Welcome

Please register for Total War Access to use the forums. If you're an existing user, your forum details will be merged with Total War Access if you register with the same email or username. For more information please read our FAQ’s here.

Categories

Modelling ranged units power - for fun!

Disposable HeroDisposable Hero Posts: 3,771Registered Users
edited June 8 in Balancing Discussions
There was a message in the locked thread that I had liked to reply to, it was about if you could make a model to show the difference between artillery and missile infantry in terms of survivability and ranged power etc.

First of all, a DISCLAIMER: This is for fun. I had a boring flight and spent two hours just playing around with the stats of most ranged infantry and the cannon-type artillery and made some very simple models. I am not in any way saying these are accurate, this is for fun but I wanted to share it anyways because in a way they are pretty illustrative. If you are only going to go on and on and on about this being inaccurate and biased, please stop reading now. If you are a bit nerdy and think these kinds of stuff is just fun to meddle with, stay tuned. :smile:

So what I did was to mash up the defensive parameters into one meta-score and the offense parameters into another meta-score. So defense is: MD, armor, speed, loose formation flag, stalk flag, parthian flag, HP, resists, shield. The resist style parametesr are used to calculate effective HP (but scaled down because they don't work on all damage types). Likewise, the parameters for range, DPS (AP scaled higher than reg damage), accuracy, stalk flag, parthian flag. I also scaled for cost.

Firstly I took the defense meta-score and plotted vs range, just to visualize the original question.


The interesting thing here, as expected, is that Jezzails and Bowshapti score unnaturally high defensive values for their range, while Deck gunners are even over-compensated having very artillery-like defensive stats but with only 245 meters range.

I then modelled the offense meta-score vs range. Here I tried to just optimize the cost-scaling to be reasonable, making the assumtion that artillery and archers should both be roughly equally cost effective. Not perfect, but it's something to steer the cost scaling a bit.



Now, the grand finale is of course then to merge the meta-scores (without the cost-scaling) and plot vs cost. Which units are undercosted and which are overcosted?



I have to say, being a really crude model and all, it was quite interesting anyways to see the usual suspects stand out. Rangers, Bugmans, Jezzails, Bowshapti score well above their value. Waywatchers are up there, but not as much as they probably should be in the real game. The model doesn't for example account for that the 360 fire on several units work as extra defense. The Helblaster was a surprise, it has excellent offensive stats, but it's hard to use in real games. I actually makes me want to try it more just to see. :) Otherwise all traditional cannons and BTs are probably underestimated a fair bit. This is more likely because of the calibration scaling. Especially cannons have really poor cal area and the model doesn't take the accuracy stat into account as I don't really know how it works. Cannons use these stats though while most missile infantry don't, except waywatchers iirc. So maybe arty and ww should be bumped up a little because of that. Otherwise WLC, carronade and Organ gun stands out as under performers, together with the BTs.

Now, having done that, I of course had to nerf Jezzials just to visualize the predicted change. I gave them -55 meters range, -30 armor and +0.5 cal area, and it looks like this:


Now, please note that I am not saying this proves anything what so ever. It's an extremely crude model and has tons of flaws. Especially alarming is that the cost-scaling is obviously completely wrong in the merged plot, you can't use the x-axis. Anyways, this is just sharing the result of a boring flight.

Make me proud, don't let me wake up to a locked thread now. Have fun, good night!

Edit:

After a few iterations and tweaking all scaling parameters on a training set that does not include the monstrous infantry I end up with a plot like this:
Post edited by Disposable Hero on
«134

Comments

  • GodOfGobbozGodOfGobboz Posts: 144Registered Users
    Hellblaster too strong. It should receive -1 or even -2 melee attack , i reckon. ;)

    (No but really ,nice work.)
  • eumaieseumaies Senior Member Posts: 4,485Registered Users
    Awesome charts :)

    Would be a fun template for analyzing some more specific hypotheses at some point (eg more targeted dimensions/stats where you’re more confident of comparability)
  • ReymReym Posts: 485Registered Users
    Nice work indeed and as said eumeis such thing could get used for debating about various hypothesis and stuff.

