So if my math is right... (character creation)

#11cge0Posted 6/14/2012 7:41:18 PM
I am using the Japanese wiki. I've been working on this for the past 40 minutes, and so far have just completed 71.2% of the operator differences for Spear in Warrior class.
---
http://last.fm/user/kingdomof
Things are bad! Things are very bad: I have it, the filth, the Nausea. ~ Jean Paul Satre
#12cge0Posted 6/14/2012 8:02:10 PM(edited)
http://i49.tinypic.com/2cpyr83.gif

Essentially, something like this would work. Take home lesson: about 50% of the operator combinations may yield trivial results. I don't have data on EX skills and I limited the analysis to warrior class, however, I'm sure different classes may have different initial states or alter the operators.

The wiki is too riddled with holes and too Japanese for me to actually do this any further without being paid. With that said, I have a few papers to read on molecular modeling...

---
http://last.fm/user/kingdomof
Things are bad! Things are very bad: I have it, the filth, the Nausea. ~ Jean Paul Satre
#13phoniex10_Posted 6/14/2012 9:07:33 PM
Thats awesome.
That seems the way i had my elemental stats listed and The combos but it was very few and i never worked on it.
To time consuming ect....
Plus so you guys can know theres no save (Like Agarest war Zero) after the prologue which takes you to the character selection screen.
---
~Unlimited Apphoenix Works~ http://www.youtube.com/user/Phoniexwings
#14cge0Posted 6/15/2012 3:39:47 AM
7*8*3 = 168 Sets of stats, if these assumptions hold. From there on we could predict general trends in terms of elemental stats and stat ranking.
---
http://last.fm/user/kingdomof
Things are bad! Things are very bad: I have it, the filth, the Nausea. ~ Jean Paul Satre
#15Uryvichk(Topic Creator)Posted 6/15/2012 6:31:16 AM
Interesting. So there's actually a fairly small set of meaningfully different characters, or at least sufficiently small that it may be possible to collect?

Also, I don't doubt your math on the 1-pair set, I just can't for the life of me get the same number when representing it graphically. I'm clearly screwing up there in a way I'm not screwing up in the other sets, but if you're correct about the operator analysis (and I don't see why you wouldn't be), there's quite possibly no point bothering if the differences are so incredibly minor that it may be possible to generalize. Still, I'd like to know how I keep getting that one wrong. :/

What interests me about your analysis is that there is an uneven total to the elemental ranks in some iterations. For example, all else being equal, AABBCD is flat-out better than AABBCE thanks to C -> B DEX and an additional Combo rank, and equivalent in elemental rank value to AABBCC (which has an additional Earth rank) and AABBCH (which has an additional Extra rank).

This, combined with other anecdotal data, suggests to me the following: elemental ranks are capped at 6, and it's possible to "waste" them. In other words, AABBCG and AABBCH are exactly the same, but AABBCH has +1 Extra over AABBCG. The only reason I can see this being the case is that AABBC produces 6/5/6/5/6/4/5/3/5/5/6/5 (or something close; G and H may each add +1 to one of these), and G adds +1 to either General, Power, Fire, or Dark while H adds +1 to Extra. Given the knowledge I have of Agarest and the fact that G and H are Scythe and Staff respectively, my guess is Scythe gives +1 Dark and Staff +1 Extra (it's either that or +1 Light). Since Scythe is trying to add +1 to 6, which is apparently the cap, nothing happens.

I suspect choices also can lower ranks, which is why Wind is at 3. Most likely some combination of Sword, Gun, and Spear lowers Wind considerably. There's 12 ranks, 4 physical and 8 magical, so I imagine each weapon has some effect on both types, probably raising and/or lowering one physical rank and raising/lowering one magical rank per pick.

What this means, and what I'm getting at here, is that there are some choices for Weiss that will have higher overall ranks, and since rank is fixed in-game and determines what skills can be used, it's advantageous to have as high a rank spread as possible. So some choices are just better. As the above example shows, some stat spreads are also just better, although cge0 didn't test bases (or else didn't note any difference). Bases are rather trivial but you never know.

My other guess is this: Minimum ranks are probably determined by class pick (weapon selection might also have different effects by class, but I can't test that). I would hazard a guess that a Warrior Weiss starts with fairly high ranks in General/Power/Combo/Range and those either can't go lower than 5 or just generally won't, while a Sorcerer Weiss probably has a better minimum to elemental ranks. Or at least a different initial spread that selections modify.

