PDA

View Full Version : Are you sure <100% hit random is working?



UserF
06-18-2013, 12:01 PM
Yes, I know, I have complained about this since Beta and Arnie assured me it was checked and is correct, however I still find it hard to believe. I consistently miss 90% and sometimes 80% (I rarely even try anything less than 90% due to how often I miss. A great example was the last game I player. The opponent drilled me twice early on for 90% with wp full power from archers and being that I was now scrambling due to those hits and the change of strategy I was forced to play I tried to strike back with my own BM at 90% with wp maxed. Guess what, I missed. So considering there were only 3 90% shots and odds were 9 out of 10 should hit I was a bit miffed, though yes, I could have been that one, and the next 7 would hit, great.

This is not an isolated experience, I have missed a 90% then a 80% (or any similar variation) in the same match, and again odds are strongly against that happening.

I also know that we should avoid taking chance shots when possible due to this fact that it is a chance, however everyone knows there are times you must do so if you want any chance of victory.

My thoughts are maybe the random chance is working but at the wrong level, i.e. at the game universe level (all matches) or even an individual match level whereas I think it should be either at a team or individual unit level. And while it may be impossible to implement tracking a units chance hits from match to match seems to be the fairest option to me. I say this as I seem to have really bad luck here and it is a bit frustrating to believe its just me. In fact in speaking with other players that have been around awhile I don't believe its just me.

Can the devs look at this again? Chance really is painful when it works against you time and time again in a game of strategy.

I'd rather have a 0% chance to hit that forces shield break over this if it continues to work against me and for my opponents.

Sincerely,

UserF/Userfno1

Esth
06-18-2013, 12:43 PM
only the devs would have enough data to evaluate this. i would also support just removing hits below armor altogether.

Kletian999
06-18-2013, 01:08 PM
Regarding your point below. That's how chance works. When you start "enforcing probability" on a micro scale it isn't random anymore- it's manipulate-able (gamblers fallacy would become true). You can hit 60% shots just like you can miss the 90%. It's all fair in the cosmic sense

___________________________
My thoughts are maybe the random chance is working but at the wrong level, i.e. at the game universe level (all matches) or even an individual match level whereas I think it should be either at a team or individual unit level

UserF
06-18-2013, 02:18 PM
not enforcing, just ensuring the random events apply at a smaller scale. If a gambler were playing the odds of the entire casino instead of the table he would probably avoid it altogether...except those slots players ;)

roder
06-18-2013, 03:16 PM
yeah the chances of missing those 90/80 shots are pretty low. even lower if they happen in one match. even lower if they happen in multiple matches. So low, that maybe one person out of all of TBS population experiences it.

You-- might be that one person. You are the chosen one.

Aleonymous
06-18-2013, 03:39 PM
Hello UserF. I've never observed something too peculiar with miss-chance at my games, so I can't share your anguish.

As you say, I generally avoid chance-shots, but I've had 50-60% ones hit and 90% miss. Just yesterday, I scored a 3WP-loaded 90% shot and thought I secured the win, only to lose it later by missing a 80% and a 70% shots in pillage (1-in-16 chances).

What you are suggesting as "smaller scale" is quite hard to implement, in my opinion, and drastically changes the game mechanics. You're probably referring to something like "in a particular game, if I miss a 90%, please make sure the following nine 90% shots I take hit-home", right? However, what happens if you hit your first 90% shot? How are the chances-to-miss your next 90% modified? Well, the 90% should be reduced to something like 8/9=88.9%. If you hit again, then it becomes 7/8... and so on until... it becomes 1/2 on the ninth time. As you see, this distorts the ARM/STR mechanic that now has to keep track of how "lucky" you've been. Now, counting in the "luck" of the opponent, and making sure both players get their due makes this dramatically complicated.

UserF
06-18-2013, 04:10 PM
No, not please make sure I hit the next 9 shots if I miss one 90%. Please make sure that the random calculations are based on a single match and not something more global. If it is indeed "right" now then great, I just have really bad luck and will continue to be unhappy when I see the odds stacked against me in a game that is otherwise decided by strategy. Luck is luck, but its hard to swallow in a game like this.