    But is talking about what is appropriate to talk about in this thread appropriate to be talked about in this thread ?
  • OrkLadsOrkLads Posts: 1,444Registered Users



    First of all, a DISCLAIMER: This is for fun. I had a boring flight and spent two hours just playing around with the stats of most ranged infantry and the cannon-type artillery and made some very simple models. I am not in any way saying these are accurate, this is for fun but I wanted to share it anyways because in a way they are pretty illustrative.

    Now, having done that, I of course had to nerf Jezzials just to visualize the predicted change.

    Start of speech: This is just for fun. Trust.

    End of Speech: So in conclusion, nerf Jezzails.

    Who didn't see that coming from the 1st line lol

  • LamentationsLamentations Posts: 118Registered Users
    If I’m interpreting this correctly, all of the cannons are bad.
  • keroro7keroro7 Posts: 220Registered Users
    this chart is awesome. it is mostly proper for units power that felt when playing games.
    so if u don't mind I have a request. i want to know the coordinates of the corsair (hand-bow) / shade, skink, skirmisher/chameleon. It is enough to just explain it in words, and thanks for ur hard works.
  • PippingtonPippington Posts: 2,026Registered Users

    If I’m interpreting this correctly, all of the cannons are bad.

    I suspect that's because the 'meta score' doesn't include penetration or splash damage (unless DisposableHero just forgot to mention those). It's treating cannons as if they can kill max 1 target per shot.

    @Disposable Hero I really like the plots but the meta scores appear to be a bit of a black box - you tell us what the inputs are but not how they're processed to come up with the final number. Without seeing the formula the plots seem hard to interpret, and I can't help but wonder how useful such a composite score can be when you're combining so many variables. Surely it's unavoidable to bake in a lot of assumptions about which stats and attributes are most valuable, which might vary from player to player (or even from unit to unit). I'd love to see similar plots for simpler numbers that address more limited, concrete questions.

    Also some random nitpicky questions:
    • What's the goodness of fit of your straight line?
    • If you fit missile infantry and artillery as two separate populations, how different are the gradients of those fits to your all-in-one fit?
    • If you change the weighting of a parameter in your meta-score, how much does the distribution change? e.g. what's the change in the position of waywatchers if you assign a 50% greater weight to parthian shot?


    Get on, Kroq-Gar, we're going shopping

  • TlaxtlanSoothsayerTlaxtlanSoothsayer Posts: 2,321Registered Users
    edited June 8
    Thanks for creating this thread, it's a very interesting approach to compare different units for sure.


    If I understand this correctly then Disposable Hero's model has
    1. An offense score
    2. Defensive score
    3. Mixed score (Defensive + offensive capabilities) vs unit cost



    I don't even know where to begin with.

    The issue is that the model outright ignores some of the most important values:
    Values such as penetration, accuracy, real damage per second, real reload time, damage from flank shots. Muzzle velocity and the resulting damage difference against fast moving or slow targets. Some projectiles get blocked by shields, others don't.
    • For example an imperial Great Cannon or Organ Guns are going to kill several enemies with 1 projectile.
    • Rangers or Sisters of Avelorn get blocked by shields.
    • The projectiles of certain guns are more than 3 times as fast as most arrows.
    • Disposable Hero's model ignores accuracy




    It also explains why the Helblaster Volleygun ends up as one of the best units in this chart. According to this model it has the best offensive capability, because it has a really high damage output (in theory).
    • 1800 (600 x 3) damage per 10 seconds according to the unit card.
    • Almost 3 times as much as a Great Cannon
    • When you test it yourself you will notice that this is not true at all, in practice it looks totally different.

    It's no surprise. In reality the Helblaster Vollery Gun isn't nearly as useful. Values such as accuracy, penetration and shots per volley play an important role as mentioned earlier on. Reload speed is another issue: Outiders have 5.4 reload speed .... in theory. But it takes them ages to reload due to the animations. Most reload speeds are an outright lie, therefore the damage per second stated by CA is wrong as well.


    Just as a side note: Did you add Ratling Guns to this model? They have a hypothetical damage output of 8136 per unit and decent survivability, speed and so on. (339 Dp10S x 24 on large unit size.)


