Har far to train UC

[ajon]

So its common to train a weapon to min delay first thing then start branching out to other skills. How does this work with UC as min delay is at 27 right? How high do you train it before starting on fighting/dodging etc. or is it generally more micromanaged and switched on and off often with other skills?

[Sandman25]

  Code:

Dungeon Crawl Stone Soup version 0.16.1-11-g651b585
Attack: Human Monk vs. yak (4000 rounds) (2015/04/24/19:10:54)
Human Monk: XL 1   Str 11   Int 10   Dex 20
Unarmed, Skill: Unarmed Combat
yak: HD 7   AC 4   EV 7

Unarmed Co | AvHitDam | MaxDam | Accuracy | AvDam | AvTime | AvSpeed | AvEffDam
         0 |      0.5 |      3 |      73% |   0.4 |   100  |  1.00 |      0.4
         1 |      0.8 |      7 |      73% |   0.6 |    98  |  1.02 |      0.6
         2 |      1.2 |      8 |      74% |   0.9 |    96  |  1.04 |      0.9
         3 |      1.6 |      9 |      75% |   1.2 |    94  |  1.06 |      1.3
         4 |      2.0 |     11 |      74% |   1.5 |    93  |  1.08 |      1.6
         5 |      2.6 |     14 |      75% |   1.9 |    91  |  1.10 |      2.1
         6 |      3.1 |     13 |      76% |   2.4 |    89  |  1.13 |      2.7
         7 |      3.6 |     16 |      75% |   2.7 |    87  |  1.15 |      3.1
         8 |      4.1 |     17 |      76% |   3.1 |    85  |  1.17 |      3.7
         9 |      4.6 |     20 |      76% |   3.5 |    83  |  1.20 |      4.3
        10 |      5.0 |     19 |      77% |   3.9 |    81  |  1.23 |      4.8
        11 |      5.6 |     23 |      76% |   4.3 |    80  |  1.26 |      5.3
        12 |      6.1 |     24 |      76% |   4.7 |    78  |  1.28 |      6.0
        13 |      6.9 |     26 |      77% |   5.3 |    76  |  1.32 |      7.0
        14 |      7.5 |     27 |      77% |   5.8 |    74  |  1.35 |      7.8
        15 |      8.0 |     30 |      77% |   6.2 |    72  |  1.39 |      8.6
        16 |      8.5 |     30 |      78% |   6.7 |    70  |  1.42 |      9.6
        17 |      9.5 |     34 |      77% |   7.3 |    68  |  1.46 |     10.7
        18 |     10.1 |     33 |      78% |   8.0 |    67  |  1.50 |     11.9
        19 |     10.5 |     34 |      78% |   8.2 |    65  |  1.54 |     12.6
        20 |     11.6 |     38 |      78% |   9.1 |    63  |  1.59 |     14.4
        21 |     11.8 |     41 |      78% |   9.3 |    61  |  1.63 |     15.2
        22 |     12.3 |     40 |      78% |   9.6 |    59  |  1.69 |     16.3
        23 |     13.2 |     42 |      78% |  10.4 |    57  |  1.74 |     18.2
        24 |     13.7 |     43 |      79% |  10.8 |    56  |  1.80 |     19.3
        25 |     14.3 |     44 |      78% |  11.3 |    54  |  1.86 |     20.9
        26 |     14.9 |     49 |      79% |  11.8 |    52  |  1.93 |     22.6
        27 |     15.5 |     51 |      78% |  12.1 |    50  |  2.00 |     24.2


As you can see you should train UC as high as you can. In "pure melee monsters" crawl I would barely train any Dodging or Fighting (unless UC was really high like 20 or so) but as long as ranged monsters exist you should try to keep balance between offence and defense.

