Whoa....time out folks - My proposal
-
Orlando the Axe
- Posts: 629
- Joined: Sat Mar 06, 2004 9:01 pm
- Location: CA
- Contact:
Thanks for the feedback also.igmo wrote:thanks for the overview CIK. i appreciate knowing the process, and respect that you all are trying to right the wrongs of 1.4.x
couple things i note:
1) the mention of a bunch of 1.4 stuff being ported in, then a note that the m2 portion was reverted back to 1.3 confuses me. you brought in all the whackyness of 1.4 only to remove it?
2) you leave yourself wiggle room for 1.5.1. i say polish 1.5.0 up till its ready, and commit to it as a last release (barring future os changes etc.) you can release a linx version as 1.5 too.
lastly, i think you would be the best person to address the lil rangetest portion of the other proposal thread. why do arch and mort have different downhill ranges in 1.5 than 1.3, if the gameplay is the same?
1)There was much good in 1.4 that wasn't in 1.3. We also wanted vTFL and all the bug fixes also.
2)I'm trying to balance the # of public betas with the desires of the team to keep polishing things. Linux is slated for 1.5.1 just because of the time frame. If there are no major game play bugs found in the 1.5.0 final then 1.5.1 will be a an optional upgrade for folks. Much like how 1.4.3 & 1.4.4 where compatible with each other. In fact if the community doesn't have any major issues with PB#2 we will go final soon after that. My rule is the last public beta should the same code used to build the final version, though sometimes I'll bend and allow a few minor very safe changes to be made. But only after much consideration. I'm also committed to keeping the development loop/public beta loop going until the community is happy with the product. We are just hoping that PB#2 will be the one.
Game play is not exactly the same, it contains a large # of bugs fixes. Many of them not even known to folks, or only know as the random OOS bug. There have been a large # of bugs fixed from PB#1, I suggest waiting and doing your testing /w PB#2 which “Will†come out early next week. We are just updating the readme's and making installers now.
-
iron
- Site Admin
- Posts: 2006
- Joined: Thu Feb 26, 2004 1:21 am
- Location: diving out of the Sun at 10 o'clock high!
- Contact:
igmo, I watched those films over & over again. As I've said in another thread, I think I need new glasses 
I really couldn't see a consistent difference in downhill ranges for archers in 1.3 and 1.5. Sometimes they take one step from your starting point, sometimes two, sometimes none, and that happens in both versions.
The cause is likely to be that the speed of the arrow isn't a fixed value, but rather a random range as defined in the tags (between 0.24 and 0.26). If the speed for a particular shot is chosen in the higher end of that range, it'll be able to fire it from slightly further away than if it were firing it a tad slower, if that makes sense.
Btw igmo, the film you posted on the bug tracker a few days ago - for some reason there's no link to it there. Could you mb host it somewhere like you did the others please?
I've just downloaded your venice test & will get back to you as soon as I can. Which may not be for a day or 3 - weekend & all...
Edited By iron on 1082768046
I really couldn't see a consistent difference in downhill ranges for archers in 1.3 and 1.5. Sometimes they take one step from your starting point, sometimes two, sometimes none, and that happens in both versions.The cause is likely to be that the speed of the arrow isn't a fixed value, but rather a random range as defined in the tags (between 0.24 and 0.26). If the speed for a particular shot is chosen in the higher end of that range, it'll be able to fire it from slightly further away than if it were firing it a tad slower, if that makes sense.
Btw igmo, the film you posted on the bug tracker a few days ago - for some reason there's no link to it there. Could you mb host it somewhere like you did the others please?
I've just downloaded your venice test & will get back to you as soon as I can. Which may not be for a day or 3 - weekend & all...
Edited By iron on 1082768046
the venice test is more objective, as you can stand right at the edge of the wall and use the tile pattern to guage distance.
i can appreciate that you cannot watch the film and know the range is different from 1.3 to 1.5. im sure u know that when you control click the unit always misses - sometimes long or short. what you need to do is repeat the same sort of test and you then will know just where you clicked and if the unit moved to get that distance into range. i can assure you there are small differences.
an ideal test would be on a map that had a 1 unit grid on it (and level and evenly sloped areas) - but im not really up for making that. i tried testing units on PG by making my line formation 32 long and 1 unit separation, so i could sorta have a ruler... but pg is not really flat, and its hard to keep the very same start position, etc.
