Automated UFO Trials

[Bomb Bloke]

Are you adding the new results to the old and sampling them as a group, or processing them seperately?

[MikeTheRed]

I'm putting/appending all fire-log data to one table. But I'm including identifiers so I can easily separate out different nights' data (and even each shot result's order in the data). So I could easily see that the original group of FFA 0 data was very different from the new one.

 

All I can think of at this point is to do another FFA 0 run, while checking carefully that everything is set up right (the soldier's FFA really is 0, etc.). And then see how it compares to the previous data.

 

Since it's probably impossible to have less than 0 accuracy - and by extension, you can't do worse than that (i.e., have more lost shots) - the second FFA 0 group is suspect at this point (perhaps it wasn't really set to FFA 0). I try to be real careful, what can I say.

 

Or maybe something else way funky happened.

[Zombie]

Since it's probably impossible to have less than 0 accuracy - and by extension, you can't do worse than that (i.e., have more lost shots) - the second FFA 0 group is suspect at this point (perhaps it wasn't really set to FFA 0). I try to be real careful, what can I say.

Well, you can do a little bit of creative editing: say weapon accuracy to 255% and then crank up soldier firing accuracy to 255% and get the overall base chance to hit down to -5%. Yeah, base chance to hit can also go above 300% when fooling around with weapon and soldier accuracies. This might help out or it may not, it depends on how the game handles negative numbers in the battlescape.

 

- Zombie

[MikeTheRed]

Wow, I didn't know that about the Accuracy Formula, Z... I took a shot at it (so to speak) and indeed, 255/255 gives -5%. I can see how setting the highest bit would cause a negative, but otherwise don't see how they got a 5 out of that. Weird.

 

For now, I just want to reproduce (or not) what happened with FFA 0. I did more math and thinking after my posting above...

 

For one thing, I found that the second FFA 0 dataset hit the target 1.9% of the time. Compare how the first FFA 0 dataset hit it 2.2% of the time, and all other (higher) FFA datasets hit pretty much in line with their FFA (FFA 10 hit 9.6% of the time, etc.). IOW, the Hits percent supports that I was using the right FFA (i.e., 0)...

 

There were a couple of subtle differences between the two groups. One was that in the first FFA 0 sample (the very first time I used BB's logger), Weapon Accuracy (WA) was set to 65 for Snap (the standard setting) and Soldier Accuracy was 1, whereas for the second dataset, WA was 99 and SA was 1. I checked in-game and both said FFA 0. The other difference is that the first set had standard laser rifle strength, and the second had strength at 255.

 

FWIW additional testing (not yet presented) showed that at WA 100 and SA 10 (FFA 10), 1.0% of shots were lost, and at WA 100, SA 25, 0.2% were lost. These are somewhat consistent (regression line) with the 1.9% lost at WA 99 and SA 1 (FFA 0)...

 

On reflection then, I am strongly curious as to whether the two accuracy values (WA & SA) still affect where Missed shots go, even if only FFA governs how many "guaranteed hits" there are. Or maybe only Weapon accuracy. IOW, maybe the big difference between what happened with FFA 0 datasets 1 and 2 was that WA was different. WA 65 is considerably lower than 99, and many more shots were lost off of the map.

 

It can easily work that way and still be consistent, IF indeed "guaranteed hits" (per FFA) are treated entirely different from misses. It's all consistent with my simple model, even... just an aspect I had assumed worked just like FFA, but never actually tested. Maybe I already have some data on it. Let me go get some more data.

[Zombie]

There were a couple of subtle differences between the two groups. One was that in the first FFA 0 sample (the very first time I used BB's logger), Weapon Accuracy (WA) was set to 65 for Snap (the standard setting) and Soldier Accuracy was 1, whereas for the second dataset, WA was 99 and SA was 1. I checked in-game and both said FFA 0. The other difference is that the first set had standard laser rifle strength, and the second had strength at 255.

Forgive me for asking, but why on earth would you start testing on a theoretical 0% FFA with SA and WA not equal to 0 themselves? This would be the only case where I would trust that the base chance to hit was indeed 0%.

