|
| Thread: Heroes' Stats and Skills Chances |  This thread is pages long: 1 2 3 4 · «PREV / NEXT» |
|
LegendMaker
Promising
Famous Hero
The Metal Specialist
|
posted February 12, 2006 12:22 AM |
|
|
Hold It Right There / Legendary Revelations
First, I have so much to say I don't even know where to begin. You people rock ! Some of you are at least as obssesive as me, and that's saying a lot. So long a path from my original idea to how I'm considering the whole thing right now, and in so short a time ! Mostly because of all your interest, suggestions, ideas, shared past experiences and inspiring point of views.
I'm enthusiastic as **** ! 
Okay, I will try to organize a bit :
1) My Original Project.
I'm still convinced it would work. However, I must agree that it's barely doable. It would require such an enourmous amount of time and efforts !
So much for my original ambitious idea. 
But that's not the end of my Quest, folks. I've made a few discoveries in the last couple of days that could very well change things. Or so I hope.
Anyways... Here it is : 
2) Look At This Beautiful Tree.
When you launch any particular game, the comp attritubes one Tree of skill advancement to that particular game. Don't ask me where it comes from or how the comp obtains it, I don't have a clue yet.
The important thing is that the Tree will remain the same whatever happens in that particular game. Save it at one point, load the save any number of times, make different choices and observe different paths as much as you might want... The Tree won't change one bit.
What I call "a Tree" here needs some explanation. Let's say, you start the game with Jeddite as starting Hero. Save the game. Somewhere in that particular saved game file, there is now some hidden but fixed Skill Advancement Plan that won't ever be changed, whatever the player(s)'(s) choices might.
For the comp, there is already one definite answer to the question "what skills will Jeddite be offered at Level 2, in that game ?". Let's assume in our example that answer is "He will be offered a choice of either Expert Wisdom or Basic Learning".
The comp has already stored the whole Tree of possibilities at this stage, including any and all particular branches the player(s)'(s) choices might lead to explore in the actual game. In pratice, it means that the comp not only knows what choice Jeddite WILL be offered at Level 2, it also knows what the next choice will be (and so on).
In our example, the answer is "IF Jeddite choose Expert Wisdom at Level 2, THEN he will be offered a choice of either Basic Fire or Basic Mysticism at Level 3 ; IF Jeddite choose Basic Learning at Level 2, THEN he will be offered a choice of either Expert Wisdom or Basic Fire".
And that won't change in that particular game, even if you load the early save and replay it a Zillion of times in a row ! It's proven. It's true. It's real. I've checked it out very thoroughly. You can countercheck it as much as you want (be my guest, the more evidence the better).
Guess where I'm going, from there on ?
Yes, you can actually explore the whole Tree and see the pattern right in front of your very eyes, without having to guess anymore !
All it takes is a saved game (a specifically designed map comes in handy, depending on what you're looking for) and some time and efforts (but not quite as much as my original project required) AND it's 100% reliable, not statistics this time around, but FACTS ! ! !
There is no need for any particular protocol for studying this. Because you don't have to do any selection anymore. You can see what happens if you choose A AND if you choose B ALSO ! ! !
(Matter of fact, I even suspect it might be pretty possible for one of you programing masters to extract the whole Tree Info all at once from the actual saved game file. But wouldn't that be "cheating" ?  Rofl).
3) Here Comes A New Challenger. Quote:
U once said to me that you want to put yourself in the mind of the AI. what makes you belive that the Ai would be tempted to pick exp wisdom instead of basic mistycism at lvl 2 for jeddite, or pick basic earth in the same situation?
Tigris, man, If I could, I would kiss you on the cheek for that comment ! You don't even know how GREAT an idea that inspired me !
I can't give you the definitive answer to that question YET. I have some more tests to run to make 100% sure. But I already have great news :
The AI Player (aka AIP) has its own, fixed protocol it uses to choose between two skills. That protocol is the same in any given game (at least for the same hero, I've tested only with Jeddite this far).
Ask the AI Player : "would you rather take Mysticism or Navigation ?" its answer will be "Navigation" any day of the week.
Ask AIP : "would you rather take Logistics or Sorcery ?" its answer will be "Logistics" any hour of the day. And so on.
Why am I so sure ? I've seen it for myself. I've designed this here map to observe the AIP selection method :
The above labyrinth has a separate Pando Box for each Level Up (with the exact number of XP points required for each). Blue Player's Hero has no other choice but to take them one by one, in the exact order I want.
Of course, Blue Player is my good friend the AIP.
