain twice as many ships as listed). Edit: I also no longer play on high caps. My system simply can't handle it. Even a long game on low caps ends up chugging. Also Keith: I'd love to play a game with you and/or X sometime, and see what kind of game you two enjoy. |
|
|
That chart and the baking data needs to go on the wiki, or at least the community part of it. It is very helpful for helping someone to choose a difficulty, and it really helps me understand the nature of the AI difficulty in this game. Draco, you deserve an Internet, or at least a few Mk. V fabricators. |
|
|
Notice that you're still getting an even worse jump with this setup. It goes from 126->200 from 7.6 to 8 in the old way, and now it is 134->225 at 8.6 to 9. There is actually going to be even WORSE granularity than before because now 8, 8.3, 8.6, 9, 9.3 and 9.6 are all crammed into 9, 9.3, 9.6, 9.8 10 which is one less option to cover that range. The 7.0-7.6 range is now just spread from 7.3 to 8.6, and those were already ok in my opinion. A better idea would be to just make 9+ difficulties 3x multiplier as well, make the 8s go in .2 increments and the 9s go in .1s, and reverse that global wave size drop (or not I guess, I think 10 is still way above what most people would care to even seriously try without cheesing settings even with the wave drop). |
|
|
TechSY: When we get the final chart I'll throw it up on the wiki. Suzera: I suppose that's true. I'll let Keith deal with it though. |
|
|
I'm with Suzera about the removing the "4x wave size multiplier bracket" and just stick with 3x for everything >=7. The hyperbolic growth is more than enough to make lvl. 9 and up insane. Not sure how I feel about the global wave size nerf though. I will have to see for myself what the new 7.6 feels like first and see if it has become too easy. |
|
|
8.6 is the new slightly harder 7.6 is all. 8.3 is a little under. If you found 7.6 too easy but 8 too hard before, 8.6 is a little harder for you. 9 is even harder than 8 was though. |
|
|
We should probably ditch the x4 and spread the x2 out a little. Right now it only effects Diff 6. |
|
|
Pushing the 2x bracket to 4 or 5 or something does seem like a good idea. |
|
|
Really the only thing that matters is the granularity of the options. It's funcational but awkward looking to just have 9, 9.1, 9.2, 9.3, 9.4 etc to provide enough granularity at that point. A text box would be probably the best functional solution, but that might be out of scope or "unsightly" and confuse newbies. Compressing difficulties 1-6 into 1.stuff and 2.stuff would help keep the 1-10 scale more useable, but it would cease to correspond as a nice math function to wave sizes. 9-10 in the upcoming patch probably comprises about 90-95% of the difficulty curve alone, but is nowhere near 90-95% of the choices. The best hybrid would be a box similar to handicaps, where you choose a % difficulty from 0% to 300% or something and make the wave growth function linear. 7 would probably be about 75% on it and the upcoming 10 would be 300%. It would also mean that 200% will be twice as many ships as 100%, instead of now where 8 is twice as many as 7 etc which would match intuition. |
|
|
A linear difficulty "curve" would be most idea, then the jumps from 1 -> 2 "roughly equate" to 5 -> 6 or 7 -> 8. That is: the % idea seems nice. |
|
|
Well, ok, since this doesn't seem to be what y'all want I'll drop the "first divisor from 2 to 3" part of the changes and instead: * Base wave calculation rule from: ** workingShips = ( ( ( AIProgressionLevel * Player.AIDifficulty ) / 2 ) / ( 11 - Player.AIDifficulty ) ) to ** workingShips = ( ( ( AIProgressionLevel * Player.AIDifficulty ) / 2 ) / ( 12 - Player.AIDifficulty ) ) So that should be a more modest impact that has a higher affect on the higher difficulties; to wit: - The change I reverted out (changing the first divisor from 2 to 3) was a general 33% reduction of wave sizes. - This new change is instead a 20% reduction of diff 7 wave sizes, a 25% reduction of diff 8 wave sizes, and a 33% reduction of diff 9 wave sizes. The change bumping the "4x tier" up from 8 to 9 is still there. Right now I'm interested in getting diff 7 through 8.6 where they should be. I'm baffled that someone would want to play diff 9 or higher with the intent to win, though perhaps that's just my own way of seeing things. And Draco18s, I'd be happy to play a game with you; my schedule is highly fragmented (two jobs, two children, etc) but I can generally manage a few hours on Sunday afternoon or evening, depending on the week. You'd probably find the experience frustrating, though, as I tend to be a fairly pause-heavy player and am not all that good ;) I did pretty well in my last game (FS on diff 7), but that sounds like a major step down from the kind of pain you prefer. |