Here is my DsMo

  Code:

5322 | D:7      | Reached skill level 9 in Unarmed Combat
5715 | D:7      | Reached skill level 5 in Dodging
6004 | D:8      | Reached skill level 10 in Unarmed Combat
6544 | D:8      | Reached skill level 5 in Fighting
  7337 | D:9      | Reached skill level 11 in Unarmed Combat
 10318 | D:11     | Reached skill level 14 in Unarmed Combat
10965 | Volcano  | Reached skill level 1 in Evocations
 15788 | Lair:3   | Reached skill level 16 in Unarmed Combat
 18299 | Lair:5   | Reached skill level 1 in Armour
 19388 | Elf:1    | Reached skill level 10 in Fighting
 19457 | Elf:1    | Reached skill level 5 in Evocations
 20197 | Lair:6   | Reached skill level 10 in Dodging
 21686 | Lair:7   | Reached skill level 5 in Armour
23116 | Lair:8   | Reached skill level 17 in Unarmed Combat


Edit. Here is elliptic's OpMo from his current streak.

  Code:

9844 | D:8      | Reached skill level 12 in Unarmed Combat
10612 | D:9      | Reached skill level 6 in Fighting
 10612 | D:9      | Reached skill level 6 in Dodging
 12679 | Lair:2   | Reached skill level 14 in Unarmed Combat

[ajon]

Got it. Looks like it takes a bit more balancing. Is it worth it? Haven't actually had a run with UC make it very far yet to know.

[Sandman25]

This is easy to learn by experience. When you barely survive an encounter with Centaur who noticed you at the edge of LoS, you will quickly realize that you should have trained Dodging/Fighting instead of another level in UC. High levels are really expensive but unfortunately it is not displayed except in wizard mode.

[rchandra]

Your original assumption "train to mindelay immediately" is also false. Well you're not wrong in that it might be common, but it's not recommended as a general practise.

UC can do fine damage, even one-handed, but I hate it without forms since it's boring and exp-intensive.

[cerealjynx]

Ars far ars you warnt to.

[ajon]

Really? I've seen that advice given very often. What is a better method for training weapon skill?

[rchandra]

For the starting weapon it looks a lot more reasonable for some if you have a positive aptitude, I was thinking of people who go to 20-26 weapon skill with nothing else for the big twohander they find early / are gifted / hope to get. But even for starting weapons I'd recommend getting some fighting/defense before going all the way to mindelay.

edit: Here's my last Gladiator. Looks like I got a little weapon skill, then trained Fighting alongside, then found an armour on d:3 so probably split training 50% weapon 25 fighting/armour. Entered Lair and reached mindelay on the war axe (16) around the same time, while also having 9 Armour 10 Fighting. At that point I think I switched to mostly armour with 5 throwing to help javelin hydras.

[celem]

Much like rchandra I dont rush mindelay. Its the fastest way to increase your DPS output, true. But thats all it does. If you are an experienced crawler then you can probably keep your character alive while skilling it up, but many players will find they spend too long being very fragile this way, its not forgiving if you mess up positioning or make bad engagements (where you should have pulled them back or outright gone around). Pushing your weapon to mindelay is the most effective route, provided you are confident that you can reduce all engagements to a 1v1 tab-fest. The moment you do not control the battlefield in this manner then you are at risk. Picking up the misc combat skills adds forgiveness above anything else.

Infact rchandra's progression on the Gl he dumped is fairly illustrative of how I skill the vast majority of characters. Play it by ear, you cant predict what you will find or when, and therefore cant predict when you will need to start gathering the other combat-related skills. GDA drops early? Split with Armour immediately. Find a nice artifact launcher? Pick up some ranged...