    Conclusions:
    It's a trap to draw overhasty conclusions, because this model does not take into account some of the most important values. For example saying that the Helblaster Volley Gun is more efficient than a Great Cannon is false, unless we are talking about very specific circumstances (the graph suggests that the Helblaster is way more cost efficient, but this isn't the case when you take all possible scenarios into account).


    My apologies, I don't want to sound too harsh. I'm aware of that I mainly focused on flaws and criticism of this model. But I really like the idea of number crunching. And I have to say that this post is way more interesting than a lot of other feedback. I bet it's going to trigger a lot of people, simply because it mentioned wArPlOcK jEzZaIls. It's a very basic model with several severe issues for various reasons nonetheless.


    Well, my criticism is also a little bit inappropriate, because Disposable Hero stated that these charts are mainly for fun, they were created within a few hours during a flight and I have to say that they are entertaining and interesting at the same time.


    Keep it up. I think something like that would be really interesting for melee infantry. The calculation becomes easier, because there are less variables involved. :smile:
  • ystyst Posts: 6,152Registered Users
    edited June 8
    Those r pretty pointless. Wheres the most relevant chart. Dmg and range?

    But of coz, we all know whats the point of this thread lol
    https://imgur.com/a/Cj4b9
    Top #3 Leaderboard on Warhammer Totalwar.
    Unit stats compare courtesy of Seal62 https://total-war-unit-compare.herokuapp.com/
  • CanuoveaCanuovea Posts: 13,348Registered Users, Moderators
    Right, thread locked!

    Erm, oh, no? It seems fine? Oh, okay, thread... uh... not locked.

    Keep it that way please.
    -Forum Terms and Conditions: https://forums.totalwar.com/discussion/172193/forum-terms-and-conditions#latest
    -Using all caps is the equivalent of shouting. Please don't.
    -The "Spam" flag is not a "disagree" flag. Have a care.
    -...No, no the "Abuse" flag isn't a "disagree" flag either!
    -5.7 Summon a moderator if someone seems to be out of line, or use the report button. Do NOT become another party to misbehaviour
  • Loupi_Loupi_ Posts: 1,323Registered Users
    Interesting modelling sir! I think the reason why the waywatchers are so close to the best fit line is because on their own they are totally fine. In a real battle there will be 3 or 4 of them, which will raise their defence and offense meta score, since they protect each other and therefore kill more.

    It would interesting to see how their position changes if you reduced their MD and HP to the level of shadowwalkers for example.


    Also where do skirmish units like deepwood scouts, shadow warrior, shades, chameleon skinks and shadow walkers fall?
  • Disposable HeroDisposable Hero Posts: 3,771Registered Users

    If I’m interpreting this correctly, all of the cannons are bad.

    Firstly one should be very careful about interpreting anything from here, but ion the case of artillery they are systematically underrated in the model. I think one big reason is because I use the calibration area at Max range as accuracy parameter and that is a pretty bad approximation. Iirc cannons and bt have accuracy and marksmanship in the 40ies while all missile infantry except ww have 20. I don't know how this influence accuracy but it sure does, for example when putting chavs on artillery you boost these stats but cal area stays the same, and the cannon become more accurate (see hammer of the witches).

    I don't think it has to do with penetration, I assume no wasted damage due to overkill so the target here might be a very big sem. Accuracy is the hardest to model here, partly because it's exactly behavior is unknown and partly because it also depends on what you're shooting at. I can easily add the total accuracy parameter though. In general I tried to be extra forgiving with cal area since otherwise all cannons, and even archers, get very weak and all powder get very strong. So we assume a rather big target here.
  • Ephraim_DaltonEphraim_Dalton Senior Member Posts: 21,095Registered Users
    All that data? Misses something very important, the rest of that faction's roster!

    Again, Skaven are missing flying units, cavalry and warhounds. Their means of influencing the flow of the battle outside of their ranged units is limited. Coincidentally, those three unit types are also the ones that cannot be thrown around by high-mass units, showing just how hamstrung Skaven actually are with their limited answers to that. Skaven are a faction whose killing power is almost completely condensed in its ranged units, but its artillery is fragile and their other damage dealers, except for the Jezzails have very low range. Pre-DLC those weaknesses pushed them to the bottom of the tier list simply because their strengths were few but their weaknesses many so unless you were a very good player and your opponent made exploitable mistakes you had a very hard time winning.