|
|
I'm also considering dropping that first "/ 2" divisor (that was added in 4.072, it's by no means venerable) and instead incrementing the base of the second half ("/ ( 12 - Player.AIDifficulty )") from 12 to 13 or 14. What do y'all think? |
|
|
Well, may as well go for a hat trick of consecutive posts: I think we could look at having a "custom" difficulty option that just pops up a textbox and you can type in a number between 1.0 and 10.0. That would definitely be for post-5.0 as there are potential problems if some of the code is not properly "hardened" against that more freeform setting. |
|
|
What, are you afraid the AI can't work with difficulty 88 and difficulty -7? :D Don't worry, I kinda know what you really mean though. About the proposed further change to the formula, I'm not sure what to say. I am having a tricky time seeing how they would differ at different difficulties. |
|
|
I'll play with some spreadsheet numbers when I get a chance Keith. And shame, Sunday doesn't work for me; I GM a D&D game. |
|
|
Dropping the first divisor and easing the curve would help. There is practically no difference in 1-4 or 5 as is right now, and they'll pretty much just be the same as one number higher is now higher after that change. The top "winnable" base ship count per AIP should probably be around 3.5 to 4 at least with no gaps between settings larger than .2 or .3 up to that point. |
|
|
Bad math post. Redoing. |
|
|
Except, wait, I am leaving factors out. Bad me. |
|
|
I should add that non-integers are fair game for that first line of the calculation, if that helps. |
|
|
aip_curves.png (72,837 bytes) |
|
|
New picture: Includes the old curve, some of the ideas that came up here, plus two I came up with trying to smooth the curve out some and drop that "if >= 6, 2, else if >= 7" chunk. The top one is probably more desired, but uses the expensive POW function (Difficulty ^ 1.5), the lower one just squaring the difficulty. In retrospect I should have done them as series 5 & 6 rather than 2 charts, but oh well. |
|
|
There's nothing wrong with doing Difficulty ^ 1.5 in the CPU-cost sense. Sure, it's gargantuanly costly compared to an add instruction, but we're talking about something that happens once per wave :) Anyway, I'm open to trying a new formula like that, but would like to hear what the rest of you think about it. |
|
|
Point. I wasn't thinking about the frequency of the call when I was fiddling. Anyway, I think Diff^1.5 gives us the happy medium and as an added bonus differentiates difficulties 1-5 a bit better. |
|
|
I love that exponential curve. It keep 9 and 10 ludicrously hard, 7 around the same, but it is much smoother overall. However, that particular equation seems to be fitted for when the 3x->4x jump happened at difficulty 8, not 9. Some tweaking of the constant and the base may need to be done. |
|
|
proposed chart.png (27,079 bytes) |
|
|
Ok, see chart. Original is pre-this-issue. K Prop is the third chart up there from one of your proposals. S Prop is my proposal. Notice that the wave sizes around the 7s area and now spread across a bunch of difficulties and scale more evenly upward up to I used (AIP * difficulty)/(13-Difficulty) Difficulty 1-3 are x1 Difficulty 4 is x1.5 Difficulty 5 is x1.75 Difficulty 6 is x2 Difficulty 7 is x2.25 Difficulty 7.3+ is x2.5 Notice the gap between 7.6 and 8 isn't so drastic, and that there is still roughly at least an equivalent to 7 and what would be 7.4 or 7.5, and covers what used to be a huge gap between 7.6 and 8, and 8.6 to 9. This may be too easy in a non-cheese game for some people since 9 is likely within doable even in original, but I imagine those people could be counted on one hand and there's far more people being held back in a too-easy difficulty in original 7.6 due to the original gap. Just add a small multiplier to the original equation or something. The count gaps are starting to get really big at the high end with this already though. As a bonus, the lower difficulties aren't so samey either. |
|
|
The 8s essentially become the gap between original 7.6 and 8, the 9s become the 8s AND the gap between 8.6 and 9, and 10 is kinda close to original 9. |
|
|
@Suzera For your proposition, 9.6 and 10 are too low. Other than that, it looks good. |
|
|