While the tester (me, here) watches from there :
I launch a new game with this map, solo mode against the AIP (which plays Blue). I get the first turn. I take my carto, then I save the game (for instance "AIP Skill Selection 01"). Then I press "End Turn". Then I save again (this time "AIP Skill Selection 01a"). Then I load the first saved game ("AIP Skill Selection 01"), multiplayer HotSeat mode. When I'm blue, I take the first Pando Box and note what choice it offers. Then I load the second saved game ("AIP Skill Selection 01a"), same mode. When I'm blue, I open my hero's stats screen and note which choice the AIP made.
This example is just how to observe the AIP's selection method for the first choice (Level 2). The process can of course be extended up to the last choice (Level 21), including any and all choices in between.
Do it a Zillion times in a row (skip right to the last choice, like : first save = very first turn ; second save = right after the AIP took the last Pando Box). Guess what you will get ? Each, any and everytime, you'll get the exact same set of skills. The AIP makes the exact SAME choices no matter how many times you ask it the same question.
After many tests, I've already gathered bits and pieces of the AIP's own protocol. I'm confident I'll find it all out in a matter of weeks (maybe even days).
However, the AIP isn't the same "person" as the comp. Tigris seems to have confused the two in his last post. But they aren't related in anyway. Except that they are both a part of the game program. Different parts, though. So, I already hear some of you (Angel is arguably the leader of the bunch  ) asking me :
"So, Legend... Since it's the way the comp selects the skills to be offered that we're interested in... Why the **** would we even give a **** to the way the AIP selects the skills which the comp offers it ?"
Long question, short answer :
"Because the AIP will do the Testing for us !" 
It's much easier and shorter to determine the AIP's selection choosing protocol than the comp's offering protocol.
Once the AIP's protocol is proven and true, it will be possible to just send the AIP (as Blue player) up against a 150 000 XP points Pando Box time and time again, note what skills it selected and in which order, and get a pretty good idea of what was the Skill Advancement Tree that particular time.
Furthermore it will always be possible to countercheck the results (I already did that myself) and here is how :
Save the game before the AIP's first turn (that's why you MUST always be Red Player). Save it again after the AIP has taken the Pando Box. Load the second save to see what skills it's selected. Then load the first save, but this time YOU are Blue. Take the same Pando and note what choices the AIP was offered by the comp at each stage. By comparing the fully Leveled Up AIP's set of skills (and, crucial info here : which Skill is in which Slot) with the choices you will get, you have a 100% chance of finding out EXACTLY what path the AIP selected each time.
Because once the game is saved, BOTH the comp's offering protocol AND the AIP's choosing protocol are FINAL (at least for this particular game).
Looking forward to read you guys' impressions.
Legend.
____________
LM   
|
|
Binabik
Responsible
Legendary Hero
|
posted February 12, 2006 12:51 AM |
|
|
Quote:
I'm enthusiastic as **** !
Legend, I have a question. How much coffee do you drink? I mean, there are a lot of people who like HOMM, but we don't pee our pants from all the exitement!
New poll
How often do you pee your pants while playing/talking about HOMM
1) Never
2) Rarely
3) At least once a week
4) Daily
____________
|
|
LegendMaker
Promising
Famous Hero
The Metal Specialist
|
posted February 12, 2006 01:01 AM |
|
|
Stay Focused ^__^
Hey, Binabik !
(I liked your last long post a lot. Thx for it. )
1) Pretty obviously my answer would be "daily". Why would I waste such time and efforts for "just a video game" if I wasn't a complete, total, miserable junkie for it in the first place ?
2) No matter how much I may love to joke around here and there myself... PLEASE keep this thread serious ! I seriously doubt I'd survive it if it was to ever end up thrown to the recycle bin.
3) So, what's your overall impression about my Revelations post ?
Legend.
PS : Not that much coffee, actually.
2 or 3 mugs a day tops. But there is some cafeine in those LITERS of Diet Coke I ingerate daily, am I right ?
____________
LM
|
|
Binabik
Responsible
Legendary Hero
|
|
posted February 12, 2006 01:33 AM |
|
|
Quote:
And that won't change in that particular game, even if you load the early save and replay it a Zillion of times in a row ! It's proven. It's true. It's real. I've checked it out very thoroughly. You can countercheck it as much as you want (be my guest, the more evidence the better).
This is what I meant earlier. You can't use restart or reload the game. You must go back to the main menu and start the game from scratch. As I said, random is not random. There's been several discussions about this and nobody has figured out how it works.
My GUESS on this is the way random numbers generators work. The program gets a "seed" from somewhere and runs it through an algorithm. That algorithm generates the random number. Very often the seed comes from the computer's real time clock which (I think) keeps time in 1 ms intervals, so it is constantly changing. The thing about random number generators is that if the same seed is used, it will always come up with the same "random" number. But with a constantly changing clock that's not a problem. But the seed can come from anywhere.