UserF
06-18-2013, 04:12 PM
FYI, I'm not complaining about losing in general, just when it comes down to chance being the deciding factor. I have nothing against losing because my opponent simply outplayed me. Those that know me realize what kind of a player I am.

Kletian999
06-18-2013, 08:35 PM
No, not please make sure I hit the next 9 shots if I miss one 90%. Please make sure that the random calculations are based on a single match and not something more global. If it is indeed "right" now then great, I just have really bad luck and will continue to be unhappy when I see the odds stacked against me in a game that is otherwise decided by strategy. Luck is luck, but its hard to swallow in a game like this.

All computers use what's called an RNG to create random numbers. When the RNG generates a number (when anyone makes a chance shot) if it's higher than the percent to hit they miss. The RNG is not effected by other people's games, it not making a sequence and coincidentally giving you the bad ones. Keep a 10 sided die/dice with you to roll every time you make a percent shot, over 200 chance shots or so the results will be within a statistical margin of error (unless your die is broken).

raven2134
06-18-2013, 10:27 PM
Simple solution. Get a piece of paper or an excel file. Note the hit/miss and the percentage chance in a table format for a set sample size at least 15 but preferably 30+. Then conduct a statistical test to validate the deviation of the result and whether it's following expected probability. You'll have verified if this is true or not. We have done this on another thread and it all checked out.

UserF
06-19-2013, 12:30 PM
on it, doing it, so far 2 games collected (played more but no <100 attempts). So far stats support my claim but still early, need more samples.

1st - 2 70% attempt 1-1, 1 50% attempt 0-1 ( I would expect 2 hits with these odds, instead only hit 33%)
2nd - 1 90% 1-0, 1 80% 0-1 (odds better than 50% but that is what we got here)

bwoneill
06-19-2013, 05:38 PM
With so few data points it's impossible to make any meaningful conclusions. You can use one of the spreadsheets to generate a scatter plot like the one below.

TBSF_RNG.ods (https://dl.dropboxusercontent.com/u/1886355/TBSF_RNG.ods)
TBSF_RNG.xlsx (https://dl.dropboxusercontent.com/u/1886355/TBSF_RNG.xlsx)

Just fill in your data in the second and third columns. The picture below was generated using data that can be found in the following post.

17569

Note that the error bars have a 68% confidence level. That means you should interpret the blue points as indicating that, given the number of samples, 68% of the time you would expect your results to be within the error bars. The orange points should be interpreted as meaning that, given your observed hit rate, there is a 68% chance that the true hit rate is within the error bars.

http://s21.postimg.org/cnjmfm4br/TBSF_RNG.png

UserF
06-19-2013, 07:29 PM
ok, so if I read this example I think it supports my claim to a bit, high % hitting less then expected while lower % is hitting higher, hmm....looks like something is off. But yes, I will keep collecting data...I have a few more games stats now and I keep track every game...

good info! BTW 68% not great....

UserF
06-19-2013, 07:30 PM
if nothing else I appreciate the discussion on this!

bwoneill
06-19-2013, 10:09 PM
The 68% confidence level comes from the standard deviation (usually called sigma) of a normal distribution and is a standard measure of uncertainty in statistics. See 68-95-99.7 rule (http://en.wikipedia.org/wiki/68-95-99.7_rule). Unfortunately there aren't many samples in the data set. The largest percentage group has only 19 events while the smallest (40%) has only 5. If we had about 1000 events each, then we could use a 2 sigma 95% confidence interval.

I believe you have misinterpreted the plot. We can ignore the point at 40% since it only has 5 events (as a rule of thumb, you can't draw meaningful conclusions from less than 10 events). All of the rest of these points are within expectations. Two points are well within the error bars and the other 3 are as close as you can get with the low number of events.

What you perceive of as a lower probability at the high end is not statically significant. There is essentially a 50-50 chance that the observed probability will be above or below the expected (OK, it's skewed a little towards the center, but it's a small effect). When you look at it, it may look as if the high end is low. But if you think about it, would you really be surprised if you flipped a coin and it came up two heads and then 3 tails? That's essentially what the plot shows.