But then it would have to go up to 11 ;) |
|
|
It's still going to be a pretty large difficulty gap in the proposed 9s even like this too, but at least it won't be a brick wall that plugs you down into a way way too easy difficulty so badly. |
|
|
Make it go to 15 then with .3s and remove all the multipliers! Problem solved. |
|
|
I can see why 9 to 9.3 should not be a sudden jump, but 9.6 to 10 should be. 10 is supposed to be unreasonable. The jump in wave sizes from 9.6 to 10 is supposed to be at least exponential compared to the jump from 9 to 9.3, and indeed it is supposed to be a super-exponential jump in difficulty. 9.3 to 9.6 is more debatable, but not as much of an issue as 9 is supposed to be the reasonable "winnable limit" EDIT: Okay, maybe polynomial wave size jump with a exponential actual difficulty jump. Either way, it is supposed to be a crazy, completely non-smooth jump in actual difficulty. |
|
|
I like mine better. :P |
|
|
Techsy: the 9.6 I proposed is about the old 8.6. The 10 I proposed is almost the old 9. The idea is to remove the huge brick walls the current difficulty settings presents. Those gap closing numbers need to go to a setting somewhere. I'm assuming no more are going to be created, so I appropriated difficulty 6 and the 9s to create a more even spread where the vast majority of players probably are, while leaving sub-5 very very weak. 10 is still probably unreasonable for almost everyone that plays AIW without cheese the settings. |
|
|
Draco: the problem with yours is that twice as many ships is way more than twice as difficult. Doubling the ship count in a wave from 1000 to 1500 can make it go from easily handled to hopeless game over. Brick wall with no inbetween. |
|
|
Ah, so you are aiming to scale the curve down. Now your analysis makes a little more sense. However, what will be the old 10? I still would want an old 10 just for the sheer bat**** insane hardness of it. (the new 10, now about the old 9, would "only" be crazy hard) EDIT: Got the old and new in my last statement backwards. Fixed. |
|
|
The new 10 would be difficulty 10 with the AI having a +300% handicap and hybrid hives and... I'm sure you'll still be able to make a suicide game even with my proposed 10. The curve I proposed is still too harsh in my opinion, but short of changing what the difficulty setting options are available, I'm not sure what's to be done with it besides something like the above, with the goal of keeping sub-5 pretty trivial, and remove the wall people experience at 7.6 going to 8. |
|
|
Did you mean the old 10 would be the new 10 with +300%...? That is, the new 10 would be easier and you would need handicaps to get a similar effect? |
|
|
Yeah, obviously. I don't think that the majority of players should be brick walled at 7.6 or 8.6 just because you can't bear to give the AI 300% positive handicap so it is spitting out tons of ships all over the place for your suicide game. The settings seem like they would be better allocated towards removing those gaps where people can be too good for 7.6 to be fun, but not good enough for 8, or 8.6 vs 9. |
|
|
Then use the squared function I put up. Diff 10, 100 AIP poops out 469 units (approx) per AI per wave versus the 593 that the Diff^1.5 does. Not enough? Try (AIP * POWER(Diff,1.5) / 2) / (12-Diff). Poops out 296 units at Diff 10 for 100 AIP (about equal to the current math on the "8.7" we don't actually have). |
|
|
Well, many of the suggestions here deal with the jump in difficulty between 7.6->8 pretty well. Some even help with the jump from 8.6->9. Some even go the extra mile and smooth the entire curve some while keeping decent difficulty all around. Now comes the tricky task of figuring out which ones would be best and if any constant factors need to be tweaked in these suggestions. |
|
|
I smoothed out the curve and reduced the higher-end numbers some. I don't want to pull down 10 [i]too[/i] far, I just wanted to make the brick wall smaller. My 9 is the current 8.6, and my 8.6 is the current 8. |
|
|
Your Diff^2 jumps over the original 7s too hard in my opinion. The idea is to reduduce the gaps for 7.6 and beyond, not make them all super-gappy instead of just 7.6 to 8 and/or 8.6 to 9. |
|
|
And the 7s felt ok to begin with. I think that's roughly the level of gap at most that should be aimed for original-7+ difficulty. Unfortunately with the quantity of settings available right now, that leaves the 9s and 10 getting bumped down, or brick walls. If the original settings could just be done in per .2 for 8s and per .1s for 9s it wouldn't really be an issue. |
