RNG (That's Random Number Generator)

[El_Condoro]

It's off topic so anyway...

What I'm trying to understand is why in the 10 die rolls there is only ONE purple/blue suppressed roll and yet it says in the text below that THREE points were suppressed.

TheGaryMouser's explanation may be the key but I'm not sure.

When a unit takes 1 kill, this also generates some suppression. The formula is (50-stars*10)%. This means that a green unit takes additional 0.5 points of suppression per every kill. This models the fact that green unit becomes disorganized and demoralized when taking damage. Veteran 5-star units take zero additional suppression because they are so tough they are immune to this effect. All the rest units lie in between these two extremes.

Thanks Rudankort. Makes sense. Maybe I should start another thread for this but can the combat calculations be explained more somewhere, sometime? They might be complicated mathematical algorithms, formulae etc etc but it would be fascinating IMO.

[Some1]

If you're told you'll be getting a 1-6 outcome in your favor, then actually getting a 0-5 or 2-4 etc is all good and well. But if there's a chance you will get a 7-2 outcome (essentially devastating your attacking unit) then that needs to be flagged up beforehand. And the best way to handle it would be to
1) use "rugged defense" to make it clear something unexpected has happened
2) make sure the player had information at his fingertips telling him there was a real risk of this happening, essentially warning him not to trust the combat predictor.

Saying "you could have looked at the detailed statistics screen" is not an acceptable excuse - it simply isn't a good user interface that expects users to go through hoops like that.

The game needs refinement:
1) reserve truly wild outcomes to exceptional results only (like "rugged defense"). Make sure they actually occur on others (like for "ambush" which currently is not nearly dangerous enough).
2) ensure the quick combat predictor gives sufficient indication an exceptional result might happen. THEN it's okay to expect the player to delve into the detailed statistics (or to simply make another suppressing attack and see if that makes the risk goes away).

There's nothing wrong with "uncertainties of battle". As long as you allow the player to make those calls himself. Getting a 1-7 outcome is totally fine, if the player knew of the risks. But you don't do that right now, not unless you bring up details screens all the time, or have considerable experience playing the game.

Knowing those risks is currently far too obscured, especially for casual players, and that is the real issue at hand.

Not categorically removing randomness. Randomness is essential.

I agree 100%. As already mentioned, the user needs to have information before, during and after a battle which allows the player to understand what happend and why.
This is especially important for new players, because if they are unable to understand cause and effect (due to hidden informations - for newbies), they wont keep playing for long.

Molve provides an excellent summary of the real core of the RNG-issue, and what could and should be changed in the future from an user point of view.

[Xerkis]

It's off topic so anyway...

What I'm trying to understand is why in the 10 die rolls there is only ONE purple/blue suppressed roll and yet it says in the text below that THREE points were suppressed.

TheGaryMouser's explanation may be the key but I'm not sure.

When a unit takes 1 kill, this also generates some suppression. The formula is (50-stars*10)%. This means that a green unit takes additional 0.5 points of suppression per every kill. This models the fact that green unit becomes disorganized and demoralized when taking damage. Veteran 5-star units take zero additional suppression because they are so tough they are immune to this effect. All the rest units lie in between these two extremes.

Thanks Rudankort. Makes sense. Maybe I should start another thread for this but can the combat calculations be explained more somewhere, sometime? They might be complicated mathematical algorithms, formulae etc etc but it would be fascinating IMO.

YES .... Please.
I stopped watching this thread too closely because as I stated way up above – I am just too confused by all these word. I need to see the algebraic formula for it to make any sense at all to me now.

[Amadeus]

I agree 100%. As already mentioned, the user needs to have information before, during and after a battle which allows the player to understand what happend and why.
This is especially important for new players, because if they are unable to understand cause and effect (due to hidden informations - for newbies), they wont keep playing for long.

Molve provides an excellent summary of the real core of the RNG-issue, and what could and should be changed in the future from an user point of view.

I apologize if this is a stupid question. I am new to Panzer Corps. What do the dice rolls exactly do?

When I watch the last combat result there are several numbers shown in red, blue and white. They are all between 1 and 99 I guess.

As I understand there is one number for each unsuppressed strength. The red numbers are mostly above 50, the blue and white mostly below 50. As I said mostly but not always. I see a kill rate too.