Again it's a guess. But it appears that the seed in HOMM comes, at least partly, from the game/map. If the game/map hasn't changed enough to generate a new seed, the program will come up with the same random number.
I might not be right, but this would explain a lot of the non-random random. It's not that the comp has preselected the entire skill tree. It's that nothing has changed the seed. Once you pick a skill, the seed *might* change. If you reload the map and pick the same skill, you could end up with the same new seed.
Try playing around with it. Go back to the main menu and start fresh. Or with a loaded save game try moving heroes around, building your town, hire new heroes, wait several days, buy a spellbook, etc. See if you get a different skill tree. I'm too lazy to try it right now, but let me know what you come up with.
Also, I've found using events easier than pan boxes when doing large tests. With pan boxes you have to move exactly to the box and go through a dialog screen. With events, you can put max XP in it. Then copy/paste and hold down ctrl-V to fill the entire map. During the test, all you have to do is quickly click anywhere on the map and you hit an event first step in any direction. When doing hundreds of tests this speeds things up a lot.
____________
|
|
Xarfax111
Bad-mannered
Supreme Hero
The last hero standing
|
|
posted February 12, 2006 02:47 AM |
|
|
Well also im not sure if Legend is only making fun out of us, i give some informations that might help.
The "Skilling Tree" will be generated right at the start of the game for every hero. Even for the heroes that are not bought and available in the tavern. All this data will be fixed and packed into a file. Period.
If uve a look into this data it looks approx. like:
Hack Offense Pathfinding Resistance Air Mysticism Logistic Artillery Wisdom Navigation Earth Blacksmith ...and so on.
Means after the start u cant change the tree. U only can influence it by taking or not.
One thing ive noticed that these pattern seems to always the same.
If the first 4 upgrades did come u could be 100% sure that Logistic will come next. That a type of possibities they use with moneyluck games to ensure that they suit to the laws.
So so with 23 possibel skills (28-5) the real chance of getting Logistics again is normally at minimum. So my guess is that there are patterns that are choosen randomly and not real possibilities.
To get better results you should concentrate ONLY on 1 hero (example Hack) and always take the next offered skill. This will give u the informations to get behind his skilling secrets.
One thing: In order to still take u serious, i would like to ask u not to suggest such nonsene-testing in alphabetical order. This will give u nothing, but nonsense and a good waste of time.
____________
|
|
LegendMaker
Promising
Famous Hero
The Metal Specialist
|
|
posted February 12, 2006 05:23 AM |
|
|
Quote:
This is what I meant earlier. You can't use restart or reload the game. You must go back to the main menu and start the game from scratch.
Sorry, but that's horribly wrong, Binabik. I've tested many times, pressing "Restart Scenario" gets a different Skill Tree each and every time ! So, like I said to Xarfax earlier in that thread, I don't see any valid reason for going back to main menu everytime. "Restart Scenario" gets the same results, at least Skill Tree wise. Quote:
Well also im not sure if Legend is only making fun out of us, i give some informations that might help.
Dear Mister X...
I'm definitely not kidding. I'm thankful for this new valuable input you just added to the mix. It pretty much confirms my latest observations and analysis. Which makes me even happier ! Quote:
The "Skilling Tree"will be generated right at the start of the game for every hero. Even for the heroes that are not bought and available in the tavern. All this data will be fixed and packed into a file. Period.
Oh yeah ! Which basically rocks, since the latest version of my project relies on that very fact to ensure it's validity and usefulness (the one I explained in details in my previous long post) ! Quote:
To get better results you should concentrate ONLY on 1 hero example
Hack) and always take the next offered skill. This will give u the informations to get behind his skilling secrets.
Yup, I agree. I've run all of my "AIP protocol" testings with Jeddite, actually.
And I also played around quite a bit with Heroes of whom I emptied ALL skill slots. The results aren't even close to final yet, but this far it strongly suggests that the comp doesn't act any differently wether or not the hero has "starting skills" or not.
My guess is that the SkillTrees are calculated by the comp at the start for all heroes, like you said, and that there is actually such things as a "Level -1" and a "Level 0" for the comp.
Our regular "Level 1" Heroes are nothing but Heroes who started at Level -1 with all skill slots empty, selected one basic skill at Level 0, and either a second basic skill or an advanced mastery of their first skill at Level 1. Then, they are handed to us Players, and the Skill Tree goes on from the specific branches of it the two first "secret" choices has led to.
If I'm right, then the "seed" of any Skill Tree (for any given hero in any particular game) is actually the secret original choice (aka SOC ) that was "offered" to the Hero at "Level 0". Hence, we already know one half of it for any given hero anytime.
