Tiering Philosophy - It's that time again

[Espinosa]

There's nothing wrong with giving metaphysical examples to see the limits of his definition--so we don't have to change our criteria simply with the advent of a new game in which it does happen. And it does happen in practice. Jill in 3-13 will never die against Ike but she needs to proc Stun to kill him.

But she'll die on the enemy phase, won't she? Ike will re-equip Ragnell and the rest of his army will help out, if even needed.

[Chiki]

But she'll die on the enemy phase, won't she? Ike will re-equip Ragnell and the rest of his army will help out, if even needed.

Jill's pretty tank by that point since one gives her all their resources, and one should keep Ike underleveled for that chapter regardless. If Ike has around 43 might, and Jill has around 31 def with a Dracoshield and transfers.. well, lol. The subhumans are much too weak to do anything to Jill by that point in the game.

[dondon151]

There's nothing wrong with giving metaphysical examples to see the limits of his definition--so we don't have to change our criteria simply with the advent of a new game in which it does happen. And it does happen in practice. Jill in 3-13 will never die against Ike but she needs to proc Stun to kill him.

please tell me where i said that redwall's proposed scenario will never happen in practice

(hint: i didn't)

[Chiki]

what is more often the case is that if something doesn't work the first time, then you're dead.

So you're saying we should ignore it because it doesn't happen much?

[dondon151]

So you're saying we should ignore it because it doesn't happen much?

for one who has a certificate in logic, you're quite eager to leap to conclusions

i am merely pointing out a problem. ain't nothin' wrong with that.

Edited June 3, 2013 by dondon151

[Chiki]

I was asking you a question, smartypants. Lol.

What did you mean, if not those two things?

Edited June 3, 2013 by Chiki

[XeKr]

It's worth noting that in most cases, exact calculations for a specific chapter are not necessary to distingush 2 characters, so many approximations can be applied. Or else other factors will be more important like stats/growths, which have set distributions instead of a bunch of branching and conditional probabilities.

In some cases strats will transition and be like a 40% 4 turn, ~60% 5 turn (~100% to complete in 5 or less turns) and we'll stop there for 4.6 expected turns. Very close to 100% of cases are captured, and any extremes are weighed so low they don't effect the expected turns much.

In recursive cases like where Grima can heal or in cases of must-have success or death, we can apply geometric series. Or even simplify with N/P calculations. You'll notice in the case of 0.4 success rate for a 1 turn or "failure" (whether by Grima healing to reset the situation or the unit dying to as to literally reset the situation). 1/0.4 = 2.5 as in Redwall's example.

Edited June 3, 2013 by XeKr

[Redwall]

Why not value failure as a whole in general, though?

Suppose we adopt the criteria in your tier list: if we fail more than 40% of the time, then we don't consider the strategy.

In my example, a one-turn clear occurs 40% of the time, and a two-turn clear occurs 24% of the time (so there is a 64% chance of clearing before turn 3). Higher-order possibilities are not accounted for with your way of doing things, so we will restrict discussion to these two possibilities.

How would your tier list criteria treat this: as a one-turn clear, or as a two-turn clear?

that's too theoretical. what is more often the case is that if something doesn't work the first time, then you're dead.

While the assumption of a 0% COD is unrealistic and imposed solely for the sake of simplicity, I don't see any problem with separately treating the chance of death should it be nonzero. I don't see any self-consistent way of "converting" COD into a turncount penalty.

[Snowy_One]

How much does your strategy rely on something specific happening and on certain characters being present? It's great to say 'this chapter can be cleared in two turns', but if it requires a specific setup and a specific strategy to even happen can it really be considered 'tier-worthy'? It's something that's based on one particular PT. Might as well just tier characters based on individual PT's if that's what happens.

[Redwall]

^You're addressing me, right? My example was only limited to that context; I agree that we should consider many contexts (combinations of characters) in tiering. The challenge is figuring out how we weight the different possibilities.

edit: can someone link me to a source describing what N/P calculations are? My only exposure to probability/statistics in the last few years has been in the context of quantum/stat mech, so I'm not familiar with the parlance of statisticians/economists.

Edited June 3, 2013 by Redwall

[Snowy_One]

I'd say 'weight them all equally'. Aside from characters who *must* be fielded you can't guarantee that any one combination, or lack of combination, of characters will be present. I'd think there would be a slight tilt towards characters with supports, but that's about as much as I'd assume.