I know you're not being paid for this cge0, and I wouldn't ask you to do further analysis out of thin air, but what if you had more data? If you let me (or someone else) know where you think we'd see the most meaningful data and have somebody test a representative sample, we might be able to just figure out the direct effect of selections on stats and element ranks, and maybe the priority system for weapon choice, and then I or someone else could simply write a FAQ that tells you how to build Weiss without the over 2000 combinations actually being tested.
#16RyoKaibaPosted 6/15/2012 12:02:24 PM
Oh lord, is this going to be as completely impossible to understand nonsense as Zero's "character creation" system, or is it better?
---
"Hate the sin, love the sinner. But sometimes it's hard to look past all that sin and find the person behind it." -Joshua Graham; Fallout: NV
#17cge0Posted 6/15/2012 12:42:26 PM
Oh lord, is this going to be as completely impossible to understand nonsense as Zero's "character creation" system, or is it better?

I think we're on the path to a decent understanding, to be fair. This isn't rocket science, we're simply attempting a reverse engineering approach that considers common programming practices.

Interesting. So there's actually a fairly small set of meaningfully different characters, or at least sufficiently small that it may be possible to collect?

Well, we'd need to cover the basic operator combinations, ie A - B, A - C, etc, where we can see the full effect of each one. I'll touch upon this further as I address the rest of this post, however, I do not own the game and I merely just stumbled in here on pure curiosity. My knowledge of the game mechanics are limited.

The total number of expected operator combinations is 36 (you can easily graphically represent this, algebra should be applicable in this setting).

Also, I don't doubt your math on the 1-pair set, I just can't for the life of me get the same number when representing it graphically. I'm clearly screwing up there in a way I'm not screwing up in the other sets, but if you're correct about the operator analysis (and I don't see why you wouldn't be), there's quite possibly no point bothering if the differences are so incredibly minor that it may be possible to generalize. Still, I'd like to know how I keep getting that one wrong. :/

Speaking from experience, try simplifying the problem. Only account for the variable 4 position from the 7 remaining. You can then multiply the number of possibilities by 8 afterward to account for the duplicate. If this doesn't work, consider your assumptions in the problem.

What interests me about your analysis is that there is an uneven total to the elemental ranks in some iterations. For example, all else being equal, AABBCD is flat-out better than AABBCE thanks to C -> B DEX and an additional Combo rank, and equivalent in elemental rank value to AABBCC (which has an additional Earth rank) and AABBCH (which has an additional Extra rank).

This, combined with other anecdotal data, suggests to me the following: elemental ranks are capped at 6, and it's possible to "waste" them. In other words, AABBCG and AABBCH are exactly the same, but AABBCH has +1 Extra over AABBCG. The only reason I can see this being the case is that AABBC produces 6/5/6/5/6/4/5/3/5/5/6/5 (or something close; G and H may each add +1 to one of these), and G adds +1 to either General, Power, Fire, or Dark while H adds +1 to Extra. Given the knowledge I have of Agarest and the fact that G and H are Scythe and Staff respectively, my guess is Scythe gives +1 Dark and Staff +1 Extra (it's either that or +1 Light). Since Scythe is trying to add +1 to 6, which is apparently the cap, nothing happens.


This is the part that is most pressing when attempting this analysis. We'd need to look at the effect of operators when we aren't originally capped out at 6! This will obviously skew the data. I will think that the base stats (class stats) play a big role. They must generate the base numbers upon which the operators work on. A quick look at the wiki should confirm this, since Battle-Mage and Warrior elemental stats are extremely different. However, if we can accommodate for this, there may be at least 36 operator analysis sets to examine in order to determine what is going on. At most, there would be 7056 (lol), but I expect it to be around a 108-216 combinations, with greater accuracy if we exceed this, however 216 should be sufficient...
---
http://last.fm/user/kingdomof
Things are bad! Things are very bad: I have it, the filth, the Nausea. ~ Jean Paul Satre
#18cge0Posted 6/15/2012 12:52:25 PM
I suspect choices also can lower ranks, which is why Wind is at 3. Most likely some combination of Sword, Gun, and Spear lowers Wind considerably. There's 12 ranks, 4 physical and 8 magical, so I imagine each weapon has some effect on both types, probably raising and/or lowering one physical rank and raising/lowering one magical rank per pick.

This is probably true. I need background information, as to the Agarest series associations for what corresponds with what weapon.

What this means, and what I'm getting at here, is that there are some choices for Weiss that will have higher overall ranks, and since rank is fixed in-game and determines what skills can be used, it's advantageous to have as high a rank spread as possible. So some choices are just better. As the above example shows, some stat spreads are also just better, although cge0 didn't test bases (or else didn't note any difference). Bases are rather trivial but you never know.

Of course some choices are going to be better than others, they definitely did not sit down and hard code each individual choice, so they cannot eliminate this fault. I don't have data on bases.

My other guess is this: Minimum ranks are probably determined by class pick (weapon selection might also have different effects by class, but I can't test that). I would hazard a guess that a Warrior Weiss starts with fairly high ranks in General/Power/Combo/Range and those either can't go lower than 5 or just generally won't, while a Sorcerer Weiss probably has a better minimum to elemental ranks. Or at least a different initial spread that selections modify.