For Jeddite, the SOC (aka Secret Original Choice) can only be one of two things : "Basic xxx OR Basic Wisdom" and "Basic Wisdom OR Basic xxx".
Since the AIP always make the same choice when
offered two basic skills when he starts with all
slots emptied (50 tests so far have 100% confirmed that, next batch of 50 should happen by monday)...
Alternatives to Wisdom can only be one of the Skills
that the AIP wouldn't choose over Wisdom. Because we
already know that Jeddite's SOC was Basic Wisdom. Then, since we also already know (still following my hypothesis) that Jeddite has again chosen Wisdom instead of the second basic skill offered at Level 1, that second unknown must also be one of the skills that the AIP doesn't prefer to Wisdom.
My guess is it's precisely what makes any "unique
Hero" "unique" : there's one common Skill Tree for any particular Hero Class, then each individual hero wanders on different branches, according to their respective SOCs. Which gives the slight differences at levelling up from one hero to the next.
When Jeddite is "created" by the comp for any game, the comp reads "skill1 : Wisdom, Advanced" and "Skill2 : none" somewhere. And that actually only means that the two first choices the comp will make are predetermined.
So the comp will take a Skill Tree at some point where Wisdom is offered at Basic two times in a row, for Jeddite. Then it will both times "choose" Wisdom. From that point, Jeddite's unique "Skill Tree" for that particular game will simply go on.
If you still take me seriously, then forget about the
primary skills for the moment (they are most probably
distributed separately following distinct patterns).
On all of the screenshots that will follow please
ignore the Hero Level and Primary Skills, only the
Secondary Skills are relevant.
According to my current theory, the "original" Jeddite (Level -1) would look like this :
Meaning, an empty hero of the Warlocks Hero Class.
Several different ways can lead to the Level 1 Jeddite we know :
1) Sometimes, the comp selects a Skill Tree at some point where the SOC (Level 0)is "BasicWisdom OR Basic Eagle Eye". Such was the case in Emptied Jeddite Test 33 :
Comp "chooses" Wisdom, then next choice (Level 1) will be "Advanced Wisdom OR Basic Earth" :
Comp "chooses" Advanced Wisdom.
The apparent result is your typical Level 1 Jeddite. But it happens to be the specific Jeddite that will give you this choice at Level 2 :
2) Other times, the comp selects this SOC (Level 0) instead (like in Emptied Jeddite Test 28) : "Basic Sorcery OR Basic Wisdom". Comp chooses Wisdom. At Level 1 the choice will be "Advanced Wisdom OR Basic Earth". Comp chooses Wisdom again.
The apparent result is your typical Level 1 Jeddite. But it happens to be the specific Jeddite that will give you that choice at Level 2 :
Again, just a theory at this point. But it seems likely judging by my results so far. Time will tell... Quote:
One thing: In order to still take u serious, i would like to ask u not to suggest such nonsene-testing in alphabetical order. This will give u nothing, but nonsense and a good waste of time.
If I didn't make it clear enough in my previous "Revelations" post, then here it is : I cancelled the whole alphabetical order basec project. Any further testings I'll do will be based on the AIP's protocol and nothing else. Hope this meets your definition of "serious".
Enthusiastically. Legend.
____________
LM
|
|
Binabik
Responsible
Legendary Hero
|
|
posted February 12, 2006 11:45 AM |
|
|
Quote:
Sorry, but that's horribly wrong, Binabik. I've tested many times, pressing "Restart Scenario" gets a different Skill Tree each and every time ! So, like I said to Xarfax earlier in that thread, I don't see any valid reason for going back to main menu everytime.
I said I didn't test it and let me know the results. I said you should go back to the main menu for two reasons. First, its just a good habit to get into when running tests. Second, because if you start with a level 21 hero and let the comp pick all skills when you start, then a restart will NOT be random.
It's a good habit because with some types of tests it does make a difference. Unless it's been tested and you know for sure it doesn't affect the particular test you are doing, you can't assume restart will give you valid results. When designing tests you have to think it through and try to find anything that MIGHT affect the test.
Starting with a level 21 hero is a good example. It would be very difficult to discover how restart affects this. Because it doesn't happen every time. And when it does happen, it's not consistent in how it works. It goes something like this:
Using restart with a high level hero can go into a repeating loop. You won't see the same skill set back-to-back. But you will see a repeating pattern that repeats itself approximately every 3-8 times. Also, this pattern won't start right away. You may have to do several restarts before the pattern begins.