|
|
Agreed. No matter what formula is chosen, the constant parameters need to be chosen such that 7 is around the same as it was before. |
|
|
What? Both of my formulas cross the current 7/7.3/7.6 almost exactly. Diff^1.5 actually comes closer, if being about 1% lower (Diff^2 being about 3% higher). |
|
|
A few more things to toss into the pot: 1) We can strip out the final "if >= 9 then 4, if >= 7 then 3, if >= 6 then 2, else 1" altogether if the results are acceptable, so the difficulty factor is concentrated into the first line. 2) Diff 10 does not _have_ to be coocoo-for-cocoa-puffs. From the first line anyway. If I'm not happy with it's insanity I can take care of supplying the extra oomph elsewhere with a targeted-only-at-10 rule, no worries there ;) Anyway, my goals here: 1) Keep 1.0-7.6 basically how they've been (before 4.072 when I smashed everything), since they seem to be suiting the desires of the people who play those levels. 2) Make 8-9.6 what the people _who actually play those levels_ (which in this discussion is Draco18s and Suzera) want them to be. How we achieve that isn't terribly important to me. |
|
|
I never though yours didn't. In fact, most of these handle the values around 7 about the same. Also, have you noticed that this discussion has easily been the longest of any bug report in mantis? I mean it deserves it; this is one of the most fundamental parts of the game we are talking about. But it still strikes me as kind of funny. |
|
|
Also, I'm not sure where you're getting your numbers for the Diff^2 function. I have that not anywhere near lining up with the original 7s. |
|
|
I have these 3.5 3.98 4.51 vs these (original) 2.63 2.96 3.35 |
|
|
keith said: "_who actually play those levels_ (which in this discussion is Draco18s and Suzera) want them to be" And me. I am playing against one lvl 8 in all my new games now. |
|
|
Diff ^ 1.5 does that perfectly. It raises the difficulty of 2-6 a small amount, but originally they were near-identical in wave size to begin with (on a per AIP scale). At 1000 AIP the numbers move from 19/42/70/107/156/450 to 14/59/122/214/349/551. Average of a 50% increase (mostly at Diff 4, getting double the wave size). @Suzera: my mistake, the Diff^2 does move the 7-7.6 around heavily. Do a 16-Diff, instead of a 14-Diff. Doesn't leave Diff 10 where I'd like it though. |
|
|
I know, the cutoff of the high end difficulty without leaving behind huge gaps is the problematic part of this, but overall I think almost everyone doesn't play 9 or above original difficulty seriously (as in no no-wave mad bombers nor handicap screwiness nor cheats), so for regular gameplay being a bit under 9 probably isn't going to hurt the game any if it allows people to more smoothely scale the difficulty to their skill level. There's always handicaps and minor factions to play with if you really want a suicide game still. Have both a polynomial and the 1/(c-x) is overly complicated for a function like this in my opinion though. After smoothing out the thing to where the nodes between 7 and 10 are somewhat evenly distributed, if the top ends up too low, it's not too hard to just stick *1.1 to *1.5 in the first difficulty equation to raise everything in all difficulties some without making the gaps worse as things go up and down in the high end like messing with changing exponents or the c in 1/(c-x). I do wonder if even original 9 is high enough as a "vanilla" top end though, but from what I have gathered, it is a very tiny minority of players that are that good. Throw on a couple of technologist AI types and some player-negative secondary settings and it's going to get bumped up a lot though and that might be beyond what is possible even with original 9 as a top end difficulty setting. I'm really shocked at how many posts are being generated in this thread though. :) |
|
|
Ok, I'm jumping back and forth between this and AVWW work and some other stuff going on, so humor me with simple statements of what to change ;) My understanding of Draco18s's proposal: "workingShips = ( ( AIProgressionLevel * Player.AIDifficulty ) / 2 ) / ( 12 - Player.AIDifficulty )" to "workingShips = ( ( AIProgressionLevel * ( Player.AIDifficulty ^ 1.5 ) / 2 ) / ( 11 - Player.AIDifficulty ) My understanding of Suzera's proposal: "workingShips = ( ( AIProgressionLevel * Player.AIDifficulty ) / 2 ) / ( 12 - Player.AIDifficulty )" to "workingShips = ( ( AIProgressionLevel * Player.AIDifficulty ) ) / ( 13 - Player.AIDifficulty )" Did I get those right? Please correct if not. |
|
|