When it comes to UC:
UC's mindelay is 27, it continues to speed up all the way. Now (hopefully) obviously you are not gonna focus it to 27 first thing. As a rule of thumb i'll focus UC as much as I feel I can afford until its around 12-15, after this point its usually still switched on but doesnt focus as I need to start spreading the points around. Remember however that there are other factors seriously influencing UC speed, if you wear body armour then it needs to be well trained as it will hamper your speed a lot, likewise shields (even a fully trained shield will add 1Aut to 50% of your UC strikes, forever). You do want to take it all the way eventually though, UC does not compare favourably with armed combat in the lategame without high levels. (caveat claws, as in below dump)

A UC char dump for comparison: TrMo^Qaz UC/Throwing
http://dobrazupa.org/morgue/celem/morgue-celem-20150114-102618.txt
Looks like UC/Fight/Dodge right from the start, Invocations early because Qaz. UC never really switches off, Fight/Dodge go off and on to stay some 5-7 levels behind UC. Invocations never really goes off either on this guy, because Im headed to 27. Throwing isnt picked up until 50k turns in (presumably once I started getting decent large rock stacks). In the end he never hits 27UC as he ascends, I think I turned it off for a bit towards the end to speed up the acquisition of shield skill and some charms. I was able to tolerate a slightly lower UC score probably because claws, even with a shield in the off-hand trolls wreck.

[Bloax]

So its common to train a weapon to min delay first thing then start branching out to other skills.

It's actually best to keep fighting 2-3 levels behind the weapon skill even at that point, or even equal to weapon skill if the fighting apt is +2 or +3 and 3-5 levels behind if the apt is much lower than the weapon skill. (As is the case with DE + Staves)

Past mindelay it's still a good idea to train weapon skill, but only if your fighting skill is somewhat higher than your weapon skill - which is to say when getting another level of weapon skill will be much cheaper in experience costs than another level of fighting.

Sandman has given decent guidelines on how much to train unarmed, which is basically "keep it high among your other skills".
An important thing to notice is that pumping 100% of the experience into unarmed is actually the best thing to do until you're up to about 9 unarmed, at which point giving a little bit into fighting will also be good due to its combined damage/hp bonuses.

[Sandman25]

Past mindelay it's still a good idea to train weapon skill, but only if your fighting skill is somewhat higher than your weapon skill - which is to say when getting another level of weapon skill will be much cheaper in experience costs than another level of fighting.

Are you talking about Be? If any other characters it is better to put XP into some spells (even in CPA). Especially true for demon weapons and qblades.

  Code:

Attack: Human Fire Elementalist vs. Geryon (4000 rounds) (2014/11/15/ 7:35:49)
Human Fire Elementalist: XL 20   Str 13   Int 21   Dex 13
Wielding: +9 demon whip of freezing, Skill: Maces & Flails
Geryon: HD 15   AC 15   EV 6