For example, call the first skill set "set 1", the second "set 2", etc. The pattern may look something like this:
random
random
random
random
set 1
set 2
set 3
set 4
set 5
set 1
set 2
set 3
set 4
set 5
set 1, etc.
These skill sets will repeat several times. If you restart enough times, it MIGHT break the cycle. But there is a good chance a whole new cycle will begin. Going to the main menu will definitely break the cycle, but a new one might start afterwards.
Starting with a high level hero is not something you would normally do. But the point is, it would be very hard to find something like this. I know of two other places this same thing happens....the heroes in the tavern and the spells in the guild. As a matter of fact, I've seen all three of these things happen at once. For example if you changed the above pattern to
skill set 1, hero set 1, spell set 1
skill set 2, hero set 2, spell set 2, etc
All three things will repeat together. But there's more. If Gurnnisson is in the tavern in set 1 with a ballista, and also in set 4 without a ballista, that will repeat. If level 2 dwelling is built will repeat. And I can't remember for sure, but I think the main hero’s primary skills also repeat. If that's right, there's a total of 6 things repeating together as a group. That's FAR from being random. Again, this won't happen every time, but it's fairly common.
BTW, you mentioned testing Jeddite and getting wisdom at level 2. Instead of letting the comp pick, do it yourself and turn down wisdom. I'm willing to bet it will be offered again at level 5 (or sooner). The pattern for wisdom with a magic hero is it will be offered every third time (or sooner). If you read my post (on page 1???) about wisdom for a might hero, it works the same way for a magic hero. Only it's every third time for a magic hero (twice as often as a might hero). I haven't tested for a magic school with a magic hero. But if it's half as often as a might hero like wisdom, then it would be offered every other level (even number levels).
Also the pattern with wisdom doesn't matter if the magic hero has had wisdom deleted in the editor, or if you start with a magic hero, like necro, without wisdom. Either way it will be offered at level 3 or sooner. If you turn it down, it will be offered at levels 3,6,9,12,15,18 or sooner.
____________
|
|
Russ
Promising
Supreme Hero
blah, blah, blah
|
|
posted February 13, 2006 04:44 PM |
|
Edited by Russ on 13 Feb 2006
|
Quote:
My GUESS on this is the way random numbers generators work. The program gets a "seed" from somewhere and runs it through an algorithm. That algorithm generates the random number. Very often the seed comes from the computer's real time clock which (I think) keeps time in 1 ms intervals, so it is constantly changing. The thing about random number generators is that if the same seed is used, it will always come up with the same "random" number. But with a constantly changing clock that's not a problem. But the seed can come from anywhere.
Again it's a guess. But it appears that the seed in HOMM comes, at least partly, from the game/map. If the game/map hasn't changed enough to generate a new seed, the program will come up with the same random number.
I would add just a little bit here. Here is what I think happens:
The map is based on the random seed, not the other way around (it would require a very complex and extremely unnecessary algorithm to code it the other way around). Each new game (or restart) a new random seed is generated. This random seed is used to determine the AI groupings/upgrades on certain map squares, or even the map generation pattern in generated maps. Like you said, the skill offering algorithm uses this seed, the existing skills, the skill table and some exceptions (such as wisdom/magic school exceptions) to come up with the new skills. So, since the seed is the same for every game and the skill table along with the exceptions are always constant, the skill tree will remain unchanged in a particular game unless you pick an external skill from somewhere (such as university) which will change the pattern since the existing skills are taken into account as well.
I would assume that the random seed is a 32 bit integer which means there are approximately 4 billion possible skill trees. So, we should concentrate on trying to figure out exactly which algorithm it uses. If I had much more spare time on my hands, I'd probably disassemble it and try to find the algorithm in the code, but that "find needle in the hay" task would probably be even more time consuming than testing, so the best we could do is try to figure the algorithm using our test results and our common sense as the guidelines (i.e. if something would be too complicated to code, we could assume that they did it the other way )
Btw, have anyone tried disassembling homm3.exe yet? If so - feel free to share your findings if you have any.
|
|
LegendMaker
Promising
Famous Hero
The Metal Specialist
|
posted February 14, 2006 03:59 AM |
|
|
Jeddite Sucks / Legendary Genealogy Hangover
Hmmpff
Well, tonight my enthusiasm isn't what it used to be. Happens often when doing research, though. Encouragin results generates enthusiasm. Depressing results, on the other hand... Not as much.
Anyways.
Here is what I wasted this evening doing :
I came up with the oh_so_brillant_idea of using an actual genealogy software to draw a full tree by teditiously repeating big time boring clicks and such... countless times. A genuine genius, am I not ? Ahem.