Edit: I think what's happening is we're seeing two different tiering philosophies clash here. One focused on what happens under ideal conditions and one focused on what happens on less-than-ideal conditions. Might it be wise to create two different tier lists?

Edited June 3, 2013 by Snowy_One

[XeKr]

@Snowy: That's a separate issue, expected turn counts don't tell you everything. Most agree that not all turns are equal, etcetc. There is still debate on things like complexity, varying contexts, etc.

The point is if we use statistics to describe things, we don't run into the issue of whether something is or is not discretely "efficient" or "reliable". We adopt expected turn counts as a tool for measuring efficiency because it combines things we value, low turn counts and high reliability. We use it as evidence when trying to decide if something is more efficient or not and reward more efficient things with a higher ranking. This is independent of "ideal" or "optimal" cases. There are indeed those cases, but we can deal with the others as well, as we're comparing things relatively.

@Redwall: <_<. I was just referring to an alternative explanation for your example which may be more intuitive for others. The method is still essentially the sum of a geometric series, 1/(1-r), we just change the wording to deal with success rates instead of failures/conditionals.

I suppose there are more rigorous statistics to be applied here, but why bother? Ideally discussions are generally accessible and the simpler the better. And as I mentioned, heavily simplified approximations will usually suffice to distinguish most characters.

EDIT: Admittedly, my explanation may be a little too "handwavey", but oh well. >_>

Edited June 3, 2013 by XeKr

[Snowy_One]

I'm not disagreeing with that. It seems that there seem to be two competing schools of thought though. One in which it's important that the player achieve as low a turncount as possible and is gunning for such a thing and one where having a low turn-count is a consequence of power though the player isn't trying for it specifically. The two are VERY different though and the philosophies on how to tier will depend on which ends up dominate. IMO the following need to be firmly established.

1) Turncounts are not the goal. They are ONE tool used to measure how useful a unit is. They are NOT the best tool of necessity nor should they be the only tool used.

2) The team and strategy are NOT established. This is because a team make-up and vary and strategies vary from player to player. A unit should not be ranked high because they can, in one strategy, reliably shave off a load of turns because that strategy may not be employed.

3) There is no such thing as 'too costly' or 'better spent elsewhere'. Only 'they don't benefit as much' or 'they benefit more'.

With those established the question then becomes 'how do we measure turn-count contributions for a character as opposed to a team?'

[Chiki]

Suppose we adopt the criteria in your tier list: if we fail more than 40% of the time, then we don't consider the strategy.

In my example, a one-turn clear occurs 40% of the time, and a two-turn clear occurs 24% of the time (so there is a 64% chance of clearing before turn 3). Higher-order possibilities are not accounted for with your way of doing things, so we will restrict discussion to these two possibilities.

How would your tier list criteria treat this: as a one-turn clear, or as a two-turn clear?

Personally what I had in mind was that there is a perfect strategy for every context, and that we have the entire playthrough and turns planned out (that if Endgame has a 65% chance of being cleared in 1 turn, then we go with that one, and so on).

[dondon151]

edit: can someone link me to a source describing what N/P calculations are? My only exposure to probability/statistics in the last few years has been in the context of quantum/stat mech, so I'm not familiar with the parlance of statisticians/economists.

pffft, N/P is just the expected value of the cumulative number of turns taken if the two possible outcomes are success or failure. nothing special here.

Edited June 3, 2013 by dondon151

[Redwall]

Personally what I had in mind was that there is a perfect strategy for every context, and that we have the entire playthrough and turns planned out (that if Endgame has a 65% chance of being cleared in 1 turn, then we go with that one, and so on).

Well, OK, but there can in general exist stages that have a 60% chance or so of being cleared in (say) two or fewer turns, with both one-turn clears and two-turn clears being very realistic possibilities. My question is, in those cases, whether you consider such clears to be two-turn clears or one-turn clears for tiering purposes if neither the two-turn clear or the one-turn clear occurs in >= 60% of your attempts. I think it is preferable to instead compute the expected turncount in order to account for both possibilities, and possibly accounting for things like three-turn clears.

[Chiki]

Well, OK, but there can in general exist stages that have a 60% chance or so of being cleared in (say) two or fewer turns, with both one-turn clears and two-turn clears being very realistic possibilities. My question is, in those cases, whether you consider such clears to be two-turn clears or one-turn clears for tiering purposes if neither the two-turn clear or the one-turn clear occurs in >= 60% of your attempts. I think it is preferable to instead compute the expected turncount in order to account for both possibilities, and possibly accounting for things like three-turn clears.