Now, if we had a few hundred events in each bin and they were still that much lower or if they were all 2 sigma or more below expectations, your conclusion might be correct.

Edit - I've added an additional column to the ODS (https://dl.dropboxusercontent.com/u/1886355/TBSF_RNG.ods) spreadsheet to calculate the chi-squared per degree of freedom (a measure of how well the data fits your expectations). A value of 1 means that the data fits the expected values with a typical amount of deviations. From the data in the above plot, I get 1.45 including the 40% point and 0.74 without it. Thus, the points above 40% show smaller than expected deviations.

netnazgul
06-21-2013, 07:39 AM
No, not please make sure I hit the next 9 shots if I miss one 90%. Please make sure that the random calculations are based on a single match and not something more global. If it is indeed "right" now then great, I just have really bad luck and will continue to be unhappy when I see the odds stacked against me in a game that is otherwise decided by strategy. Luck is luck, but its hard to swallow in a game like this.
Everything you say here and before that is based on your consideration that your %%-hits are based on previous hits of your particular currently striking unit / your team / current battle / all battles / etc. This precondition is totally wrong in case of actual random generators used in most systems (and of course in Factions) in which every generated number is close to completely independent from others, so you cannot presume you will hit 9 90% strikes if you missed one. Actually you can miss 10 out of 10 90% strikes in a row; that would be a ((1-0.9)^10)*100% = 0,00000001% chance, but it's still not zero.

UserF
06-28-2013, 09:53 AM
Agreed, and therein lies the problem as far as I am concerned.

So the RNG works as it is supposed to, I just don't agree that the results are what strategy folks consider palatable. I would love to see <100% chance hits removed or simply run a extremely large number of RNG calculations and then average them to give the result for a single shot so that the "odds" are more realistic per match since that is the granularity we care about, i.e. winning each match not figuring out what my average is over my entire career.

Aleonymous
06-28-2013, 12:11 PM
...or simply run a extremely large number of RNG calculations and then average them to give the result...

That is by definition identical to what is actually done! Example:



===================
RNG Seed -- Outcome // 90%-chance ==> Miss when Seed<0.1
===================
0.4371 -- Hit
0.1754 -- Hit
0.1383 -- Hit
0.2290 -- Hit
0.0312 -- Miss
0.6202 -- Hit
0.5683 -- Hit
0.6351 -- Hit
0.0231 -- Miss
0.9782 -- Hit
===================
In this random seed of just 10 samples, the hits were 80% instead of 90%. If I'd drawn more sample (I drew those from Matlab "rand" routine), it would have converged to 90%.

So, what are you proposing by saying "averaging" them? That we impose a rule like "Hits were more than the misses in this simulation, so the outcome is hit"? That violates the logic of it and (if I get it right) it means that everything >50% would hit, everything <50% would miss and everything =50% would hit/miss randomly.

Another (potentially unrelated) issue: Most pseudo-random RNGs typically have a "state" (e.g. a number) which they are initialized in. This means, that once they're set to a given state, they will give the SAME random number sequence, EVERY time. If, by any chance, you noticed that the first 90% hit attempted in a match misses systematically, that could mean that the RNG-state is not set correctly (i.e. it's not reset/randomized at each battle/day/computer). I seriously doubt that's the case here, because most RNG states are initialized by a pseudo-random number, for instance some operation on the digits of the current machine-time (e.g. 2013Y/06M/28D/19h/57m/35s).

raven2134
06-29-2013, 01:49 AM
Anyone see Rensei's thread? Let's avoid the silly "Randomize the Randomizer" course this is taking haha

Aleonymous
06-29-2013, 03:29 AM
"Randomize the Randomizer"

Referring to this, eh? :D

https://dl.dropboxusercontent.com/u/99633386/TBSF_fGpossd_vNOT.gif
(based on this (http://stoicstudio.com/forum/showthread.php?1675-Factions-Randomizer&p=22187&viewfull=1#post22187), Butters)