Okay, so an actual Skill Tree (neither stats nor probabilities, an actual unmasked Skill Tree) looks exactly (not approximately) like this :
Squares are "A" Skills (the ones you'll find on the left slot), circles are "B" Skills (the ones on the right slot). Any choice offered can give two distinct choices on next level, which each separate in two more possibilities and so on.
I couldn't get farther than Level 6 (64 Skills in 32 couples). I did try Level 7, but typing 128 times hated words such as "Mysticism" would kill what's left of me.
Here is this Tree's AIP Digest :
LOFAO, aren't we ?
Well, not as funny as you might think. Actually not so funny at all. Check what the possible choices were :
And just to make the results easier to read, here is the selection of All the Skills that were offered at Basic this time around :
I was hoping I'd find actual loops. I didn't. Yet, I did find many, many clones amongst the choices.
Twins :
Quadruplets :
And even an almost whole family of clones (except the two slight variations I emphasized with arrows here) :
"So, Legend... What's your conclusion ?"
Well, three rather important points :
1) There must be loops. It would take an army of computerized ants to verify any further, though. By Level 21, it's precisely 1,048,576 entries to compare ! Still better than Russ' 4 Billions, I guess. If anyone feels like it, I'll gladly hand all my test files to that person (or entity ?) !
2) Jeddite is pure ****. Time to face it. 
3) If you wanna die, I know a fairly quick way : you say the word "genealogy" to my face. Simple as that ! LMFAO
Legendary Rest.
____________
LM
|
|
dimis
Responsible
Supreme Hero
Digitally signed by FoG
|
posted April 20, 2006 09:07 PM |
|
|
First question
One question just to verify some things ...
"Say you get offered a magic skill on level 4 which you take it. What happens when hero reaches level 8?"
____________
The empty set
|
|
maretti
Responsible
Supreme Hero
|
|
posted April 20, 2006 09:29 PM |
|
|
You might be offered a magic skill but you cant be sure. If you are unlucky you will be offered magic again as late as lvl 16 or 17.
____________
Crag rules, Orrin and Ivor suck
|
|
dimis
Responsible
Supreme Hero
Digitally signed by FoG
|
posted April 22, 2006 01:57 PM |
|
Edited by dimis on 24 Apr 2006
|
On the "Always New Skill Advancement" policy
I don't know if there is a bug in my thoughts and the implementation that I made or we indeed know little about the internals of hero-levelling up and offered skills, but regarding the "Always New Skill Advancement" policy I can briefly give three examples:
CRAG-HACK
AIR MAGIC : 550187845632 paths out of 1198794632832.
ARCHERY : 414135529920 paths out of 1198794632832.
ARMORER : 370776222144 paths out of 1198794632832.
ARTILLERY : 453665864448 paths out of 1198794632832.
BALLISTICS : 453665864448 paths out of 1198794632832.
DIPLOMACY : 78199867584 paths out of 1198794632832.
EAGLE EYE : 148864979520 paths out of 1198794632832.
EARTH MAGIC : 550187845632 paths out of 1198794632832.
ESTATES : 148864979520 paths out of 1198794632832.
FIRE MAGIC : 397320576192 paths out of 1198794632832.
FIRST AID : 78199867584 paths out of 1198794632832.
INTELLIGENCE: 78199867584 paths out of 1198794632832.
LEADERSHIP : 323121044160 paths out of 1198794632832.
LEARNING : 270648335232 paths out of 1198794632832.
LOGISTICS : 414135529920 paths out of 1198794632832.
LUCK : 212777523648 paths out of 1198794632832.
MYSTICISM : 212777523648 paths out of 1198794632832.
NAVIGATION : 148864979520 paths out of 1198794632832.
NECROMANCY : 0 paths out of 1198794632832.
OFFENSE : 1198794632832 paths out of 1198794632832.
PATHFINDING : 453665864448 paths out of 1198794632832.
RESISTANCE : 370776222144 paths out of 1198794632832.
SCHOLAR : 78199867584 paths out of 1198794632832.
SCOUTING : 453665864448 paths out of 1198794632832.
SORCERY : 78199867584 paths out of 1198794632832.
TACTICS : 453665864448 paths out of 1198794632832.
WATER MAGIC : 0 paths out of 1198794632832.
WISDOM : 1198794632832 paths out of 1198794632832.
GURNISSON
AIR MAGIC : 4922243424 paths out of 11162570688.
ARCHERY : 3393006624 paths out of 11162570688.
ARMORER : 3015267120 paths out of 11162570688.
ARTILLERY : 11162570688 paths out of 11162570688.
BALLISTICS : 3742466112 paths out of 11162570688.
DIPLOMACY : 606977280 paths out of 11162570688.