CIK recommends repeating a test on 1.5b2, and i will.
i can appreciate that you cannot watch the film and know the range is different from 1.3 to 1.5. im sure u know that when you control click the unit always misses - sometimes long or short. what you need to do is repeat the same sort of test and you then will know just where you clicked and if the unit moved to get that distance into range. i can assure you there are small differences.
an ideal test would be on a map that had a 1 unit grid on it (and level and evenly sloped areas) - but im not really up for making that. i tried testing units on PG by making my line formation 32 long and 1 unit separation, so i could sorta have a ruler... but pg is not really flat, and its hard to keep the very same start position, etc.
CIK recommends repeating a test on 1.5b2, and i will.
actually, here is a better concept for a rangetesting map:
have an invincible target dummy in the center of 1 unit separated concentric circles (drawn on the mesh). place artil units u want to test at say the 50 unit line, then click to attack the dummy and see what ring they stop on when they fire.
the same map could be made level, with a 30% down slope to the dummy and with a 30% up slope. any map wiz wanna crank that out in 20 mins and confirm or deny the ongoing discussion about range? im sure i could do it, but it would take me hours.
have an invincible target dummy in the center of 1 unit separated concentric circles (drawn on the mesh). place artil units u want to test at say the 50 unit line, then click to attack the dummy and see what ring they stop on when they fire.
the same map could be made level, with a 30% down slope to the dummy and with a 30% up slope. any map wiz wanna crank that out in 20 mins and confirm or deny the ongoing discussion about range? im sure i could do it, but it would take me hours.
igmo, here's another suggestion for your test. As Iron stated, most Myth units use randoms settings. To really see if the engine is affecting anything, you'd need to make a special archer tag, or whatever, with static instead of random ranges. Then test this unit on both 1.3 and 1.5. The alternative is to average the results of a hundred tests, much as we did when trying to replicate TFL for vTFL the first time around. Honestly, one thing we learned when making these tests over a year ago is that perceptions of what's accurate can vary quite a bit.
P.S. Say hi to some of the other old timer PoOps for me.
P.S. Say hi to some of the other old timer PoOps for me.
ChrisP wrote:To really see if the engine is affecting anything, you'd need to make a special archer tag, or whatever, with static instead of random ranges.
that thought has occured to me... i was about to say that i thought it could ruin the test - as that random is (imo) being affected by other code in the 2 versions, so making the random fixed might make the problem go away.
on further thought, i think it would make any difference smaller, but the difference would remain.
otherwise, yeah, you do hte test a bunch of times and find the circle where the unit stops. the centers of the circles should not be the same, even if the circles themselves overlap.
now, how are you coming with that testing map? you can go ahead and populate it with de-randomed units if you want.
oh yeah, and PoOp has a pretty active forum if you wanna poke in and say hey... http://traininghall.com/forum/poop/
okay then
so i made a plugin as described a few posts up... it's available at: http://home.austin.rr.com/aoffice/rangeTest-map.zip
the plugin has 3 meshes where an arch, dorf, fetch, lock, mort and soul can attack an invincible target dummy on level, uphill and downhill terrain. the groundplane is marked with numbered concentric circles which are 1 myth unit apart. (meaning if you ran a 1 unit separated formation perpendicular to the circles, a unit would stand on each line.)
i then ran 5 tests for each condition on both versions 1.3 and 1.5. those films are available here: http://home.austin.rr.com/aoffice/rangeTest-films.zip
i recorded the distance each unit shot at, and a summary of that can be found here: http://home.austin.rr.com/aoffice/rangeTest.htm
on that chart the average for 1.3 is subtracted from the average for 1.5 to show the average deviation (in the blue field.) if the number is negative, 1.5 shoots shorter for that unit -- positive means 1.5 shoots longer.
in general, the fetch, lock and mort are pretty similar in the two versions. the arch, dorf and soul ranges differ more. the largest difference is .5 units - which is about one archer step. one thing to note is that if you were in an uphill/downhill archer battle, you'd need to add the appropriate deviations together to get an idea of the net affect. as an example 1.5 archers shoot .45 units shorter downhill, and .5 units longer uphill. the net affect would be very slight in terms of an advantage either way, yet would require you as a player to position your archers in a differnt place and thus would "feel" weird.
so, big whoop you say? those differences seem pretty damn small you say? ok, mb, but an assertion from projectMagma is the gameplay is the same. a frequent assertion from the players is that range seems different. well, the players are right - the ranges are different.
if the unit tags are unchanged, the physics or some other aspect is altering the end result.
so i made a plugin as described a few posts up... it's available at: http://home.austin.rr.com/aoffice/rangeTest-map.zip
the plugin has 3 meshes where an arch, dorf, fetch, lock, mort and soul can attack an invincible target dummy on level, uphill and downhill terrain. the groundplane is marked with numbered concentric circles which are 1 myth unit apart. (meaning if you ran a 1 unit separated formation perpendicular to the circles, a unit would stand on each line.)