Suzera also proposed a much more in depth wave size multiplier for difficulties system. These new values are kind of needed for <=9 difficulties to remain balanced with this new equation. You can see it in 0002572:0009426 |
|
|
Oh, right, Suzera's would include a change from "multiply by (if >= 9 then 4, if >= 7 then 3, if >= 6 then 2, else 1)" to Difficulty 1-3 are x1 Difficulty 4 is x1.5 Difficulty 5 is x1.75 Difficulty 6 is x2 Difficulty 7 is x2.25 Difficulty 7.3+ is x2.5 |
|
|
Yeah, I have a new set of multipliers in the later step to replace the 1/2/3/4 multiplier step function as well. Difficulty 1-3 are x1 Difficulty 4 is x1.5 Difficulty 5 is x1.75 Difficulty 6 is x2 Difficulty 7 is x2.25 Difficulty 7.3+ is x2.5 My proposed difficulty 7 just happens to be exactly the same as the original difficulty 7 too. Edit: Beat me to posting AGAIN keith. |
|
|
Its a little complicated, but try this on for size. It removes the "X-Diff" in the denominator entirely and almost 1:1 matches up with the original equation. =(AIP * POWER(Diff+1,(Diff/4.6)) /2) / 5 9.3 is the only difficulty (higher than 7.6) which is actually harder than the original. 7.6 gets 4% more ships (126 -> 131 @ AIP 100). I'd suggest =(AIP * POWER(Diff+1,(Diff/4.75)) /2) / 5 Just to ease out the jumps at the higher difficulties some. Adding new graph. |
|
|
aip_exponential.PNG (74,015 bytes) |
|
|
Double ships is not double difficulty. When it starts getting up to original 9 amounts of stuff, even 10% more could potentially turn things from "hard" to "you can't win" instead of from "hard" to "10% harder". Going from 10 to 20 ships is easier than 1000 to 1010 mathematically, partcularly when you are at the player ship cap power limit, which I think even original 9 was pushing pretty hard, much less nearly twice that many ships. The absolute number of ships difference is the gap, not the ratio between it and the previous setting. |
|
|
You also still seem to have the wrong numbers for ships/aip for original, at least. Way too low. Edit: All of them seem wrong actually. |
|
|
Then we need to know that maxima. Just how many ships can the player realistically deal with given 4 planets worth of knowledge? If we can approximate that, then we can start to determine where, exactly, our curve needs to lie, even if that means making 7-7.6 easier. More near-linear function that causes a 10% rise in ship counts per difficulty step, topping out at 235 per AI player per wave, at Difficulty 10. =(AIP * POWER(Diff,1.4) * 1.25) / 5 |
|
|
Players have more than 4 planets of knowledge, and it really depends on a lot of stuff like bonus ship choice, whether the AI get autobombs and lots more. It really needs to be more of a line though, not a curve. |
|
|
4 planets of knowledge gives you an AIP of 100, which is the point. At 100 AIP you're only just starting the game, therefore at diff 8ish you should still be able to hold off the waves with the ships that you can research with only 4 planets of knowledge. (4 planets assumes taking 3) |
|
|
But at AIP 110 you'll have 5 planets of knowledge. If 3k knowledge isn't worth 20 AIP that's a separate problem, and is aside from trying to smoothe gaps out of the curve. It would be better to rebalance a smaller part of the game knowledge per planet or knowledge cost of things if it isn't worth 20 AIP, not change the entire difficulty curve around. It looks like you have 3/8 what I do on every figure for some reason. That's fine though, at least we're still looking at the same thing mostly except I think your curve in an earlier diagram was off still. The X^A/(B-X) ones. |
|
|
Might the 3/8ths have to do with the fact that I'm working with low ship caps? I'm basing my math off what occurred in my game. |
|
|
I'm not doing any cap adjustments. I'm just doing steps 1 and 8 since a lot of the rest is either random or based on ship cap, and it's a lot simpler than working back from an in game number. Edit: Difficulty -> ship cap. |
|
|
Fair enough. The curve remains consistent even with the added multipliers. It was just easier for me to look at my "ships sent" and go "eh, 235. That's pretty good for that" based on my experiences. |
|
|
My numbers are just workingships per AIP I guess, assuming floors aren't messing with things. This changes reinforcements too right Keith? |
|
|
I do have a column for workingships. I just didn't divide that by AIP. |
|
|
Oh right, working ships is before the last multiplier thing. Close enough anyway. If this doesn't impact reinforcements as well it's probably not worthwhile to change anything except just implement a custom box so people can type in 8.789 for difficulty. |