EAGLE EYE : 1167828624 paths out of 11162570688.
EARTH MAGIC : 4922243424 paths out of 11162570688.
ESTATES : 1167828624 paths out of 11162570688.
FIRE MAGIC : 3500464464 paths out of 11162570688.
FIRST AID : 606977280 paths out of 11162570688.
INTELLIGENCE: 606977280 paths out of 11162570688.
LEADERSHIP : 2606688000 paths out of 11162570688.
LEARNING : 2164543776 paths out of 11162570688.
LOGISTICS : 3393006624 paths out of 11162570688.
LUCK : 1685943072 paths out of 11162570688.
MYSTICISM : 1685943072 paths out of 11162570688.
NAVIGATION : 1167828624 paths out of 11162570688.
NECROMANCY : 0 paths out of 11162570688.
OFFENSE : 11162570688 paths out of 11162570688.
PATHFINDING : 3742466112 paths out of 11162570688.
RESISTANCE : 3015267120 paths out of 11162570688.
SCHOLAR : 606977280 paths out of 11162570688.
SCOUTING : 3742466112 paths out of 11162570688.
SORCERY : 606977280 paths out of 11162570688.
TACTICS : 3742466112 paths out of 11162570688.
WATER MAGIC : 0 paths out of 11162570688.
WISDOM : 11162570688 paths out of 11162570688.
ORRIN
AIR MAGIC : 8443191168 paths out of 22541901120.
ARCHERY : 22541901120 paths out of 22541901120.
ARMORER : 5076331200 paths out of 22541901120.
ARTILLERY : 5076331200 paths out of 22541901120.
BALLISTICS : 7313028480 paths out of 22541901120.
DIPLOMACY : 4210288320 paths out of 22541901120.
EAGLE EYE : 2266045920 paths out of 22541901120.
EARTH MAGIC : 5980489344 paths out of 22541901120.
ESTATES : 5878841760 paths out of 22541901120.
FIRE MAGIC : 3180066048 paths out of 22541901120.
FIRST AID : 2266045920 paths out of 22541901120.
INTELLIGENCE: 1176317760 paths out of 22541901120.
LEADERSHIP : 22541901120 paths out of 22541901120.
LEARNING : 4210288320 paths out of 22541901120.
LOGISTICS : 5076331200 paths out of 22541901120.
LUCK : 3275398080 paths out of 22541901120.
MYSTICISM : 2266045920 paths out of 22541901120.
NAVIGATION : 7313028480 paths out of 22541901120.
NECROMANCY : 0 paths out of 22541901120.
OFFENSE : 6622835520 paths out of 22541901120.
PATHFINDING : 4210288320 paths out of 22541901120.
RESISTANCE : 5076331200 paths out of 22541901120.
SCHOLAR : 1176317760 paths out of 22541901120.
SCOUTING : 4210288320 paths out of 22541901120.
SORCERY : 1176317760 paths out of 22541901120.
TACTICS : 6622835520 paths out of 22541901120.
WATER MAGIC : 10606222080 paths out of 22541901120.
WISDOM : 22541901120 paths out of 22541901120.
Now, how did I get these results? I made a program scanning the entire search space of available skills when "Always New Skill Advancement" (ANSA) policy is in effect, i.e. you always choose the first new skill you are offered. I 'll soon make another thread with a description of it, where we can talk about the internals and the various algorithms implemented; of course you will have a copy of the source code, since this is going to be an open-source project.
But please give me some time to settle some things. I 'll edit this post again (add a reference) once the new thread is made.
EDIT:
Thread is ready and you can find it here. Awaiting comments and ideas there so that we don't spam this thread.
EOE.
According to the above the expected probability for Tactics/Crag-Hack combination is 37,8435015 %.
Moreover, reagrding Gurnisson, the expected probability for Logistics when one follows the ANSA policy is 30,3962834 %.
But these numbers are not even close to what Angelito encountered in his tests.
____________
The empty set
|
|
angelito
Honorable
Undefeatable Hero
proud father of a princess
|
|
posted April 24, 2006 08:48 AM |
|
|
Because i never follow the ANSA rule maybe?..
____________
Better judged by 12 than carried by 6.
|
|
dimis
Responsible
Supreme Hero
Digitally signed by FoG
|
posted April 24, 2006 12:10 PM |
|
|
Quote:
Because i never follow the ANSA rule maybe?..
I thought you did some tests with Crag-Hack and Gurnisson following this strategy. Have I misinterpreted what you've written some posts earlier?
____________
The empty set
|
|
angelito
Honorable
Undefeatable Hero
proud father of a princess
|
|
posted April 24, 2006 12:22 PM |
|
|
I thought u refer to game experiences.