    300 range and the survavibility on the Jezzails are needed. If you want to nerf them, something else has to be buffed to compensate unless it's your goal to just nerf Skaven to trash tier status again.
  • Disposable HeroDisposable Hero Posts: 3,771Registered Users
    keroro7 said:

    this chart is awesome. it is mostly proper for units power that felt when playing games.
    so if u don't mind I have a request. i want to know the coordinates of the corsair (hand-bow) / shade, skink, skirmisher/chameleon. It is enough to just explain it in words, and thanks for ur hard works.

    Thanks! I did not include every unit, I tried to first stick to units without for example poison or other "extras" that makes it harder to fit. As vanilla as possible. Corsairs I could definitely add though and shades too. The skinks I left out for now so that I didnt have to bother with poison initially.
  • Disposable HeroDisposable Hero Posts: 3,771Registered Users

    If I’m interpreting this correctly, all of the cannons are bad.

    I suspect that's because the 'meta score' doesn't include penetration or splash damage (unless DisposableHero just forgot to mention those). It's treating cannons as if they can kill max 1 target per shot.

    @Disposable Hero I really like the plots but the meta scores appear to be a bit of a black box - you tell us what the inputs are but not how they're processed to come up with the final number. Without seeing the formula the plots seem hard to interpret, and I can't help but wonder how useful such a composite score can be when you're combining so many variables. Surely it's unavoidable to bake in a lot of assumptions about which stats and attributes are most valuable, which might vary from player to player (or even from unit to unit). I'd love to see similar plots for simpler numbers that address more limited, concrete questions.

    Also some random nitpicky questions:
    • What's the goodness of fit of your straight line?
    • If you fit missile infantry and artillery as two separate populations, how different are the gradients of those fits to your all-in-one fit?
    • If you change the weighting of a parameter in your meta-score, how much does the distribution change? e.g. what's the change in the position of waywatchers if you assign a 50% greater weight to parthian shot?
    Yeah, that's why I try to emphasize that this was something I mostly did for my own entertainment, not to prove anything. The formula are a mess more or less for anyone looking from the outside, but I can try to elaborate a little.

    The key here though is that all input is the in game values, so there is no bias there. The model itself is arbitrary in the way that all scaling values and the way all these parameters are merged are trial-and-error, based on some general intuition how this should be working. However, that said, the important thing that makes it useful for something is that all value are then treated exactly the same by the model. So the effect of shield block chance will be weighted the same for all units with shields. All units with PR, or stalk, or parthian will be weighted the same. So if there is a general trend here and some units are outliers, then it does mean something. It can be hard to pin down exactly what creates the offset, in order to get that you need to change parameters back and forth and look at the live updates in the graphs to see the effect on all units simultaneously.

    To your questions:
    - r2 is 0.4, room for improvement but there's also a limit to how good we could expect it to be. Maybe 0.6-0.7 is achievable.
    - I didn't do that but I could check later
    - hard to describe, things like parthian are just added as a blanket increase for now, but i should integrate it into the effective dps value for sure. A strong unit with stalk should benefit more than a weak unit with stalk. Range, accuracy and ap-ratio are directly scaled into the effective dps, I just didn't get that far with the other "extras" yet. Could be fixed.

    The first thing I will look at though is the range scaling and the cost scaling and accuracy skill implementation. These should bump up artillery a fair amount, and also jezzails so they will look stronger because of this.
  • Disposable HeroDisposable Hero Posts: 3,771Registered Users

    Thanks for creating this thread, it's a very interesting approach to compare different units for sure.


    If I understand this correctly then Disposable Hero's model has

    1. An offense score
    2. Defensive score
    3. Mixed score (Defensive + offensive capabilities) vs unit cost



    I don't even know where to begin with.

