Experience Documented

[Molve]

Please don't turn this thread into an argument, Flint.

Well, that's what forum discussions are for, aren't they?

Please don't make it sound as if I am repressing your freedom to post on these forums. I was asking you to take the "overstrengthing is totally worth it" spiel outside this particular thread; not more, not less.

My only comment regarding your numbers is: okay so you have proved that better is... better. But is it worth it? That absolutely vital issue is 1) not something you can prove as long as you keep excluding costs, and 2) still something I feel is best discussed in another thread, as to not clutter the XP discussion of this thread.

[Molve]

I provided some test results in the RNG thread, and extreme results (mostly unfavorable) were far too likely.
You can check out the exact numbers in that thread.
Once I can muster the energy, I'll make some more statistics.

I already checked my RNG, and on a big number of tests (that's what Molve requested above) it is almost perfectly uniform. I know that sometimes it can generate long sequences of poor (or good) rolls, but in the context of predictor discussion this is completely irrelevant. Any quirks of RNG average out over time.

(This is not to say there is nothing to fix, in background I'm thinking how to improve the situation with RNG. Maybe some more advanced algorithm will work better, there is a bunch of different pseudorandom algorithms out there, but the problem is, it is hard to even formulate criteria for an ideal RNG in a game like this.)

Thank you, Rudankort.

By "almost perfectly uniform", are you talking about the sequence of RNG numbers or the correlation between actual combat outcomes visavi the "quick predictor" hovering thing?

Because I should probably clarify that at least I am almost completely uninterested in RNG seeds and the random sequences themselves. I'm sure they're fine, but above all: that's not at all what I'm concerned about.

From my perspective, a programmer (you) would probably not have to do anything at all regarding the RNG or how the random number sequence algorithm. Instead, what you (again probably, it's your code after all) should focus on, is "cooking" the numbers after the combat routine has spitted them out.

That is, instead of trying hard to fine-tune all the parameters going into a combat, simply accept that it's entertainment first and simulation only second. It's perfectly fine to tweak the numbers afterwards and save a lot of work and anguish trying to do the impossible (creating the "perfect" combat simulation).

If the combat predictor says 0-10, an actual outcome of 6-4 is simply unacceptable (turning the predictor into a joke). So why don't you simply add a dozen IF statements that catch these outliers, and cook them before sending them on to the rest of the program?

(I know this isn't exactly on topic either, but I needed to respond lest Rudankort veers too far off into a RNG discussion. I respect Rezaf's wishes, but feel that issue is much less urgent than the seemingly similar but probably completely different predictor accuracy one)

[Rudankort]

Though, I fear you're set in your opinion about this and will not change it nomatter what I or anyone else says.
I don't really know what else to say, it'll take you a couple of minutes - you didn't deny that - but your flat out refuse to do it.
You have already spent more time reading and replying to posts explaining that you're reluctant to do it than it would have cost you to just implement it. End of story, I guess. I cannot convince you because you don't want to be convinced.

You are right that I could spend the time on more useful things, but hey, I did not write on the forum for a long time. What would you say if I abandoned you again?

As for the feature in question, it is not about time. When I change anything in the game, I want to be sure the change is good and improves it for a big majority of players out there. If you know something about programming, you should also know that you better not change the code unless there is a very good reason for this. Your "2 minute" estimation is a pure speculation, you never know how much time the change is going to cost you, in case it causes an issue and you need to debug it. If the issue is rare and does not repro on your PC, it can take DAYS to collect the info and pin the cause, and will ruin the experience for some of the players. And this is only top part of an iceberg...

At least the MP-stuff-relies-on-it is a good reason, though, honestly, would it be a problem to make reseeding NOT work in multiplayer games or when starting a replay? I know some programming myself, but it's somewhat difficult to predict, as it depends on how you have implemented this.

...and the bottom of an iceberg is, every option you add to your code complicates it, adds more paths of execution, more tests you need to do etc. It is impossible to calculate how much time you will spend in perspective, extending and maintaining more complex code. You can argue that it depends on code organization, but in any case, more complex functionality requires more complex code, there is no escape from that. And yes, the more changes you do, the more time you need to spend on refactoring too.