 

- Zombie

[MikeTheRed]

Forgive me for asking, but why on earth would you start testing on a theoretical 0% FFA with SA and WA not equal to 0 themselves? This would be the only case where I would trust that the base chance to hit was indeed 0%.

Eh, I was just ditzing WA below my usual 100 setting (for other results), knowing that it'd round to a FFA of 0. Prior to this, I can't recall anyone suggesting anything other than that FFA alone governs hit and miss results.

[Bomb Bloke]

There is a possibility that the figure used for the practical calculations is not rounded even if the displayed figure is. Whether this would account for such a large gap, I dunno. But it's worth keeping in mind.

 

Regardless, it seems to me that there should also be a deviation in horizontal angles of fire if it really is an accuracy issue of some sort. For it to be a recording issue would be a lot more complex... All sorts of possibilities there. The logger runs the same code with every repitition, and none of it is random. External system circumstances, on the other hand, are...

 

Hence my original recommendation was to simply remove all "misfire" results from the records, as there was no way of telling which ones really were misfires and which were simply high shots until a roof was stuck on the map (which I didn't do originally because I was lazy). So I suppose now is a good time to do that. Shove the new version in your GAME_98 folder, but be warned that you'll probably have to start all your sampling over from scratch (as the new angles of fire that can be recorded will be competing with your original results percentage-wise).

 

Now for there to be a misfire, the bullet must go backwards, or not at all.

Map.zip

[Shortstop4313]

There is a possibility that the figure used for the practical calculations is not rounded even if the displayed figure is. Whether this would account for such a large gap, I dunno. But it's worth keeping in mind.

 

Regardless, it seems to me that there should also be a deviation in horizontal angles of fire if it really is an accuracy issue of some sort. For it to be a recording issue would be a lot more complex... All sorts of possibilities there. The logger runs the same code with every repitition, and none of it is random. External system circumstances, on the other hand, are...

 

Hence my original recommendation was to simply remove all "misfire" results from the records, as there was no way of telling which ones really were misfires and which were simply high shots until a roof was stuck on the map (which I didn't do originally because I was lazy). So I suppose now is a good time to do that. Shove the new version in your GAME_98 folder, but be warned that you'll probably have to start all your sampling over from scratch (as the new angles of fire that can be recorded will be competing with your original results percentage-wise).

 

Now for there to be a misfire, the bullet must go backwards, or not at all.

 

Great reading here, guys... Very interesting reverse engineering.

[MikeTheRed]

Thanks for that post, ShortStop... you brought the thread to my attention again...

 

BB, I can't believe it but somehow I didn't catch your previous post. So I have been collecting loads of data over the past few weeks, using the old map. I have a lot of combinations of soldier accuracy and weapon accuracy. I'm sure it's still worthwhile in and of itself... especially given how much time it takes to collect a lot, when one is running their PC overnight. (Calendar time, not personal time.)

 

But I'm having trouble finding time to analyze this data. (Christmas games ) If anybody wants to take a whack at it, I can send it to you. It's in MS Access atm.

[Shortstop4313]

Another interesting question to try to deduce the answer for...

 

Does crouching actually reduce your chance of getting hit?

[zaimoni]

Another interesting question to try to deduce the answer for...

 

Does crouching actually reduce your chance of getting hit?

I already tested that, in slow motion.

 

Crouching reduces the chance of being hit by calculated misses. So it's of more use for inaccurate aliens. Calculated hits pretty much unswervingly head for the centroid of the target.

 

Reportedly, crouching also makes Blaster Bombs miss you if they're traveling horizontally and they directly target your "square".

[BladeFireLight]

Has anyone tested the effects of smoke on the aliens chance to hit or shoot?

 

-Blade FireLight

[MikeTheRed]

Not that I know of... the work done above is the first I'm aware of. It's for a simple one to one face-off, and only for the X-COM unit's stats while shooting an alien (not vice versa), on a clear level unobstructed board. And it's taking a while.

 

Good to see ya!