    The issue is that the model outright ignores some of the most important values:
    Values such as penetration, accuracy, real damage per second, real reload time, damage from flank shots. Muzzle velocity and the resulting damage difference against fast moving or slow targets. Some projectiles get blocked by shields, others don't.
    • For example an imperial Great Cannon or Organ Guns are going to kill several enemies with 1 projectile.
    • Rangers or Sisters of Avelorn get blocked by shields.
    • The projectiles of certain guns are more than 3 times as fast as most arrows.
    • Disposable Hero's model ignores accuracy
    Not quite for everything, let's see. Penetration is not ignored, but no wasted damage due to overkill is assumed. I also use real dps, the raw dps is calculated from damage, ap, model count and reload time. Animations are of course not included. Velocity is not part of it, and tbh it's hard to do properly anyways. If we start to assume dodging and infinitely good opponents then every ranged unit in the game does 0 damage so I am not sure how productive that is. :smile: Block chance by shield and PR for example is scaled down a bit to simulate that it's not always in play. Same with MD because it only matters in melee, and same with armor because it only mitigates non-ap damage etc. Nothing is exact, everything is scaled and approximated. The key is that if the model correlates well with a subset of units that are generally accepted as well balanced, then any outliers from this line stands out due to a higher density of positive or negative parameters per cost.

    It also explains why the Helblaster Volleygun ends up as one of the best units in this chart. According to this model it has the best offensive capability, because it has a really high damage output (in theory).
    • 1800 (600 x 3) damage per 10 seconds according to the unit card.
    • Almost 3 times as much as a Great Cannon
    • When you test it yourself you will notice that this is not true at all, in practice it looks totally different.

    It's no surprise. In reality the Helblaster Vollery Gun isn't nearly as useful. Values such as accuracy, penetration and shots per volley play an important role as mentioned earlier on. Reload speed is another issue: Outiders have 5.4 reload speed .... in theory. But it takes them ages to reload due to the animations. Most reload speeds are an outright lie, therefore the damage per second stated by CA is wrong as well.
    I did include reload etc, but what makes the hellbaster stand out is that it's much more accurate than the organ gun at long range. That let's it's damage scale much higher than the organ guns. Does the DPS stated by CA take animations into account???

    Just as a side note: Did you add Ratling Guns to this model? They have a hypothetical damage output of 8136 per unit and decent survivability, speed and so on. (339 Dp10S x 24 on large unit size.)
    I removed them for now, partly because of the extra slow effect that I should account for, and partly because the DPS broke the graph, it was so insanely high lol. The main reason why I didn't want that high dps was that my calculated dps didn't match the dps stated by CA so there's something fishy with the stats of ratlings.

    Conclusions:
    It's a trap to draw overhasty conclusions, because this model does not take into account some of the most important values. For example saying that the Helblaster Volley Gun is more efficient than a Great Cannon is false, unless we are talking about very specific circumstances (the graph suggests that the Helblaster is way more cost efficient, but this isn't the case when you take all possible scenarios into account).
    Yeah I am not saying that. This is just what the model spits out. Range should probably be reinforced, but the main problem for hellblaster is that there's some negative synnergy there, like very slow artillery with shorter range and direct fire which can be blocked. That's a negative synnergy that can never be described, that I have to just accept. It does however show how bad the organ gun is because it doesn't even have the potential of the Helblaster.

    My apologies, I don't want to sound too harsh. I'm aware of that I mainly focused on flaws and criticism of this model. But I really like the idea of number crunching. And I have to say that this post is way more interesting than a lot of other feedback. I bet it's going to trigger a lot of people, simply because it mentioned wArPlOcK jEzZaIls. It's a very basic model with several severe issues for various reasons nonetheless.


    Well, my criticism is also a little bit inappropriate, because Disposable Hero stated that these charts are mainly for fun, they were created within a few hours during a flight and I have to say that they are entertaining and interesting at the same time.


    Keep it up. I think something like that would be really interesting for melee infantry. The calculation becomes easier, because there are less variables involved. :smile:
    No problem, these things I am completely open about and I have no claims with these figures. It's totally not perfect, just a bit interesting. It sort of was a bit better than I had expected though so that's why I shared it.
  • Disposable HeroDisposable Hero Posts: 3,771Registered Users
    Loupi_ said:

    Interesting modelling sir! I think the reason why the waywatchers are so close to the best fit line is because on their own they are totally fine. In a real battle there will be 3 or 4 of them, which will raise their defence and offense meta score, since they protect each other and therefore kill more.

    It would interesting to see how their position changes if you reduced their MD and HP to the level of shadowwalkers for example.