|
|
Right, and that other column I have "ships sent" is just mulipliers, so my final value (ships per AIP) should be the same as yours. |
|
|
Was away for a while, just catching up. My gmail gave up trying to keep all the updates in one conversation at 100 replies and rolled over into another one. "This changes reinforcements too right Keith?" This algorithm is in the same method but is only used when the IsReinforcing parameter is false. So no, no impact at all on that. We can look at the reinforcement algorithm if you'd like, but it'd be a post-5.0 thing for changing anything there, as my 4.072 change didn't impact it, etc. Just don't want to mess that up so close to release. Anyway, I'm thinking I'll switch the current implementation to use Suzera's suggestion now for this next release candidate (assuming I beat Chris to the svn). I didn't realize before that our fixed-point math library doesn't have a pow function yet and am taking the path of least resistance on that, and iirc the ^2 method had unintended effects on the 7-7.6 range. Other than that I was leaning towards the ^1.5 method. We'll see how this goes in real testing, and go from there. |
|
|
No probs Keith. We'll see how it works out. |
|
|
Again, I am sorta begging here, but can we get a chance to test this new wave calculation soon? |
|
|
Ok, the code now reads: FInt workingShips = ( AIProgressionLevel * player.AIDifficulty ) / ( 13 - player.AIDifficulty ); and much later (on the main thread) if ( player.AIDifficulty <= 3 ) numberUnits *= 1; // does nothing, I know, just consistency in expression else if ( player.AIDifficulty <= 4 ) numberUnits *= FInt.FromParts( 1, 500 ); else if ( player.AIDifficulty <= 5 ) numberUnits *= FInt.FromParts( 1, 750 ); else if ( player.AIDifficulty <= 6 ) numberUnits *= FInt.FromParts( 2, 0 ); else if ( player.AIDifficulty <= 7 ) numberUnits *= FInt.FromParts( 2, 250 ); else numberUnits *= FInt.FromParts( 2, 500 ); (the old x2, x3, x4 is gone, of course) Those are now the only wave-calc relevant changes from 4.072, let me know if something else was intended. |
|
|
Hey Keith, it just occurred to me that there's another way to calculate X^1.5 without needing an explicit power function. Not that I'm trying to snip in ahead of Suzera's solution. But X^1.5 is equal to X * SQRT(X). If you guys have a SQRT for floats you could do it that way. |
|
|
We have a sqrt for FInt (fixed-int, it's just an Int64 << 12 and thus subdivides each integer into 4096 parts), so yea that would work :) We'll see how Suzera's method works in 4.073 and can consider the ^1.5 method or others (and, very likely, some hybrid) from there :) Thanks to all for the non-trivial amount of work on the math and graphs, etc. |
|
|
What I meant was, reinforcements would need to follow waves or else things would be different AI offense/defense ratiowise starting in the new 8s. Assuming reinforcements followed wave scaling with difficulty before. Since the AI is somewhat weaker on defense anyway though, I guess it's wouldn't be too bad if that was the case since offense would be relatively weaker than defense with this in, leading to potentially scaling it up for a tougher defense compared to offense than before. I don't know the logic behind reinforcements to really say anything more about it. I always assumed they tapped off the wave calcs somewhere. |
|
|
And just for the extra inch of research: http://stackoverflow.com/questions/3518973/floating-point-exponentiation-without-power-function :) |
|
|
I'll remember that if I ever need pow(float,float) and don't have it :) But we're actually not using float here, just FInt. Floating-point is evil-evil-non-deterministic-across-machines-evil ;) Though actually in the AI thread we can use float if we really need to since it only happens on one machine (it just can't send float back to the main thread, but that's easy). |
|
|
True, floats are kind of evil, but I think the basics might be the same for fixed-point rationals (which you use your FInts like, sometimes). |
|
|
Anyone care to summarise the changes this makes? Specifically will my current game now have weaker waves? ( 7.6 difficulty ) Are there any difficulty ranges equivalent to between 7.6 and 8 now? I too often found 7.6 too easy and 8 too much of a jump to get used to. |
|
|
I believe that the 4.073 method will yield: 7 unchanged. 7.3 and 7.6 slightly harder. 8+ much easier (and increasing more gradually). |
|
|