My question is what do the dice rolls exactly do? How are they diced?

[Tarrak]

I apologize if this is a stupid question. I am new to Panzer Corps. What do the dice rolls exactly do?

When I watch the last combat result there are several numbers shown in red, blue and white. They are all between 1 and 99 I guess.

As I understand there is one number for each unsuppressed strength. The red numbers are mostly above 50, the blue and white mostly below 50. As I said mostly but not always. I see a kill rate too.

My question is what do the dice rolls exactly do? How are they diced?

If you look a line above the dice rolls there is a line saying something like for example: Miss: 45%, Suppress: 11%, Kill: 44%. Now you roll a die for each unsuppressed strength point (well there are exceptions from this for units with non standard Rate of Fire, mostly heavy artillery). In this case we are looking now at every roll from 1 to 45 would mean a miss and they are shown white on the result screen, from 46 to 57 a suppression and this is shown in blue and from 58 to 100 a kill which is shown in red.

The chances for a miss, suppression and kill are calculated depending on the attacking unit appropriate attack value and the defending units defense value. This can be modified through different circumstances and terrain modifiers but if its the case it is listed in the combat log.

To determinate who shots first the initiative value of both units is compared to each other. Here again different modifiers like for example ambush are applied. The unit with the higher value get to shot first but it does not always mean with the full strength. This depends kind of on the difference between the initiative values and was explained in detail somewhere in this thread before but sadly i got lost somewhere in the middle of the discussion

[Amadeus]

Thanks, that helps a lot. Too much luck for my taste. A clear yes for the poll.

[Molve]

Again, the poll is worded this way: "Would you feel the Panzer Corps game play experience would be better if the combat prediction numbers were 100% accruate (sic), except for rugged defense encounters?"

At least one "No" vote (mine) should not be taken as a suggestion "everything's fine now, nothing needs changing". It only means "the suggestion to remove all variability (reaching 100% accuracy) would be very bad indeed."



If you instead reformulate the question to read: "Would you feel the Panzer Corps game play experience would be better if the combat prediction numbers were substantially more reliable, excepting ruggeds, and the risk of getting such a rugged encounter would be indicated somehow*?"...

...then my answer would be YES, and the poll results would (again IMHO) be much more interesting to evaluate.


*) Indicated how you might ask? Well, perhaps by coloring the combat prediction green, yellow and red to indicate a negligible, potential and considerable risk of something going very wrong (read "get a rugged result"). Essentially, the color represents whether enough variables are unknown (a high risk) or almost all variables that goes into the predictor are known entities (such as no entrenchment, a clear initiative gap between attacker and defender - including mass attack effects, no adjacent hexes in the fog that might contain supporting artillery etc etc)

Essentially having the program putting information at the player's fingertips, making the UI more friendly by not requiring the serious player to delve into subsequent details windows before every attack (only those the predictor warns you are in red or yellow). As long as perhaps half of all combats can be green, you've just cut the amount of dialogs the serious player needs to browse through in half!

[TheGrayMouser]

No offense but what is really the ? here? Is it simply wanting the prediction to be more accurate or do people want the nature of the combat mechanics changed?

I bring this up because can the prediction even be made more accurate at all, with out assaulting the player with a ridiculous amount of statistical data??

Lets assume its just to make the prediction more accurate. Perhaps someone with true statistical skills can come up with a chart or something but the way i see it is this

Assume two exactly equal units:
as it stands predictor states 3-3 odds Ist , each unit is rolling 10 % dice and causing kills suppression, quite a bit of variance is possible here already assuming tru random % dice rolls.

the prediction I would guess, is merely the likly average or mean? of what could happen.
How much variance could there be , I have no idea but no doudt you could prove it out and if so, to what purpose?
Would it be remotely usefull if the predictor ran/displayed like this: (completely made up #'s here)
3-3 23.5% chance
3-2 17.5%
2-3 17.5%
4-3 13%
3-4 13%
5-2 7.6%
etc
add likly 10-15 more lines of etc (hell maybe more, although it would be finite)
0-0 .0005% chance

BTW that doesnt even scratch the surface on how complicated( and thus how wider the deviation could get) because this doesnt even factor in the REAl results when the random intiative is rolled when combat is actually done , nor does it illustrate what happens when units take supression from kills ( the amount varying based on the units experiance, an undocumentd feature /combat mechanic)
With so many variables and conditionals ON TOP of those variables, i really dont think its possible to make the predictor more "accurate" unless you just want to deluge a player with statistical charts, or if you want to completly change how combat works in the ist place.