But my tests were very short ones. I leveled them both up 30 times. This is for sure not enough for statistical values. It was only made to show how hugh the differecne between "prefered skills" and "ANSA" is.
____________
Better judged by 12 than carried by 6.
|
|
dimis
Responsible
Supreme Hero
Digitally signed by FoG
|
|
posted April 25, 2006 06:19 PM |
|
|
I realized today morning that the results I posted above regarding Gurnisson, Crag-Hack and Orrin are all trash. Reason is that I had a logical error on the preliminary version of the ANSA executable and as a result all computations were wrong. Problem was that I treated heroe's skills on level 7 as I ought to on level 6 (wisdom exception) and also for a similar reason I offered Magic_Schools only on level 5 instead of level 4. Moreover, I had the problem with branching factor regarding Wisdom on the transition from level 5 to level 6, which I discuss on the other thread.
Soon with correct results on probabilities only.
- dimis -
____________
The empty set
|
|
dimis
Responsible
Supreme Hero
Digitally signed by FoG
|
|
posted April 26, 2006 10:15 AM |
bonus applied. |
Edited by angelito at 19:16, 11 Nov 2009.
|
ANSA policy and respective results
For more info on obtaining these results, check out this thread.
The excel file containing these results is here.
- dimis -
Edit by angelito
Very nice. This was exact what i was looking for. Now everyone can look up "his" favorite heroes. Well done dimis! +qp applied
____________
The empty set
|
|
angelito
Honorable
Undefeatable Hero
proud father of a princess
|
|
posted April 26, 2006 01:04 PM |
|
|
Nice work here dimis.
One questions for the better overview:
Is it possible to put every class in a excel table and then post a pic of each table?
............Air / Artillery / Balistics / ......
Tazar......30,5 / 38,7 / 26,2 / .......
Gurnisson..30,5 / 100 / 28,3 / ....
Krellion......../...../....../...........
Shiva.........../...../....../..........
Hope u get my point though...
Would be much easier to compare different heroes that way...
____________
Better judged by 12 than carried by 6.
|
|
dimis
Responsible
Supreme Hero
Digitally signed by FoG
|
|
posted April 26, 2006 01:16 PM |
|
Edited by dimis on 28 Apr 2006
|
excel related
yes it is. I can write a small script for this. I believe excel documents are plain documents and columns are decided by "TAB-spaces", right? (sorry, but I 've stopped using excel the last 4 years so I don't remember much ... ) If that's the way things are, I can automate the process easily and generate the table you want.
I was also thinking about making graphs (pictures) for each Class of Heroes, so that someone can easier see the distinctions for each skill between heroes of the same Class.
EDIT - 26 April 2006:
Yeah, no problem with making a spreadsheet. Tab-spaces are enough to generate a good file for calc (equivalent to excel). But it's going to take some time. After all I have to wait long for heroes starting out with just one skill (Rashka was an exception from those starting with 2 skills because he already had wisdom and thus no limitations on branches leading to his 6th level - total time: 1799.75 seconds).
EOE
EDIT 2 - 28 April 2006:
You can view above an excel (actually calc) snapshot on heroes' skill chances when following ANSA policy. Thanks for the idea Angelito. Numbers are much more handy this way.
EOE
____________
The empty set
|
|
Binabik
Responsible
Legendary Hero
|
|
posted April 28, 2006 11:21 PM |
|
|
Dimis: I can't open that xls file. Could you upload a delimited txt file, like tab or comma delimited?
Also, how would I calculate getting one or more skills from a group of skills? For example with Fafner, the first four skills of Air, Archery, Armorer and Artillery. What would be the odds of getting at least one of those four, 2/4, 3/4 etc.? I don't need an answer to that specific group of skills, but a formula to use for any group of skills.
Testing must still be used to find out how skill selection works. The ANSA table shows expected results IF we know all there is to know on skill selection. Through testing, we would have to look for results that are significantly different from the expected ANSA results.
Xarfax mentioned earlier that there may be skill groups. Some of the testing I've done shows the same thing. I haven't tested enough to say for sure, but the results look suspicious.
Basically it looks like the AI is attempting to offer a variety of skills. For example it tries to create a balance between fighting skills, map skills and "other" skills like estates or learning. I haven't been able to identify how many groups there are, or which skills are in each group. But it does look like it tries to create a balance.
Also, I was discussing with Legend earlier about the difference between doing a restart and going back to the main menu and starting a new test. I've discovered that doing restarts DOES go into a loop. So test results doing restarts are not valid. However, doing restarts does create some interesting ideas for further testing, which I won't go into right now without verifying some things.
____________
|
| |
|
|