Maces & Fl | AvHitDam | MaxDam | Accuracy | AvDam | AvTime | AvSpeed | AvEffDam
         0 |      5.9 |     29 |      82% |   4.9 |   110  |  0.91 |      4.4
         1 |      5.9 |     30 |      81% |   4.8 |   105  |  0.95 |      4.6
         2 |      5.8 |     30 |      81% |   4.8 |   100  |  1.00 |      4.8
         3 |      5.8 |     33 |      81% |   4.7 |    95  |  1.05 |      5.0
         4 |      6.3 |     32 |      82% |   5.1 |    90  |  1.11 |      5.7
         5 |      6.0 |     32 |      82% |   5.0 |    85  |  1.18 |      5.9
         6 |      6.2 |     35 |      82% |   5.1 |    80  |  1.25 |      6.4
         7 |      6.5 |     35 |      82% |   5.3 |    75  |  1.33 |      7.1
         8 |      6.5 |     34 |      82% |   5.3 |    70  |  1.43 |      7.6
         9 |      6.9 |     33 |      84% |   5.8 |    65  |  1.54 |      9.0
        10 |      6.8 |     35 |      83% |   5.7 |    60  |  1.67 |      9.5
        11 |      6.9 |     33 |      83% |   5.7 |    55  |  1.82 |     10.4
        12 |      7.2 |     39 |      83% |   6.0 |    50  |  2.00 |     12.0
        13 |      7.1 |     36 |      83% |   6.0 |    50  |  2.00 |     11.9
        14 |      7.5 |     38 |      84% |   6.3 |    50  |  2.00 |     12.7
        15 |      7.5 |     38 |      83% |   6.3 |    50  |  2.00 |     12.6
        16 |      7.7 |     38 |      84% |   6.5 |    50  |  2.00 |     13.1
        17 |      7.8 |     39 |      84% |   6.5 |    50  |  2.00 |     13.1
        18 |      7.9 |     38 |      84% |   6.7 |    50  |  2.00 |     13.4
        19 |      8.2 |     45 |      84% |   6.9 |    50  |  2.00 |     13.9
        20 |      8.1 |     39 |      85% |   6.9 |    50  |  2.00 |     13.8
        21 |      8.4 |     41 |      83% |   7.1 |    50  |  2.00 |     14.1
        22 |      8.5 |     43 |      85% |   7.2 |    50  |  2.00 |     14.5
        23 |      8.7 |     40 |      84% |   7.3 |    50  |  2.00 |     14.7
        24 |      8.8 |     45 |      85% |   7.6 |    50  |  2.00 |     15.1
        25 |      8.8 |     42 |      84% |   7.5 |    50  |  2.00 |     15.0
        26 |      9.0 |     41 |      85% |   7.7 |    50  |  2.00 |     15.4
        27 |      9.2 |     44 |      85% |   7.9 |    50  |  2.00 |     15.7


[dornbeast]

My current method is to aim for 10 in the weapon, followed by my primary defense and Fighting. Then I look for mindelay.

I'm still trying to work out how casters go, although "10% fail unless yellow" seems to be a decent target.

On the other hand, I've ascended once, and died countless times, so what do I know?

[Hurkyl]

The closer you get to mindelay, the more valuable another point of skill is. Especially so with fast weapons. If you have 10 skill in short blades and a short sword is your primary means of killing things, raising that to 12 skill will do a lot for you. If you have 10 skill in crossbows and an arbalest is your primary means of killing things, raising to 12 skill doesn't do nearly so much.

If you've skilled reasonably and you're getting close to mindelay (and are using a weapon as a main way to kill dangerous things), it's probably worth shutting everything else off to get there faster. But if you're not so close, there's much less hurry.

[daggaz]

The closer you get to mindelay, the more valuable another point of skill is.

Considering that attack delay goes down more or less linearly with weapon skill, while the the cost of weapon skill goes up more or less exponentially, I have to say I have some pretty serious doubts about your analysis, here.

You get the most bang for your buck with the first points in weapon skill, and less and less as you move on. Thats why it is optimal to rush some points into weapon skill first hand, and then back off a bit when it starts costing a lot more, putting those points instead into cheaply getting your defenses/fighting up. Unless you are wielding a dagger on a kobold or something similar, it's almost never a great idea to rush to min-delay, even if you are close.. The only reason fast weapons seem to outshine the slower weapons in your analysis, is because they require such a lower position on the exponential curve to hit min-delay in the first place. Getting that last point in longblades so you can swing your claymore just a tiny bit faster is certainly not as good an idea as putting say 10 points into dodging and adding several points of EV to your defenses.

[dowan]

But he said another point of skill, not another point of xp invested in the skill. And it's true, going from 1.1 to 1.0 delay (about 10% more damage) isn't as big as going from 0.8 to 0.7 delay(about 12% more damage)

My math might not be exactly right here, but the concept I'm describing is true, losing 0.1 delay is more and more valuable the lower your delay already is, because 0.1 is a larger chunk of it.

In the most extreme example, that doesn't actually happen in crawl, if you went from 0.2 to 0.1 delay, you'd be doubling your damage.