    Also where do skirmish units like deepwood scouts, shadow warrior, shades, chameleon skinks and shadow walkers fall?

    Cheers! I could make that test later, I will need to improve the model a bit more first.

    DWS and SW are in the graph, none of them stand out. Shades can easily be added, for skinks and walkers I need to add in a factor for poison that I don't have now.

  • KurnothHunterKurnothHunter Posts: 234Registered Users

    All that data? Misses something very important, the rest of that faction's roster!

    Again, Skaven are missing flying units, cavalry and warhounds.

    i guess that's why it is called for fun, only ranged units

    honestly, guys, we have already told everything we could about new DLC stuff, lizards, skavens, jezzails, doomflayers whatsoever

    I guess CA heard enough (and more than enough) from everyone around here


    interesting data, can it model other stuff? like damage from charges chart? or armor and physical resistance correlations?

  • MadDemiurgMadDemiurg Posts: 2,381Registered Users
    Any model that shows that hellblaster is good, cannons are bad and jezzails have better def than bowshabti is way way off. There's so much wrong with this I wouldn't even know where to start.

    Team Skaven

    Team O&G

  • TlaxtlanSoothsayerTlaxtlanSoothsayer Posts: 2,321Registered Users
    edited June 8


    I did include reload etc, but what makes the hellbaster stand out is that it's much more accurate than the organ gun at long range. That let's it's damage scale much higher than the organ guns. Does the DPS stated by CA take animations into account???


    That's the issue with the "damage value is the damage over 10 seconds" stat on CA's unit cards.

    It doesn't take reload animations into account. This is the reason why I mentioned Outriders: In theory they have 5.4 seconds reload speed and 34 Dp10s.

    In practice they take 9 or even 10 seconds instead of the stated 5.4 seconds to shoot. The reload animations cut the real Dp10s of Outriders almost in half.

    This is the case for many other units, too. For example Thunderers. Or Hawk Riders take longer to reload as well. It pretty much affects every unit. It's just that some units have really fast animations and others take ages to shoot, so the Dps of some units is close to their Dp10s on the unit card, for others not so much.

  • GeneralConfusionGeneralConfusion Posts: 908Registered Users
    Looks cool! I would definitely be interested in refining the values going into such a chart until real data can be extracted; analysis of this kind is very susceptible to the GIGO effect, obviously.

    As for Ratling Gun DPS; the reason the stated values are absurd is, I think, because the calculated values the game displays don't account for the time it takes to fire. Each Ratling Gun model fires a volley of 18 shots with something like a third or a quarter-second between them - so the full volley takes 5 seconds or so to fire - and then stops and reloads for about another ~4-5 seconds. The game, however, doesn't account for those 5 seconds of firing; it just assumes that all the damage is instantaneously done at the start of a 'firing cycle', and then the reload animation starts. That pushes the displayed DPS way up.
  • Disposable HeroDisposable Hero Posts: 3,771Registered Users

    Looks cool! I would definitely be interested in refining the values going into such a chart until real data can be extracted; analysis of this kind is very susceptible to the GIGO effect, obviously.

    As for Ratling Gun DPS; the reason the stated values are absurd is, I think, because the calculated values the game displays don't account for the time it takes to fire. Each Ratling Gun model fires a volley of 18 shots with something like a third or a quarter-second between them - so the full volley takes 5 seconds or so to fire - and then stops and reloads for about another ~4-5 seconds. The game, however, doesn't account for those 5 seconds of firing; it just assumes that all the damage is instantaneously done at the start of a 'firing cycle', and then the reload animation starts. That pushes the displayed DPS way up.

    Aye, I put in some more time this morning and got it up to 0.68, better than I had hoped so I'll see how far I can push it.
  • Disposable HeroDisposable Hero Posts: 3,771Registered Users


    I did include reload etc, but what makes the hellbaster stand out is that it's much more accurate than the organ gun at long range. That let's it's damage scale much higher than the organ guns. Does the DPS stated by CA take animations into account???


    That's the issue with the "damage value is the damage over 10 seconds" stat on CA's unit cards.

    It doesn't take reload animations into account. This is the reason why I mentioned Outriders: In theory they have 5.4 seconds reload speed and 34 Dp10s.