i then ran 5 tests for each condition on both versions 1.3 and 1.5. those films are available here: http://home.austin.rr.com/aoffice/rangeTest-films.zip
i recorded the distance each unit shot at, and a summary of that can be found here: http://home.austin.rr.com/aoffice/rangeTest.htm
on that chart the average for 1.3 is subtracted from the average for 1.5 to show the average deviation (in the blue field.) if the number is negative, 1.5 shoots shorter for that unit -- positive means 1.5 shoots longer.
in general, the fetch, lock and mort are pretty similar in the two versions. the arch, dorf and soul ranges differ more. the largest difference is .5 units - which is about one archer step. one thing to note is that if you were in an uphill/downhill archer battle, you'd need to add the appropriate deviations together to get an idea of the net affect. as an example 1.5 archers shoot .45 units shorter downhill, and .5 units longer uphill. the net affect would be very slight in terms of an advantage either way, yet would require you as a player to position your archers in a differnt place and thus would "feel" weird.
so, big whoop you say? those differences seem pretty damn small you say? ok, mb, but an assertion from projectMagma is the gameplay is the same. a frequent assertion from the players is that range seems different. well, the players are right - the ranges are different.
if the unit tags are unchanged, the physics or some other aspect is altering the end result.
Great job igmo! You've gone above and beyond the normal endevours of public beta testers and really provided some useful information here.
The question I'd pose is what would be a significant deviation. Afterall, some deviation is to be expected when there is a random factor. Unfortunately it's bee a long time since I've taken any stats courses, and I don't have time to break out my books at the moment.
I'm sure someone will be able to answer this question for us.
The question I'd pose is what would be a significant deviation. Afterall, some deviation is to be expected when there is a random factor. Unfortunately it's bee a long time since I've taken any stats courses, and I don't have time to break out my books at the moment.
I'm sure someone will be able to answer this question for us.
yup, i'm not a statistician either.
my feeling is that the fetch and lock are definately close enough to be within deviation. and, yes, they are essentially unaffected by slope.
the archer seems indicative a true variation, the dwarves maybe, and the souless is a big who knows. souls seem to have pretty wide variation within a single test in 1.5 play (more so than in 1.3)
a bigger sampling would im sure be a help. i'll redo the test when 1.5b2 comes out, and hope someone else can find the time to run tests too.
my feeling is that the fetch and lock are definately close enough to be within deviation. and, yes, they are essentially unaffected by slope.
the archer seems indicative a true variation, the dwarves maybe, and the souless is a big who knows. souls seem to have pretty wide variation within a single test in 1.5 play (more so than in 1.3)
a bigger sampling would im sure be a help. i'll redo the test when 1.5b2 comes out, and hope someone else can find the time to run tests too.
Hehe, I'm glad you made the test mesh instead of insisting I do it. 
I looked at your results, but I don't think 5 tests per unit can prove anything. There are at least two, and possibly three random factors involved. The first is the attack velocity as I mentioned before.
Another variable that occured to me is attack accuracy. Archers have a fairly innacurate attack when unvetted, and this is probably a significant variable in your test. If you don't want to make monster tags with static instead of random values, one thing that will help make your test more controlled is to vet each unit to 5 kills prior to testing and measuring.
Finally, many projectile tags have a random velocity setting as well. Offhand, I don't remember if archer arrows or dwarf bottles use a random value, but if they do, then that also can skew your results.
Again, I recommend either using tags without random values or testing with at least 100 shots instead of just 5. And if you're concerned that the random generation may have been changed, and do the test both ways, and the 100 shot test shows significant deviation while the non-random test does not, then we'll know something is wrong with random generation.
Anyhow, keep up the good work. Oh, and I tried to use the PoOp forum a couple months ago, but you have to register for the damn thing, and I'm just too lazy.
I looked at your results, but I don't think 5 tests per unit can prove anything. There are at least two, and possibly three random factors involved. The first is the attack velocity as I mentioned before.
Another variable that occured to me is attack accuracy. Archers have a fairly innacurate attack when unvetted, and this is probably a significant variable in your test. If you don't want to make monster tags with static instead of random values, one thing that will help make your test more controlled is to vet each unit to 5 kills prior to testing and measuring.
Finally, many projectile tags have a random velocity setting as well. Offhand, I don't remember if archer arrows or dwarf bottles use a random value, but if they do, then that also can skew your results.