If I see a clear benefit, all the above becomes irrelevant. My job, as a developer, is to cater for player's needs. I guess, people from the beta team can tell you that I'm always open to feedback and it is not so hard to convince me. In fact, in many cases convincing is not required because the benefit is clear. I just slap my head and wonder why I did not think about that idea myself. But if I don't see a benefit - sorry, I'm not going to change the code. Forums are great for getting feedback, but there is always a danger. A small but very vocal group of people can make you believe that something is important, while in reality it is not. For this reason I don't take anyone's word that something is important, I must understand why it is really like that. With reseeding the saves it is simply not the case (yet?).

[Molve]

I don't really know what else to say, it'll take you a couple of minutes - you didn't deny that - but your flat out refuse to do it.
You have already spent more time reading and replying to posts explaining that you're reluctant to do it than it would have cost you to just implement it.

Sorry rezaf - I see your point, but you're starting to come across as if you somehow are entitled to coding changes just as long as they're small enough.

If you are a programmer you should know that there are no such thing as a "quick fix", not when you work in a larger team for a responsible company anyhow.

Yes, the code change itself might just take a few minutes to implement. This is irrelevant. Any change, no matter how small, needs to be tested, documented, checked-in to the code repository, distributed, and delivered.

In short, please reconsider your approach here. You're antagonizing the team, and that most certainly isn't going to help your cause.

I hope you can see this message as the sincere attempt at mediation as I'm intending it to be.

Best Regards,
Molve

Edit: Ninajed by R himself

[Fimconte]

In my mind reseeding saves would only fix the symptoms, not the root cause.
Which is that you can have considerable streaks of bad (or good) RNG,
where the combat predictor gives you:
0-8, 2-6, 0-4, 0-2
odds and in reality you get:
6-2, 4-0, 4-1, 5-1.

And yes, perhaps over 100000 "rolls" it averages out, but it's still incredibly frustrating to see frankly absurd rolls at times.
(0 entrenchment infantry on "Clear" dealing crippling blows to Panzer IVG when predicted a 4-0 results in a 2-6)

[rezaf]

Like I said, you cannot be convinced because you don't want to.

About the two minute estimate: You MUST have a routine that parses the savefile, loads the random seed stored therein and sets it as the base for a newly instanced RNG. If you'd slip in a snippet that just ignored the value stored in the savegame and cooked up a new one, that's all it takes.
No further consequences, no code branching, no complicated stuff coming up. Easy and simple.

I can totally relate to the notion of not wanting to make the codebase more complicated, believe me - I know enough programming to see the wisdom in that.
I still disagree with the "I don't add or change ANYTHING unless I feel it's absolutely necessary" notion. Giving players (for me it'd be users, as I don't code games) options is a great thing, even if it only affects a small portion of the userbase or is even only useful to modders or whatever.

But fine, let's end this here. You think my point of view is worthless and nobody else wants a randomize seed option, it's your good right.
I really tire of this.

Please don't leave the forum again, maybe there's other people who are not such morons as me who have nothing but idiotic ideas.
They deserve your attention, even if I don't.

Edit: Rudankort, about your reply, I could probably outline a way to do this, but I don't want to start a coding discussion on top of everything.
Molve might be right. I guess it's a personality fault that I don't like it when a game designer forces HIS line of thought upon me this way, especially when it's common practice in other games to do it differently. I get it that you think this is egoistic and retarded of me, and since it's your code and not open source, that's all there is to say. I'm not going to make another post about this, which is why I simply edited this post. Hope that settles this for now.
_____
rezaf

[Rudankort]

By "almost perfectly uniform", are you talking about the sequence of RNG numbers or the correlation between actual combat outcomes visavi the "quick predictor" hovering thing?

By "almost perfectly uniform" I mean that all numbers in the range are equiprobable. Random distribution has nothing to do with sequences, because every dice roll is independent from the others. And even if you got ten 1s in a row in a range [1;100], it does not mean 1 becomes less probable in the next (eleventh) roll than it normally is. For this reason, long sequences of "good" and "bad" rolls do not indicate that RNG is faulty, although it can still be bad for the game. That's exactly what I'm thinking about, in between the other tasks.