I'm going to assume the programmers were lazy (I've never met one that wasn't) and simply said "Make the class represent a base set of numbers/stat ranking/base stat variables."

I know you're not being paid for this cge0, and I wouldn't ask you to do further analysis out of thin air, but what if you had more data? If you let me (or someone else) know where you think we'd see the most meaningful data and have somebody test a representative sample, we might be able to just figure out the direct effect of selections on stats and element ranks, and maybe the priority system for weapon choice, and then I or someone else could simply write a FAQ that tells you how to build Weiss without the over 2000 combinations actually being tested.

I suppose niche humor regarding how most scientists are well underpaid isn't very self evident, lol. I wouldn't mind if I had some data (in English...) that had decent examples. A good start would be getting all 36 of the unique linear combinations of operators for each class. This is the bare minimum of data I'd put my stamp of approval on. If you can put it into an excel file, even better. I just don't want to go through stripping the data from the Wiki (especially since it has major holes), analyzing it, and coming up with a half assed answer.
---
http://last.fm/user/kingdomof
Things are bad! Things are very bad: I have it, the filth, the Nausea. ~ Jean Paul Satre
#19cge0Posted 6/15/2012 1:00:16 PM
Honestly, there is nothing special when it comes to me having a magic touch when it comes to data analysis. I'll oversee the project and advise, but I laid out my proposal already, and it's free to use, have fun with it. If push comes to shove and it doesn't work out, then I just need to think of something else. You're equally capable as me at this point.
---
http://last.fm/user/kingdomof
Things are bad! Things are very bad: I have it, the filth, the Nausea. ~ Jean Paul Satre
#20Uryvichk(Topic Creator)Posted 6/15/2012 2:36:38 PM(edited)
I already have an Excel sheet set up and I'll have plenty of free time to tinker once the game is out in English. I'll let you know what I get from there.

To give you some background, the reason we never figured this crud out for Agarest Zero is that, in Agarest Zero, you didn't pick the individual option but a "card" which had two weapons on it, in some combination of 5 (or maybe it was 6). As a result, it was incredibly difficult to isolate the effects of a single weapon type in the manner you suggested by using a locked-in two pair combination with one additional pick and seeing what happens with the remaining choice. Consequently, I only managed to figure out the following things about stats:

1) Stat ranks are altered in some fashion by picking weapons, but it's not as simple a matter as say "2x Greatsword = A -> S STR." I could never get enough out of Zero to get any good numbers from it. I'm sure there are also baseline stat growth rank spreads for each class (Warriors have a higher base STR or VIT, etc.), but again, we could never find them for Zero.

2) This is the part that was confusing; it seems that weapon assignment works on a "priority" system, where a weapon that has fewer picks may nevertheless end up assigned even if something else has more picks. For example, even if you picked a ton of Scythe cards for a Sorcerer Sieghart (who has a high Scythe priority), it was possible to end up with Sword/Rod instead. Sword/Rod appeared to have higher priority... and I never quite figured out what that priority was, even after hundreds of tests. It was roughly along the lines of "if a higher-priority weapon is in the top 3 options, it gets preference even over other choices." Compounding that, I think stat spreads were possibly dependent on weapon choice! It was a mess.

My guess is Weiss operates on similar rules. Warrior for example probably prioritizes Sword first and foremost, then probably Spear, and then who knows specifically. Sorcerer generally prioritizes Sword/Rod (is that even a thing anymore in 2? I mean it's there in the creation screen) and Scythe. I would not be shocked if picking something like Greatsword/Dagger/Dagger/Sword-Rod/Rod/Scythe still produces a Sword-user, for example, since that sort of thing happened in Zero. I don't think it's worth speculating on that until the game's out and I can get more data though.

3) There were just too many Sieghart variations. Actually, now that I think on it I'm not sure if there were more Sieghart options than Weiss options... I think there were 26 cards and you picked 5. Order didn't matter but class choice did. Anyway.

I'm enthusiastic about this project because it's eminently more controllable. Since the elements have all been separated out to individual components, I think it's better off. Also, the character creation this time actually suggests what stats and such are influenced by the weapon selection. I can't read Japanese, but in the character creation video from MasterLL's channel you can see the stat names (STR, VIT, etc.) listed in the description.

So my hope is in two weeks when the game is in my hands, I can get better data on it. Push comes to shove, I have a PVR recorder and free time, so I could just record myself checking all 2000-some combinations, at least until such time as we have the data sufficient to maybe make some educated guesses.

My fervent wish is that they simplified the internal logic of character creation and it really is as simple as something like "Gun = +1 Fire/-1 Wind, Max is 6 Min is 3." But given the studio behind this game, my guess is it's a heck of a lot messier than that.

Awesome work so far though. I'm going to just get some more data when the game's out in America and come back to this thread.

EDIT: I'm also not a scientist, when you say "unique linear operators," can you elaborate on what that is? If I know what you're talking about I'll know what to look for.