[HeinzG]

Hell Molve, want me to buy new glasses ?!?

But honestly, I support your thoughts. Take my "No" vote as Molve stated it above.

[Amadeus]

Again, the poll is worded this way: "Would you feel the Panzer Corps game play experience would be better if the combat prediction numbers were 100% accruate (sic), except for rugged defense encounters?"

If you instead reformulate the question to read: "Would you feel the Panzer Corps game play experience would be better if the combat prediction numbers were substantially more reliable, excepting ruggeds, and the risk of getting such a rugged encounter would be indicated somehow*?"...

...then my answer would be YES, and the poll results would (again IMHO) be much more interesting to evaluate.

Why don't you make a new Poll. That is not that difficult I guess. I already vote yes and would like to do it again...

[IainMcNeil]

The only way to make teh predictions more accurate is to change teh mechanics so results are more predictable and this would in my opinion really break the game mechanics.

A bug part of being a good general is planning for the bad outcomes. Usually the best generals are those who hope for the best but plan for the worst. E.g. don't put yourself in a position where a bad result leaves you exposed if you can avoid it. This doesn't mean you don't take chances - you can't win without taking risks. But you have to take them in a way that you minimize the downside and maximize the upside. Any individual result can go against you but over the course of a battle you'll win.

[HeinzG]

Ok, thank you Ian for sheding some ligth on this topic.

If it ist this way, no changes should be made. Would change gameplay to the worse.

So my vote goes stil for a clear No

[Razz1]

The problem with randomness, is the fact that it is too great.

Just tone it down.

Try playing the Allies and you will see the problem ten fold.

[Amadeus]

The only way to make teh predictions more accurate is to change teh mechanics so results are more predictable and this would in my opinion really break the game mechanics.

A bug part of being a good general is planning for the bad outcomes. Usually the best generals are those who hope for the best but plan for the worst. E.g. don't put yourself in a position where a bad result leaves you exposed if you can avoid it. This doesn't mean you don't take chances - you can't win without taking risks. But you have to take them in a way that you minimize the downside and maximize the upside. Any individual result can go against you but over the course of a battle you'll win.

Pity, because one additional point for Newbies that we don't talk about yet. You could hardly get a feeling when luck has a substantial matter. How could I plan when the outcome is unclear and untransparent. I wish a bit more of "Easy to learn and difficult to master."

[IainMcNeil]

I think the issue is not randomness, it is lethality. Early war combats are fine. Late war combats are more lethal and swings as a result are larger. There is no way for randomness to effect one side more than the other.

Lethality is something we've been discussing internally and bringing the late war game baalnce back in line.

[Molve]

The only way to make teh predictions more accurate is to change teh mechanics so results are more predictable and this would in my opinion really break the game mechanics.

A bug part of being a good general is planning for the bad outcomes. Usually the best generals are those who hope for the best but plan for the worst. E.g. don't put yourself in a position where a bad result leaves you exposed if you can avoid it. This doesn't mean you don't take chances - you can't win without taking risks. But you have to take them in a way that you minimize the downside and maximize the upside. Any individual result can go against you but over the course of a battle you'll win.

I think the issue is not randomness, it is lethality. Early war combats are fine. Late war combats are more lethal and swings as a result are larger. There is no way for randomness to effect one side more than the other.

Lethality is something we've been discussing internally and bringing the late war game baalnce back in line.

With respect, Iain, you're changing the subject.

Of course you can make the predictions more accurate without changing the underlying mechanics. (Not that I agree you can't change the mechanics without breaking the game, but that's another separate issue) You could make the combat predictor give you a hint of how reliable the prediction is. Summarize the unknowns and give this to the player instantly.

The question is revolving around accessibility. A "beer and pretzels" game like this should be accessible. Requiring a player to double-check predictions against statistical data given by pop-up dialogs isn't. End of story.