If the combat predictor says 0-10, an actual outcome of 6-4 is simply unacceptable (turning the predictor into a joke). So why don't you simply add a dozen IF statements that catch these outliers, and cook them before sending them on to the rest of the program?

It all boils down to this. You think that this is unacceptable. I think that it is not only acceptable, but also the right thing to do, as long as such result has low probability. If we establish that prediction is 0-10, but in reality 6-4 is more probable that 0-10 outcome, this would be a serious bug. But I have never seen this, prediction gives you a pretty good idea of what you can expect. There are singular bad results, but they are rare.

Now I agree that some combats are more prone to error than others. If your good expected result depends only on high initiative, you can get beaten in case of a bad initiative roll. If your good result is based on high initiative AND supported by good defense rating, it becomes very unlikely that you end up beaten up. I think that this does not only represent the reality better, but it also gives the game more depth. It takes learning to understand how the combat is resolved, how different unit stats work together to produce the end result. But if you master this, you can avoid a lot of issues with seemingly random bad results. Players who beat the game even on the hardest levels (like Rommel and Manstein) without any reloading in the process consistently prove this.

[Rudankort]

About the two minute estimate: You MUST have a routine that parses the savefile, loads the random seed stored therein and sets it as the base for a newly instanced RNG. If you'd slip in a snippet that just ignored the value stored in the savegame and cooked up a new one, that's all it takes.
No further consequences, no code branching, no complicated stuff coming up. Easy and simple.

And at this point your replays break. The replay must be able to reproduce the whole game, from first turn to last, including all combat results and all random events, perfectly. You change the seed in the middle, you break the sequence. And if I do what you suggest in SP and not MP, it is an IF and a branch.

[macattack]

And yes, perhaps over 100000 "rolls" it averages out, but it's still incredibly frustrating to see frankly absurd rolls at times.
(0 entrenchment infantry on "Clear" dealing crippling blows to Panzer IVG when predicted a 4-0 results in a 2-6)

This is a statistical rarity and an argument that it is indeed realistic combat results. It did not happen often, but there were many examples of unusual combat results in WWII. So as long as the 4-0 predicted result only rarely results in a 2-6, I would call this good programming.

[Molve]

It all boils down to this. You think that this is unacceptable. I think that it is not only acceptable, but also the right thing to do, as long as such result has low probability. If we establish that prediction is 0-10, but in reality 6-4 is more probable that 0-10 outcome, this would be a serious bug. But I have never seen this, prediction gives you a pretty good idea of what you can expect. There are singular bad results, but they are rare.

Now I agree that some combats are more prone to error than others. If your good expected result depends only on high initiative, you can get beaten in case of a bad initiative roll. If your good result is based on high initiative AND supported by good defense rating, it becomes very unlikely that you end up beaten up. I think that this does not only represent the reality better, but it also gives the game more depth. It takes learning to understand how the combat is resolved, how different unit stats work together to produce the end result. But if you master this, you can avoid a lot of issues with seemingly random bad results. Players who beat the game even on the hardest levels (like Rommel and Manstein) without any reloading in the process consistently prove this.

Yes, I think you are right.

You are targeting Panzer Corps towards the master player.

Panzer General was instead targeted towards the casual player. And I would have liked PC to retain that focus.

In other words: I don't care about the players beating the game on Rommel or Manstein. I'm sure they don't need to look up the details screens, because they have mastered all the subtleties already.

But that doesn't help the general experience for the less-hardcore player.

You think it's fine to get wild results as long as they're rare. That's not good enough, being told your unit was struck by a thunderbolt from a blue sky "don't worry it doesn't happen often".

The issue is that the game doesn't make it easy enough to see the warning signs when this is going to happen. Allowing the player to cancel his attack when this danger is imminent. Indeed, I'm not even convinced there are warning signs - it could well be that these results are unavoidable as results of pure and unadulterated bad luck.

But that makes for a bad game experience. Again, as a simulation I'm sure the hardcore players see this as something good. As a game experience, and specifically a PG game experience, it can be extremely frustrating.