If you're below 8 before, stay the same or MAYBE go down one step. If you're at 7.6 before but it was too easy and old 8 was too hard, 8, 8.3 or 8.6 might be better for you now. 9 is roughly where 8 was. 9.3 now is roughly old 8.3 9.6 is roughlly old 8.6 9.8 is roughly the halfway point between old 8.6 and 9 10 is a little easier than 9 used to be. In summary, everything below 8 is roughly the same, but everything after 7.6 now scales roughly how the difference between 7, 7.3 and 7.6 are instead of being such huge jumps, particularly 7.6->8 and everything above 9. |
|
|
Looks like you should just put in an extreme mode ( dif 11 ) and everything will be sorted! sounds good to me anyway. |
|
|
I'd probably just put in a x2 or x3 or whatever multiplier for diff 10. Selecting Diff 10 is supposed to be the "I lose" button ;) Honestly I'm not worried about that until we're settled on the main calculation. |
|
|
Once we have it I'll have to Nuke It From Orbit (The Only Way to be Sure) and follow that up with a Double Godlike. ;D |
|
|
Instead of trying to make this work with math equations to scale from "win without effort" to "you're going to lose the first wave always for sure" with the current way it is set up, it would be better to just change how you calculate wave, reinforcement etc sizes to a linear function and make difficulty a text box. It's not going to work right with the limited amount of options that exist now with that exponential function in there. The right way in my opinion would just make the step 1 wave size be (AIP * Difficulty %/20) and make the difficulty go from 10% to 400% in steps of 10% and remove the step 8 multipliers. That nets a range from .2 ships/AIP (about double what original 1 difficult was) to 20 ships/AIP (which is about what original 10 was) in difficulty steps of .2/AIP per 10% and do the same thing with reinforcement numbers. As an added bonus, 200% would be twice as many ships as 100%. Really intuitive. It's still going to be a more than linear mathematical difficulty curve though, but far better than what is there now. That's assuming you don't just want a text box for difficulty instead of a bunch of %s like handicap. |
|
|
Where are we on this? Wave sizes still too high? |
|
|
Hard to say, Keith. I got hit with 24 electric shuttles and lost (you saw that). I also got hit with >30 Spire Minirams and lost (another mantis post floating around about how minirams don't suicide properly with regards to damage penalties, IIRC). So I'm not sure. I still feel that about 200 AIP is not the proper AIP level for the AI to get 2x-3x ship-cap sized waves though. Not even on diff 7. It means that 400 AIP is 4x-6x and that 1600 AIP (which used to be winnable) is 16x-24x! (Multiplied by mark level, but even so, half as many units that are twice as strong is still roughly the same, and Mk1 has a 1.0 multiplier and Mk2 has a 0.9, so it's not even "half as many"). |
|
|
Thanks for the reminder on the miniram (and rams, and blades) thing, just put in a fix for that for 5.001. Couldn't find the mantis record for some reason. Anyway, assuming I've managed to whack all the cheap-AI-trick moles at some point, let me know how the waves are. On 200 AIP => 2x/3x-ship-cap-waves, I'm honestly not sure what measuring stick to use there. As in, 150 Bombers on low? How big is your fleet by the point you have 200 AIP? Or is it unrealistic to commit all/half/whatever of your fleet to defend against a midgame (early-ish mid) wave? And 1600 AIP was a lot easier than intended earlier. Basically it used to be that you could get to a level of power where AIP was almost totally irrelevant because the wave sizes would cap out at 2000. Of course, you'd be facing MkIV/V ships, but if you could handle it, you could just nuke everything, etc. Whether 1600 is winnable or "losing is just a matter of time" will depend a lot on game settings, your position, etc. But I think it's safe to say that at that point the AI is very likely to kill you. But 200 and 400 (and even 800, if you've got enough stuff) shouldn't be "you die now" territory. Like you, I'm not really sure where that stands right now, but it seems generally better than it was. |
|
|