Think of it this way. An LTC player (like me) is not going to be satisfied with anything more with the lowest possible turn, as long as it's within the limits of reliability. And I would restart over and over again until I get what I want.

But not all players are like me and aim for perfect play, so I don't really know what to do. Perhaps it might be best to compute for both. It depends on whether we value perfect anal play or not.

[Redwall]

My post was indeed written under the assumption of perfect play. What I was trying to say was that even with perfect play, you can, for example, have Morgan fail to kill Grima on turn 1 due to the RNG but succeed in killing him on turn 2; you can also have him successfully kill Grima on turn 1 due to the RNG. In both cases, perfect play is assumed, but going by your current definition of reliability, at most one of these two possibilities (a one-turn clear or a two-turn clear) is going to count, when I think both (and possibly higher-order possibilities resulting from turn 2 RNG-screwage, and not from imperfect play) should count.

[Chiki]

Okay, by perfect play I also meant that I can't imagine an LTC player settling for a 2 turn when a 1 turn is available within the limits of reliability. I'm sure they would restart.

[Snowy_One]

Think of it this way. An LTC player (like me) is not going to be satisfied with anything more with the lowest possible turn, as long as it's within the limits of reliability. And I would restart over and over again until I get what I want.

But not all players are like me and aim for perfect play, so I don't really know what to do. Perhaps it might be best to compute for both. It depends on whether we value perfect anal play or not.

This is kinda the problem. People who play for LTC are going to end up with completely different teams and strategies than people who don't. After all, their needs are different. A person not playing for LTC may still want to complete the map fast, but doesn't have the same need for rescue-drop and shove/smite chains that an LTC player does. This changes the criteria, focus, and... everything... about how the person will play. Is Mia or Zihark better for wrath? For a LTC player it's a simple question of 'which saves me more turns and might I save more by giving it to Mist so she can 1RKO one guy in one chapter'. For someone who doesn't the answer is less clear. Zihark's AVO (more survivability if he fails and consistant) vs. Mia's vantage (first strike crit) and Mist might not even be on the table.

[Chiki]

This is kinda the problem. People who play for LTC are going to end up with completely different teams and strategies than people who don't. After all, their needs are different. A person not playing for LTC may still want to complete the map fast, but doesn't have the same need for rescue-drop and shove/smite chains that an LTC player does. This changes the criteria, focus, and... everything... about how the person will play. Is Mia or Zihark better for wrath? For a LTC player it's a simple question of 'which saves me more turns and might I save more by giving it to Mist so she can 1RKO one guy in one chapter'. For someone who doesn't the answer is less clear. Zihark's AVO (more survivability if he fails and consistant) vs. Mia's vantage (first strike crit) and Mist might not even be on the table.

Why should we care about people who play casually? I mean, why should I care about some otaku's playthrough in which he played with every character who has big breasts?

[Snowy_One]

Why should we care about people who play casually? I mean, why should I care about some otaku's playthrough in which he played with every character who has big breasts?

The converse question could be asked 'why should someone care about someone who plays only for LTC'. But the thing is, he and you are one in the same. Your criteria is different, but as far as 'tiering' is concerned, you've both thrown actual tiering out in favor of personal criteria (breast size vs. fastest turn completion). You're both using strategies that are limited (yours focusing on low turn counts with a specific team, his likely on pairing them up ) and probably 'specific' in that people going in just from looking at said list aren't going to know how to make the characters good. But, here's the kicker...

In the end the Otaku and you BOTH have to play the game. The Otaku with his romance/fetished supports and you with your minmaxed characters. Which playstyle is more likely to reflect me, the person I suggested FE to at Gamestop, and everyone else I know who plays this game in terms of how it's played? Yours may be 'faster'. It may even be argued to be 'better', but unless I follow the pre-established strategies for a LTC, it's also irrelevant.

If we just wanted LTC there would be no point to a tier list. Just find the characters and team that beat the game the fastest. The whole point to a tier list is that this is NOT always happening, or else there would be no need for tiers, just 'used to beat game' and 'not used'.

[Chiki]

Democracy wins. Simply put, we've decided that what allows us to measure the differences between characters is going quickly. No other rational alternative has been approved, and there never will be one. You nor anyone else is going to change that.

Edited June 3, 2013 by Chiki

[Hawk King]

Chiki, you say a 60% reliability or better should go for tier list discussion. What percent of reliability do you use for your LTC play?

[Chiki]

As far as I can go without losing my sanity.