But as a game, Panzer Corps fails to cushion regular gamers. Not only is the predictor unhelpful, there is no notion of allowing the player to be in control over which risks he's taking.

I'll readily admit my memories of PG games might be flat out wrong. But they still tell me PG was much better - less "unrefined" if you will - than PC in putting the player in charge over the destinies of his units.

Best Regards

[Vaughn]

I am not a programmer and although I understand some statistics, math isn't really my thing. Most of this conversation is going over my head. I'm just a guy that really likes this game.

Thanks for posting the screenshots of examples of what experience can do but I don't know if it is enough. It seems that experience is not as important as it should be considering what it takes to reach 5 stars. It seems that heroes have more of an impact on actual gameplay than experience does. Heroes are a wonderful addition to the game BTW.

I don't know what the problem is, and I have no clue what the solution would be, so I have no idea what I am even contributing to this conversation.

[Molve]

And yes, perhaps over 100000 "rolls" it averages out, but it's still incredibly frustrating to see frankly absurd rolls at times.
(0 entrenchment infantry on "Clear" dealing crippling blows to Panzer IVG when predicted a 4-0 results in a 2-6)

This is a statistical rarity and an argument that it is indeed realistic combat results. It did not happen often, but there were many examples of unusual combat results in WWII. So as long as the 4-0 predicted result only rarely results in a 2-6, I would call this good programming.

I fear you're talking about very different things.

As a simulation it might be "realistic". But as a game - is it fun?

My point (can't vouch for anyone else) is that a "wild" result from a regular combat outcome should probably never deviate more than a few points from the predicted outcome (if we want a predictor regular players can trust).

A "wild" result from a "rugged defense" is something different. Then it's okay to get those outrageous differences (from the predictor) if and only if the player is told outright this "rugged defense" could happen in this particular combat.

Why? Because that would have told the player he's in charge, he's in control, he's the one with all the facts. It would allow the player to choose between calling off the attack (until the defender was softened more) or to risk it.

As the game is now, you could get almost as devastating outcomes (devastating for your faith in the predictor, not devastating for the health of your unit) with no prior warning, with no way to avoid or deflect the difference.



Do you now see why you're almost talking as if in a completely different language when you say things like "statistical rarity" or "realistic combat", and that this has nothing to do with "good programming"?

[impar]

Rudankort, could you give your opinion on making the predictions more reliable the more experience a unit has?
Is it even feasible?

[El_Condoro]

A "wild" result from a "rugged defense" is something different. Then it's okay to get those outrageous differences (from the predictor) if and only if the player is told outright this "rugged defense" could happen in this particular combat.

IMO there is only so much hand-holding that the game should do and the rest is up to the player and his experience. Rugged defence odds - the exact percentage - can be seen using the Ctrl-click attack before making the actual attack. I think what you're asking is for a warning. That is there, too, but not in that form - it's the entrenchment of the defender and your own experience. You will know that anything entrenched over 4 has a pretty good chance of a RD. Less down to zero but still possible. If you want the exact chance - Ctrl-click. Personally, I don't think any more warning than that is necessay. It's certainly more than we ever got with PG2.

[Rudankort]

Yes, I think you are right.

You are targeting Panzer Corps towards the master player.

Panzer General was instead targeted towards the casual player. And I would have liked PC to retain that focus.

Panzer General was more random than Panzer Corps. Mainly because of the profound effect initiative had there. You win initiative - you shoot first. For this reason an unlucky initiative roll could completely ruin the combat for you. And for the same reason the odds were much more off the mark than in Panzer Corps. This was especially clear in early war. In the later war, experienced units had so big initiative bonus that they won initiative most of the time, so this issue became less pronounced. But this also made elite units really unstoppable, they could fight 10 combats in a row and never take a single loss.

So yes, your memory serves you wrong here.

In Panzer Corps our design goal is to address both master and casual players. For master player there is some depth to the game, and I don't want to reduce it. For casual players there are predictions, which give you a pretty decent estimation of what you can expect. They may not be ideal, but come on, there are so many games out there where you get no clue whatsoever how a combat may end. You attack with no indication at all, and only from experience you can learn which attacks are safe and which are not.