Also, the "hope for the best and plan for the worst" bit isn't really addressing the user-interface issues. At worst, it's an excuse not to have to do anything. More charitably, it could also be that you have played so much you have become desensitized to the frustration felt by newcomers as well as veterans remembering the old panzer general series, who in my mind, spent more effort on these issues.

So yeah, the issue is very much randomness. Specifically: the unreliability of the predictor. Once you've fully suppressed your target and taken other predictions, I don't mind getting a 1-7 outcome instead of a 0-9 outcome. That's just life, and predictions should NOT become anywhere close to 100% (as asked by this poll).

The thing is instead that prior to taking these precautions the predictor might still show you will win handily. And then, when your attack results in 6-2 (without a "Rugged" indicator even!) that is a "wtf?" moment. Completely needless frustration on part of the player.

Feel free to discuss late game lethality. It's just not the subject here (the problems discussed are just as great in the early war as in the late war), and please don't fix that in the belief it would address the complaints that prompted Kerensky to start this thread.


Sincerely yours,
Molve

[TheGrayMouser]

Molve, nobody is forcing players to cross referance predictions vs results, you dont need to do that to enjoy the game and to be good at the game, it is a beer and pretzels style game AND it does capture the easy to play , difficult to master feel.

Why dont you believe the developer when he says you CANt make the predictor more accurate without changing the game mechanics? Three or 4 posts up I came my two unscientific cents of why you cant , there are too many conditionals on top of quite complicated statistical probablities (remember your rolling on average 20 percentile dice per combat and thats just for starters) Roll two 6 sided dice with one trying to beat the other to win. How many outcomes are there ? 36 I believe , yet the more dice, the more variables exponentially more outcomes, but yet the predictor itself, to be usefull needs to be simple 7-1, for example . Please explain mathematically how you can make a pediction of random #'s more accurate, gambling / lottery players around the world would love to know

IF you really want the predictor to be more accurate, just edit the data files. Basically reduce the effects of initaive to .05 ( or even 0) effect vs the current 20% modifier and Im pretty sure the combat results will be much much closer to the prediction.

[Some1]

Even if i assume that the combat predictor can't be made more accurate, i still see two worthwhile improvements:

1.) If not already done, shape the output of the random generator so that the values are normally/gaussian distributed, see http://en.wikipedia.org/wiki/Normal_dis ... stribution (if i have understood the wikipedia entry correctly). While this wouldn't stop wild swings of the combat results, it would at least reduce the number of those.

2.) As a few time already mentioned by Molve, the user should be provided with an indication of how trustworthy the displayed combat prediction really is. There i see two factors:

• The number of hex field still under the fog of war neighbouring the enemy unit, e.g. one field is bad, and more than one basicly invalidates the result of the combat predictor.

• The possible/likely span of results. The primitive solution would be calculating the worst and the best possible results, and using those somehow as an input for the combat prediction GUI. The 'correct' solution could be to iterate through all possibilities to get the probability distribution. This could than be used to calculate the best and worst result based on a probability of e.g. 95%, and use the difference of the two results as an input for the combat prediction GUI. The 'correct' solution would be computationally expensive, but this shouldn't be a problem with todays hardware, as long as it is not too expensive and as long as it is not used by the AI.

[Xitax]

Hmm, the scope of the distribution (gauss curve or however it actually is) would be nice to know but I'm not sure we know what that is unless the devs can illuminate us. Has anyone graphed it to see what it actually looks like?

As to showing the span of results, iteration is a bad and costly model (in terms of cpu cycles), but a reasonable model would be to show the best and worse result within one standard deviation of the mean. However having said that I am against it because it will clutter the GUI and I don't want any more stuff in there. I don't have a problem with knowing that the combat predictor is not 100% accurate and the numbers that we do get there are a good enough to give me a feeling for possible results as they are right now.

[Some1]

The 'correct' solution would be computationally expensive, but this shouldn't be a problem with todays hardware, as long as it is not too expensive and as long as it is not used by the AI.

I have just done a rough estimate of the effort and realized, that the 'combat dice' has 100 sides. So in case of no Artillery/Flak support for the attacked unit, only the number of iteration needed for a 10 strength attacking unit, and without the iterations of the attacked unit and the iteration for the variable initiative, i estimate 100^10 = 1E20 iterations. A brute force approach would have been a _very_ bad idea.