Again, I recommend either using tags without random values or testing with at least 100 shots instead of just 5. And if you're concerned that the random generation may have been changed, and do the test both ways, and the 100 shot test shows significant deviation while the non-random test does not, then we'll know something is wrong with random generation.
Anyhow, keep up the good work. Oh, and I tried to use the PoOp forum a couple months ago, but you have to register for the damn thing, and I'm just too lazy.
Hey Guys - this has inspired me to check something similar we have noticed in playing almost 1,000 Rocket Dorf Fest games over the past year.
The "dud" and "rebound" of the standard Rocket Dorf is similar to that of a Dorf but not the same. What we notice a lot, however, is that "duds" (the projectile "promotion" rather than the non-media "detonation") seem to come in pairs (dud-dud) more often than you would statistically expect. I keep thinking it has to do with the pseudo-random number generator used - or perhaps the granularity of the promotion percentage (I've always wondered if the smallest increment is really 0.033).
I had hoped to make a test map for this with an invincible unit on a map that I could just target with a Rocket Dorf from afar, so I could just click on it and then start taking notes as to Hit, Rebound, Dud. This would also be good because the recharge time would be consistent. I'd do it for 100 shots.
Now that I've made a 1.5-compatible version of RDF 4, I could also have it with films if needed. I'll see if I can whip out a test this weekend if at all possible. This "double-dud" effect just seems more noticable than you would expect.
The "promotion on detonation fraction" for the one example I would use is 0.065 - in my mind this is 6.5%, so 6.5 out of 100 shots should be duds. The odds of two of these in a row happening as often as we seem to see them appears odd - but then, it could be that we are shooting 10,000 shots in a typical 6 hour Match!
I'll see if I can't set this up during the weekend.
Hey Doob! Did you really go to the Galapagos? If so, how was it!
[ P.S. My baak@orderofhpak.com e-mail is off-line until the host gets my account reset after they've switched servers ]
The "dud" and "rebound" of the standard Rocket Dorf is similar to that of a Dorf but not the same. What we notice a lot, however, is that "duds" (the projectile "promotion" rather than the non-media "detonation") seem to come in pairs (dud-dud) more often than you would statistically expect. I keep thinking it has to do with the pseudo-random number generator used - or perhaps the granularity of the promotion percentage (I've always wondered if the smallest increment is really 0.033).
I had hoped to make a test map for this with an invincible unit on a map that I could just target with a Rocket Dorf from afar, so I could just click on it and then start taking notes as to Hit, Rebound, Dud. This would also be good because the recharge time would be consistent. I'd do it for 100 shots.
Now that I've made a 1.5-compatible version of RDF 4, I could also have it with films if needed. I'll see if I can whip out a test this weekend if at all possible. This "double-dud" effect just seems more noticable than you would expect.
The "promotion on detonation fraction" for the one example I would use is 0.065 - in my mind this is 6.5%, so 6.5 out of 100 shots should be duds. The odds of two of these in a row happening as often as we seem to see them appears odd - but then, it could be that we are shooting 10,000 shots in a typical 6 hour Match!
I'll see if I can't set this up during the weekend.
Hey Doob! Did you really go to the Galapagos? If so, how was it!
[ P.S. My baak@orderofhpak.com e-mail is off-line until the host gets my account reset after they've switched servers ]
igmo wrote:yup, i'm not a statistician either.
I am a bit of an amateur statistician and if someone would like to provide some data points, I would be glad to put in some analysis time.
Ideally I am going to need about 30 data points per unit per version per test type (i.e. distances of 30 archer shots in 1.3 and 30 shots in 1.5b1 for uphill, repeat for downhill).
... Sorry to reply to my own post, but I decided to see how fast I could whip out this test!
It's actually running right now - lol - I built a test using an invincible unit from my StoneHeart RDF map (a large fence post used in the Cage mesh), made it a hunting unit and set an RDF "Mini" at work firing at it. I put it on the center flat area and he is on the flat area as well.
He'll just keep firing at it for the next 10 minutes - lol!
Then I'll watch the film (at 2x speed) and we will see what the results are. This is currently for 1.4.3 and I will do it for 1.5 B1 next. I may not get the results posted here until tomorrow though - but I will post it then.
It's actually running right now - lol - I built a test using an invincible unit from my StoneHeart RDF map (a large fence post used in the Cage mesh), made it a hunting unit and set an RDF "Mini" at work firing at it. I put it on the center flat area and he is on the flat area as well.
He'll just keep firing at it for the next 10 minutes - lol!
Then I'll watch the film (at 2x speed) and we will see what the results are. This is currently for 1.4.3 and I will do it for 1.5 B1 next. I may not get the results posted here until tomorrow though - but I will post it then.