[njvack]

Well, he also said it gets important to rush weapon training as you approach mindelay, which is not necessarily (or, I think, usually) the case. Training defense/magic/invo/evo/fighting/stealth may be more appropriate depending on where your character is.

Getting that last point of Axes for your exe axe, or long blades for that triple sword, is really really expensive. Maybe you could get Haste online instead.

[daggaz]

The only skill I think of in this context is the skill you can put points into: [weapon type]. You dont put points into "min-delay". I dont even know why I am responding to this argument, it is beyond the pale of obtusely arguing semantics for the win. I would also add once again, that 'points' in min-delay scale linearly so that last point is not worth any more than the first one, although it does cost a lot more.

[Zooty]

Sure, weapon speed depends linearly on skill. But you don't care about speed, you care about damage, which depends inversely on speed (+whatever damage you get from higher skill). You probably get more damage per Xp at low skill levels, but you definetly get more damage per skill level at high levels.

[Pollen_Golem]

damage, which depends inversely on speed DELAY

You probably get more damage per Xp INVESTMENT at low skill levels

Huhh, whaa...

Edit: OK fixed.

[duvessa]

by speed, Zooty meant delay

[Hurkyl]

Also, the cost to gain the next skill level increases more or less linearly, not exponentially.

(the experience to skill point ratio decreases over time, but that affects all skill levels equally)

[Hurkyl]

Well, he also said it gets important to rush weapon training as you approach mindelay, which is not necessarily (or, I think, usually) the case. Training defense/magic/invo/evo/fighting/stealth may be more appropriate depending on where your character is.

Getting that last point of Axes for your exe axe, or long blades for that triple sword, is really really expensive. Maybe you could get Haste online instead.

The second-to-last point was really expensive too. And so was the third-to-last point. If getting haste is better than getting the last point of skill, then it was almost surely a better than the second-to-last point, or the third-to-last point.

I ran the numbers to confirm what I had guesstimated. Going from 22 -> 24 skill in axes increases your damage* by 12.5% (0.9/0.8 = 1.125), at a cost of 4350 skill points. Going from 22 -> 26 skill in axes increases your damage by 28.6% (0.9/0.7 ~ 1.286); at the same rate as above this would be worth around 9950 skill points (4350 * 286 / 125) -- but it only actually costs you 9300 skill points.

So, yes, at least in this situation, you are getting a better return on your investment as you continue to train the skill, even after you account for the increasing cost of skill levels.

Okay, sure, sometimes you can be in a weird place where something new comes up (I just found a book with haste!), or for some reason the strategic value of doing more damage is decreasing relative to other concerns, or maybe your previous choice to train your weapon skill was a mistake and you've only just now realized it.

But usually you're not in a weird place, and so when you get close enough to mindelay, there is a sort of snowball effect where the more you train the skill, the more worthwhile it is to continue training the skill.

---

I can't decide if the worth of more damage should be measured additively or multiplicatively; if the latter, then going from 22 -> 26 should be worth about 9280 (log(1.286) / log(1.125) * 4350) skill points, so the worth of training the skill remains about constant.

*: I'm only counting rate of fire. I'm not counting accuracy or the multiplier to base damage.

[dowan]

So, if you do factor in the increased accuracy and damage, it's even more worth it, by some amount.

But, as you said, all this only makes sense if the best way to improve your character at the moment is to increase your damage output, and if you don't have access to some other, cheaper way to increase your damage output.

But essentially, if it was worthwhile to increase your weapon skill from 0 - 1, it's worth it even more to increase it from 1 - 2, and this pattern remains true until hitting min-delay.

Spoiler: show

All these conversations make me realize, nobody is trying that hard to figure out how to play crawl optimally. People are trying harder to refute common advice, even when that common advice is actually correct. So really, the meta game of crawl takes place on this forum, crawl.exe is just a minigame in which you can sometimes prove what you were saying on the forum. Of course, logical proof doesn't beat popular consensus, so really, playing crawl.exe is quite suboptimal if you're trying to actually win the real game. Optimal play calls for snarky one liners, and avoiding making any factual assertions.