    In practice they take 9 or even 10 seconds instead of the stated 5.4 seconds to shoot. The reload animations cut the real Dp10s of Outriders almost in half.

    This is the case for many other units, too. For example Thunderers. Or Hawk Riders take longer to reload as well. It pretty much affects every unit. It's just that some units have really fast animations and others take ages to shoot, so the Dps of some units is close to their Dp10s on the unit card, for others not so much.

    Aye, I am aware but i don't know if there's anything to do about it. I don't feel like clocking every unit manually and I don't want to selectively edit input... But the effect is big in a few cases I reckon, like wlc vs great cannon.

    I didn't include any cav yet either but I probably should. That will help estimate to value ca put to speed vis a vis dps. Maybe, it will introduce new problems though with cb and ma being more significant.
  • PippingtonPippington Posts: 2,026Registered Users

    The key here though is that all input is the in game values, so there is no bias there. The model itself is arbitrary in the way that all scaling values and the way all these parameters are merged are trial-and-error, based on some general intuition how this should be working.

    [...]

    It can be hard to pin down exactly what creates the offset, in order to get that you need to change parameters back and forth and look at the live updates in the graphs to see the effect on all units simultaneously.

    That sounds very much like it'd be vulnerable to being (consciously or subconsciously) sculpted towards confirming the kind of result you were expecting before you set out.


    Get on, Kroq-Gar, we're going shopping

  • Disposable HeroDisposable Hero Posts: 3,771Registered Users

    The key here though is that all input is the in game values, so there is no bias there. The model itself is arbitrary in the way that all scaling values and the way all these parameters are merged are trial-and-error, based on some general intuition how this should be working.

    [...]

    It can be hard to pin down exactly what creates the offset, in order to get that you need to change parameters back and forth and look at the live updates in the graphs to see the effect on all units simultaneously.

    That sounds very much like it'd be vulnerable to being (consciously or subconsciously) sculpted towards confirming the kind of result you were expecting before you set out.
    Yes, there is that danger but I try to avoid that as much as I now can by having a dataset that includes all features in several places and fit them all at the same time. It's a common problem with parametrization though, it's always biased on the training set... So I'd want the training set to be diverse and not have unique points.

    Anyways, there's a limit how far it's reasonable to push this, consider it a fun little exercise just.
  • SarmatiansSarmatians Posts: 3,260Registered Users
    I think game 1 reload animations are in general longer and wonkier.

    Take cannons and WLC. Looking at unit card, WLC fires about 5% faster (18.9 vs 19.8 seconds). In reality it fires almost 50% faster, iirc, managing to fire off almost three volleys for cannon's two volleys.
  • Disposable HeroDisposable Hero Posts: 3,771Registered Users
    So here's an improved fit, what I did was mainly increase the impact of range and add the marksman/accuracy skill to add into cal-area.



    All in all not too different from the earlier version but the over all fit is better.
  • Disposable HeroDisposable Hero Posts: 3,771Registered Users

    So here's an improved fit, what I did was mainly increase the impact of range and add the marksman/accuracy skill to add into cal-area.



    All in all not too different from the earlier version but the over all fit is better for artillery now.

  • GeneralConfusionGeneralConfusion Posts: 908Registered Users
    I mean I do still have to agree with the earlier comments that say that if your model measures a Hellblaster as being more efficient than a Cannon the model has gone wrong somewhere, lol.

    Keep working at it though!
  • Disposable HeroDisposable Hero Posts: 3,771Registered Users

    I mean I do still have to agree with the earlier comments that say that if your model measures a Hellblaster as being more efficient than a Cannon the model has gone wrong somewhere, lol.

    Keep working at it though!

    Consider it proof of that I am not tailoring it to be what I want it to be. :smile:

    I will make some tests later, but iirc it also has multiple shots, like ratlings, so the dps may be over estimated for the same reason. Otherwise in pure stats it has that much potential. It's huge dps, mainly ap at 295 meters with actually quite good accuracy. The weakness is the same as a cannon so nothing to counter balance that except range. What the model can never catch is the negative synergy here of direct fire in combination with shorter range and very bad mobility. In a way ca thinks it's better than cannon too since it costs 350 gold more.... :)
Sign In or Register to comment.