I'll hold off on any true comments to the sizes of waves until I see 5.001, but essentially my thoughts (in general) are: > As in, 150 Bombers on low? Generally I've been seeing 200 bombers on low, it fluctuates obviously, but can be as high as 270 even for only 150 AIP (that's from one AI, the other AI getting its own chunk, which if its also bombers is a "you die now" regardless of whatever math we're going to put to the wave size calculation). 200 bombers vs. (tops) 98 fighters generally ends in the bomber's favor. With the newer multiplier this may no longer be true, but is the basis of my complaint. > because the wave sizes would cap out at 2000 True. The last game I played that got up that high in AIP we had so many defenses that I don't think even current wave sizes would have given us pause, simply because we could put down every single turret on one wormhole, had 2 superforts, a radar jammer II, Mk1-4 beam cannons, and two caps of Fortresses 1-3. Waves are one of those things that I think by the Midgame should be possible to hold off with only turrets, otherwise by the midgame your entire operating cycle is stopping a wave and you have no resources or time you can use to expand. Exo-galaxy waves, counter attack waves, CPA, sure, you need your fleet, but those are rare enough that 1) you can do other things and 2) have time to get into position. > But 200 and 400 (and even 800, if you've got enough stuff) shouldn't be "you die now" territory. Agreed. We'll have to see where things are now, of course, but I'd actually like to GET to 400 AIP sometime. |
|
|
I think what Draco is saying that the current extra-ship per AIP ratio is overall too high. I seem to remember from the 3.0 days that the base wave size to AIP relation was some complicated formulla that in the end, was sub linear. Now that it is linear, this has profound impact on the pacing of the game, especially now that there are many explicit changes making low AIP wins much harder. Also, the extra ship per AIP ratios may need to be scales down a bit around 7 and up. However, linear growth is kind of steep. If AIP+x doubles wave sizes, then AIP+nx will (n-1) times the wave sizes, by the very nature of being a linear relation. It might not be possible to find a reasonable x (for every difficulty) that can be reasonable given the current AIP pacing of the game. I think it is possible, but if it isn't, you will have to go back to a sub-linear growth. |
|
|
I'm saying its possible. But essentially, yes. |
|
|
Um, yes to bumping down the extra ship per AIP for 7 and up, or yes to creating a sub-linear relationship between base wave size and AIP? |
|
|
Sorry, sublinear relationship. |
|
|
At first I was thinking you are askin to much, but then I remember a very good point brought up in this report, a linear increase in wave size leads to a super-linear increase in difficulty, thanks to player ship caps and resource caps. (notice, I am speaking asyntopically here, this may not hold in the early game) Thus it would stand to reason that a proper sub-linear increase in wave size would give a linear increase in difficulty (again, might not be completely true for the early game) So your idea has merit. Though it makes me sad that a nice simple linear relation may be too hard to balance. EDIT: it seems like you want a linear or slightly above linear relationship of difficulty to AIP. I'm not sure what chris and kieth want though, they may be happy with a super-linear increase in diffuculty with respect to AIP. Either way, I think they would be wiling to flatten this curve a bit. |
|
|
If someone has a sublinear formula they want tried, I'm all ears. There's also the consideration of perhaps reducing the mark-level-based multiplier to something lower. So MkII getting a 0.6 multiplier instead of 0.9, etc. MkIII having maybe 0.45, MkIV maybe 0.35. That would make it a difficulty increase ((0.6*2 = 1.2) > 1.0, etc) but not as steep ((0.9*2 = 1.8) way > 1.0). It makes sense for the humans to have the "shallower" drop off on mk levels so they feel like they're actually getting something _better_ for their investment of knowledge. But the AIs waves get upgraded "for free" at a certain AIP. That transition point is supposed to be of great concern, but it will become that anyway because fighting the higher mark stuff is much harder than the numbers initially imply. |
|
|
Well, can the suggestions for a sub-linear formula be put on a new report, this one is getting way too long. I would also like to point out that all the hard work done here would not go to waste. The formula that this issue brought forth deals more with the base wave size to AI difficulty. That is to say, it deals with a formula for a different multiplier than the one that would be changed by a new base wave size to AIP relationship. |
|
|
Go ahead and open another ticket. |
|
|
Okay, we can continue the discussion about the base wave size to AIP relationship at 0002771 |
- Anonymous