[and into]

So by this logic, as soon as one starts to train UC exclusively, one should train UC to 27? Yeah good luck with that. if you actually try that out on crawl.exe you will see why it generally isn't a good idea.

I don't see why it is a shame that people are concerned with trying to give helpful advice to new players rather than figuring out a semi-mythic "optimal" skill investment strategy.

[dowan]

No, not at all. Just realize, the more weapon skill (before min-delay) you already have, the more return you're getting on additional investment.

Being able to kill stuff in less auts is not always (or even usually) the best way to improve your character. UC 27 won't help you survive a centaur at the edge of your LOS. So exclusively pumping UC training instead of balancing it against fighting, armor, dodging, etc is a bad idea, but not because the UC skill itself is any less efficient.

If you make a logical and correct assertion, but back it with incorrect facts, when someone finds out the facts are incorrect they may assume your logical assertion was in fact incorrect. The logical and correct assertion is that it's a bad idea to just pump a weapon skill to min-delay, ignoring all other skills. The incorrect fact is that it is less and less efficient to put experience into that weapon skill.

It's a great thing that this community tries to give good advice, and generally it succeeds at this. The problem is when the community is repeating good advice it heard from someone who understood it, and try to apply it incorrectly, because they don't understand why it's good advice. Even worse is when the reason given for it being good advice is wrong, because a logically minded person is likely to dismiss very good advice just because the people giving it explained it wrong.

It's like saying that breaking up packs is a good idea (true) because pack monsters give each other a passive to-hit and damage bonus (not true). If someone code dives and sees there's no such pack bonus, they might assume the advice itself is wrong.

[duvessa]

No, not at all. Just realize, the more weapon skill (before min-delay) you already have, the more return you're getting on additional investment.

This is generally wrong, skill cost tends to increase faster than the speed increase increases.

[tasonir]

Also it seems like everyone is focusing on the increase in damage because of reduced attack delay, which sounds like they're giving weapon advice. Dirty, dirty weapon advice. UC scales differently, because it receives meaningful damage increases from skill in addition to the damage delay. The most important skill level to gain is going from 0 to 1, which results in a 33% damage increase (without counting delay increase), much more than 26->27. The fastest scaling period of UC is the first 5-10 levels; this is why transmuters are the weakest background, but monks are fairly decent. Those first few levels are brutal. You want UC to be around 8 before it starts to be reasonable. Around then you start getting only 10% more per skill level, so the gains are slowing down. Attack delay gains increase, but that's pretty gradual and it mostly gets covered up by the slower (percentage) damage gains.

You can see this in sandman's fsim chart. From 0 to 5 you go from .4 to 2.1, a 520% increase. From 22 to 27 you get about a 50% increase. Starting from 0 skill is somewhat contrived, you probably started with 2 for transmuter: 0.9 to 3.1 (skill 7) is a 344% increase. If you're a monk going from 4 to 9, 1.6 to 4.3 is a 268% increase. Attack delay is not the be all and end all of UC. Early ranks are worth more than later ranks, the opposite of weaponry.

Claws giving a flat bonus will flatten the curve early on somewhat, of course, but early levels are still more important than later ones, it just isn't as extreme.

[duvessa]

weapon skill also increases damage per hit...

[dowan]

No, not at all. Just realize, the more weapon skill (before min-delay) you already have, the more return you're getting on additional investment.

This is generally wrong, skill cost tends to increase faster than the speed increase increases.

I ran the numbers to confirm what I had guesstimated. Going from 22 -> 24 skill in axes increases your damage* by 12.5% (0.9/0.8 = 1.125), at a cost of 4350 skill points. Going from 22 -> 26 skill in axes increases your damage by 28.6% (0.9/0.7 ~ 1.286); at the same rate as above this would be worth around 9950 skill points (4350 * 286 / 125) -- but it only actually costs you 9300 skill points.