You think it's fine to get wild results as long as they're rare. That's not good enough, being told your unit was struck by a thunderbolt from a blue sky "don't worry it doesn't happen often".

This is completely different. Pure random events are frustrating, and that is the reason why PzC does not have random breakdowns for example (some people requested this feature because it is more realistic). But, as I said, really bad results usually have some ground under them. If you lose a couple more points than you anticipated, it is not a huge problem. In the next battle you will lose two points less. If you suddenly lose a whole unit because of bad luck, it is bad, but personally, I have seen that... a single time during all the time I work with the game. And even then, it was a battle between infantry units in close terrain, which is by definition with low defense on both sides and high casualties. It does not take master degree to avoid such a thing - suppress the enemy, bomb it from the air, use mass attack, attack with the strongest unit first, and there is very little risk.

If for you personally losing a few points of overstrength due to bad rolls is unacceptable, perhaps for you a better tactics is to avoid overstrength. It is your strategy and your choice. Choice is good. Isn't it?

[_Flin_]

Please don't make it sound as if I am repressing your freedom to post on these forums.

Hi Molve, this was not my intention at all I just thought we had a little misunderstanding about what we were actually talking about.

My only comment regarding your numbers is: okay so you have proved that better is... better. But is it worth it? That absolutely vital issue is 1) not something you can prove as long as you keep excluding costs, and 2) still something I feel is best discussed in another thread, as to not clutter the XP discussion of this thread.

Concerning 1)
Well, actually I do not want to "prove" anything. Anyone can play as he likes. If you do not think experience is worth it, I can understand that for a lot of situations.
I am usually not in the situation to be starved of prestige. And there are parts of my force that hardly ever get attacked. For example Level Bombers or Artillery. For those, elite replacement and overstrenghting was a no-brainer for me. It is interesting that you have another opinion. I dont see the point why I should NOT elite replace these units. For me it is worth the cost, and having an array of elite units makes playing the game a lot easier. The possibility to disband units for money in the deployment screen makes me having enough prestige whenever I want to. And if I do a horrible mistake, I go back a few turns and load again.
So... I cannot understand why anyone shouldn't elite replace their units, and overstrength them, if it's worth it. You sound as if you do not think it is ever worth it. I disagree with that. But maybe I just understood you wrong.

Concerning 2)
This thread is about the power of experience and overstrengthening. And if it is not powerful enough. Kerensky tried to show that it is powerful. I agree with that. And I think that it is worth to have experienced and overstrengthened units, while it is impossible to have ONLY units like that, especially in the tank/infantry department (not to mention recons ). And since there is no "common questions" thread about the exact workings of experience, I thought I'd sprinkle in some more details.

[deducter]

Here is an images from my test of Closed Beta DLC 41, Rommel difficulty, at the end.


Panther-esque armor vs T34.

My panzers all had 2-3 stars experience, vs 0 experience for the AI.

How can I afford to overstrength everything on Rommel? Because the experience made my panzers nigh-invincible, as shown in the first screenshot.

I shudder to think how much more powerful my units would be with even better experience modifiers.

The infantry loss with an auxiliary, and the AT loss because I put it in a forest without artillery cover.

Kerensky EDIT: Had to snip one, sorry.

[Kerensky]

And that's a good example of the big picture.
Experience by itself? Not a big effect. High experience vs no experience when the high experience has leader bonuses too? That can quickly spiral out of control.

Sure Russian units can be brought up to par by giving them experience, but the Russians of 1941 weren't experienced troops. To give them experience as a balancing mechanism would really fly in the face of history, something we're trying to stay truer to.

Once we get to DLC 1942 (next installment BTW) and beyond, then Russians can have all kinds of experience on them to even the playing field because it is historically appropriate, and good game play (1941 was an easy year for the Germans until Winter, while 1942 and onwards were anything but 'easy')

[deducter]

Aww, that second screenshot is the most telling though. Ah well, I'll post it again when the DLC is released.

[Kerensky]

Yup sorry, once we go live feel free though.
Between here and there though, expect balance changes.