So, yes, at least in this situation, you are getting a better return on your investment as you continue to train the skill, even after you account for the increasing cost of skill levels.

Is this example incorrect for some reason?

[and into]

I ran the numbers to confirm what I had guesstimated. Going from 22 -> 24 skill in axes increases your damage* by 12.5% (0.9/0.8 = 1.125), at a cost of 4350 skill points. Going from 22 -> 26 skill in axes increases your damage by 28.6% (0.9/0.7 ~ 1.286); at the same rate as above this would be worth around 9950 skill points (4350 * 286 / 125) -- but it only actually costs you 9300 skill points.

So, yes, at least in this situation, you are getting a better return on your investment as you continue to train the skill, even after you account for the increasing cost of skill levels.

Is this example incorrect for some reason?

I don't know what duvessa is referring to, specifically, but I will say that Hurkyl's point is clarifying in certain respects, but potentially misleading in others. njvack's initial point is about opportunity costs for training a weapon to a very high level, and the opportunity costs do in fact change as the total amount of experience being invested in a skill increases.

As I understand it, Hurkyl's point is that, until you hit min-delay, the "skill points invested vs. increase in damage" ratio tends to improve slightly with greater investment, even though the total cost of skill points increases as the weapon skill gets to higher levels. This may generally be true. However, there is an important difference between investing 4350 skill points for a 12.5% increase in damage, vs. 9300 skill points for a 28.6% increase in damage, because in the latter case there is much more you could "buy" with those skill points, and other skills do not necessarily have a linear-ish relationship between amount invested and your return. In fact, in many practical situations, many skills will in this respect prove quite different from pre-min-delay weapon skill.

To take one clear example, getting an important god ability (e.g., Finesse) reliably useful can be extremely valuable for a particular character, whereas getting it to 80% failure is almost worthless; in such cases you basically have a situation in which you want to devote at least X amount of experience into invocations, or not bother investing anything.

So, just because it makes sense to devote 4350 skill points into a weapon skill doesn't mean it will make sense for the same character to devote 9300 skill points into the weapon skill, even if the return in damage for the investment is slightly better. (And this is ignoring new loot, nota bene.)

All of this is to say that the actual judgments about how much to invest in weapon skill, and when, are non-trivial, and are not exhausted simply by virtue of the fact that "skill investment:damage increase" remains roughly constant, or slightly improves, up until you hit min-delay. (I don't think Hurkyl was necessarily saying otherwise, but I don't want the numbers Hurkyl shared to be misinterpreted.)

[tasonir]

weapon skill also increases damage per hit...

Fighting also increases damage per hit.

What is your point?

[Hurkyl]

As I understand it, Hurkyl's point is that, until you hit min-delay, the "skill points invested vs. increase in damage" ratio tends to improve slightly with greater investment, even though the total cost of skill points increases as the weapon skill gets to higher levels.

Just for the record, I only claim this as you near min-delay, and I hold some skepticism regarding crossbows since they bottom out at a rather slow speed.

This may generally be true. However, there is an important difference between investing 4350 skill points for a 12.5% increase in damage, vs. 9300 skill points for a 28.6% increase in damage, because in the latter case there is much more you could "buy" with those skill points, and other skills do not necessarily have a linear-ish relationship between amount invested and your return. In fact, in many practical situations, many skills will in this respect prove quite different from pre-min-delay weapon skill.

And my claim is that, barring skilling mistakes, this only happens under unusual circumstances; e.g. a new opportunity opened up, or for some weird reason you judged that getting that there was some sort of damage plateau where doing enough damage to get over the hump is rather important, but being able to do even more damage isn't. I don't think this happens in crawl -- certainly not often -- except maybe if you're doing massive damage and wasting a lot of damage with overkilling.

To take one clear example, getting an important god ability (e.g., Finesse) reliably useful can be extremely valuable for a particular character, whereas getting it to 80% failure is almost worthless; in such cases you basically have a situation in which you want to devote at least X amount of experience into invocations, or not bother investing anything.

So, just because it makes sense to devote 4350 skill points into a weapon skill doesn't mean it will make sense for the same character to devote 9300 skill points into the weapon skill, even if the return in damage for the investment is slightly better. (And this is ignoring new loot, nota bene.)

Allow me to phrase how this scenario looks differently:

• You think it's a good trade to spend 4350 skill points to get a 12.5% boost in damage rather than get Finesse which would boost your damage by 50% in an emergency

and once you've done that,

• You think it's a bad trade to spend 4950 skill points to get a 14.3% boost in damage, rather than get Finesse which would boost your damage by 100% in an emergency
• You think it's a bad trade to spend 4350 skill points to get a 10.3% boost in damage*, rather than get Finesse which would boost your damage by 100% in an emergency (by only getting 75% the way from 25 -> 26)

Okay, that could happen, but that's not a very wide margin for things to line up right; it's far more likely that the first trade was bad, or that either of the second set of trades are good. And if you are in that margin, it's narrow enough that the right choice is only a tiny bit better than the wrong choice.

The only realistic scenario I can come up with where this makes a difference is in the drastic change category -- e.g. you are preparing to move to a new branch (or to get a rune) and you think you need Finesse for it, but you don't think Finesse will be very useful until that point, so you put it off as long as you can.

*: Just in case someone thinks this is in contradiction with my previous claims, note that the remaining 600 skill points to finish up the level afterwards are a very good deal.

[duvessa]

No, not at all. Just realize, the more weapon skill (before min-delay) you already have, the more return you're getting on additional investment.

This is generally wrong, skill cost tends to increase faster than the speed increase increases.

I ran the numbers to confirm what I had guesstimated. Going from 22 -> 24 skill in axes increases your damage* by 12.5% (0.9/0.8 = 1.125), at a cost of 4350 skill points. Going from 22 -> 26 skill in axes increases your damage by 28.6% (0.9/0.7 ~ 1.286); at the same rate as above this would be worth around 9950 skill points (4350 * 286 / 125) -- but it only actually costs you 9300 skill points.

So, yes, at least in this situation, you are getting a better return on your investment as you continue to train the skill, even after you account for the increasing cost of skill levels.

Is this example incorrect for some reason?

Sorry, I should have been more specific. This example is probably correct, because it deals with really high skill levels really close to min delay. But without really high skill levels close to min delay (and therefore a very rare weapon) it's not true very often. For example, skill 12 to skill 16 costs slightly more skill points than skill 0 to skill 12, and even on a great sword will increase damage by less than 50%, whereas 0 to 12 will likely increase damage by more than 100%. 14 to 15.65 costs the same skill points as 12 to 14, and also doesn't increase great sword damage by as much as 12 to 14, without tons of slaying or slaying-like effects.

[ajon]

Wouldn't finesse boost your damage by 100%? If it is doubling, and you get two attacks in the time you normally get one, then you will do 100% more damage. Your normal damage is only 50% of the damage you do with finesse.

[Hurkyl]

Wouldn't finesse boost your damage by 100%? If it is doubling, and you get two attacks in the time you normally get one, then you will do 100% more damage. Your normal damage is only 50% of the damage you do with finesse.

Yes, I meant 100%: I forgot to flip the number around the right way.

[Hurkyl]

This example is probably correct, because it deals with really high skill levels really close to min delay. But without really high skill levels

I hadn't realized that the rate at which the cost of a skill level grows decreases as you get to very high skill levels.

E.g. 11 -> 12 is 15.4% more expensive than 10 -> 11, but 23 -> 24 is merely 7.1% more expensive than 22 -> 23.

(note: these are the only ones I checked; it's possible that these particular examples are outliers)