Talk:Psionics
Just had a thought on the formula!
What if the "Psionic Combat Strength" formula was a ' + ' rather than a ' * ' ?
It would mean the example Soldier would have a PCS of 111, rather than 1520, and the defending Muton would have:
Panic Chance = 76% MC Chance = 56%
These numbers seem more reasonable, don't you think?
--Danial 15:32, 19 Nov 2005 (PST)
Right, it sounds much more reasonable now. My actual results were 1192 successes out of 2420 tries = 49.26% success rate. The MCer was right next to MC victim, and we know distance introduces some kind of decrease to base chance. That's the good news. The bad news is that I also tested psi str 95, skill 44 (and some higher skill levels) and never failed to MC once with these better MCers. Using your revision, a psi str 95, skill 44 guy should have had a success rate of 84% vs. a beginner Muton soldier (psi str 25, skill 0), not 100%. My higher-skill guys were also right next to their MC victims. So, to me the first example makes sense relative to your idea (fits base chance, plus a little distance adds a little more), but the second example counters it.
Still, it sounds WAY better than 1500% :)
I sure wish psi testing wasn't so labor intensive. I got those 2400 counts doing all that experience testing... not interested in doing it again, just for psi equations. At least not at the moment. :P ---MikeTheRed 21:12, 23 Nov 2005 (PST)
Question: why is the Sectopod's psionic resistance listed as N/A? I've mind-controlled sectopods (or at least one square of a sectopod) on many occasions. I usually have it shoot itself, since it can target one of its other, non-MC'ed squares. From my anecdotal experience, their psi resistance seems somewhat high. --Papa Legba 14:19, 2 February 2006 (PST)
Answer: If you look on the alien summaries in the wiki they all have a PSI resistance listed, except the Sectopod. Since it's not listed anyplace I put N/A. Feel free to put it where it belongs in the list if you have enough experience to feel comfortable ranking it. --Darksun
Darksun, I took the liberty of signing your name for you, and making a separator line. Why not say a word about yourself under User:Darksun and also, use four dashes (----) to demarcate entries. For more on the basics of wiki'ing, see NKF's "Community Portal" at the bottom of the homepage. Past all that - welcome, fellow XCOMMIE! We were all wiki noobs once. But not everybody loves XCOM. You must. Welcome aboard!
Proceeding right along to flagellating, laugh - Where are you folks talking about Sectopod psi resistance being "N/A"? I see their Psi Strength as 100, and their Psi Skill as 0 at Beginner; Psi Strenth 116 and Skill still 0 at Superhuman. In this respect, they match Cyberdisks. What do you folks mean by Psi "resistance"? Do you know the math behind the psi equations? I haven't been able to figure it out. But the stats I just listed are ones hacked out of the game... Zombie is going to present a full precis' some time soon.
Great speaking at ya ... keep on contributing, Darksun!! --MikeTheRed
Alien psionic resistance
I haven't tried mind-controlling Cyberdiscs much, but if, as I've read elsewhere, both Sectopods and Cyberdiscs have have a Psi Strength of 100, they should probably both be classed "extremely high". --Ethereal Cereal 10:30, 4 May 2006 (PDT)
Psionic testing
I started doing some testing to figure out the psionics formula.
Initial finding: distance takes the hypotenuse into account. I'm not sure if the actual Pythagorean formula is used, but I can say this much: an Ethereal was able to panic a moderately-psi-weak soldier at 30 squares along the horizontal but not at 30 squares along the diagonal.
Second finding: the difference in "difficulty" between MC and panic is definitely 20 points (although that may not be percent). A soldier with 97 Psi strength, 0 Psi skill standing adjacent to an Ethereal Leader (60 str/45 skill) could not be panicked at all, but 96 Psi str could (often). The same soldier with 77 Psi str couldn't be mind-controlled, but could at 76 Psi str. With further testing, I see that 60/45 vs. 97 should succeed occasionally; I might retest this for greater precision, although the +20 still seems to hold.
Third finding: A soldier with 99 Psi str/0 skill could just barely be panicked by an adjacent Ethereal with 63/45. I couldn't count how often the Ethereal succeeded, but it was something like 1%-2% of the time. The same soldier was panicked just as infrequently by an Ethereal with 45/63 (str & skill reversed), and much more often by an Ethereal with 54/54. This pretty strongly suggests the attack portion of the formula involves psi str & skill multiplied together.
Data points collected so far:
- Ethereal, 63/45, just barely panicked soldier, 99/0
- Ethereal 40/40 just barely panicked soldier 75/0
- Ethereal 27/27 (multiplied together = 729) just barely panicked soldier 57/0
- Ethereal 10/73 (multiplied = 730) just barely panicked soldier 57/0
- this confirms that attacking is based on str*skill
- Ethereal 18/18 just barely panicked soldier 49/0
- Ethereal 1/1 just barely panicked soldier 43/0
- Ethereal 1/1 just barely panicked soldier 1/210
- this confirms Psi skill / 5 is used in calculating defense; 1 + (210/5) = 43
I also tested Ethereal 0/1 (no psi str, 1 pt. skill); it still attacked, but did not attack at 0 pts. skill no matter how high psi str was (no surprise there, otherwise all units would do psi attacks).
I ran the points through a curve-fitting program and got a nice and simple answer:
attack strength = psi str * psi skill / 50 defense strength = psi str + (psi skill / 5)
Panic Attack chance = 44% + attack strength - defense strength Mind Control Attack chance = 24% + attack strength - defense strength
The formula has held up under limited further testing; I'm satisfied it's correct. I invite others to test it more than I have.
The impact of distance has yet to be tested, but it should be easy to figure out now.
--Ethereal Cereal 22:03, 26 May 2006 (PDT)
There is a reasonable approximation to the Pythagorean formula (in the plane) that could have been used, that doesn't require loops or floating-point math:
max( | Δx | , | Δy | ) + ½min( | Δx | , | Δy | )
where Δ_ := _2 - _1.
It overestimates slightly. I haven't thought through how this would generalize to 3D. [Yanked from Moria, Angband, and variants...think it's too basic to copyright, though.]
One way to test whether the distance estimator is no greater than true Pythagorean formula by working out the greatest distance (pure x or pure y) that MC/Panic is not 0%, then testing on a diagonal with a "computed same". An integer-math overestimator (like the above) would by 0%.
Muliple of 5 allows replacing the diagonal with a suitably scaled 3-4-5 triangle...should work as well for testing.
--Zaimoni 2:38PM, 26 May 2006 (CDT)
Wow... good approach to teasing this out, Ethereal!
Am I doing this right? I plugged in numbers for Muton, 25/0, Me 95/44, and got 83% MC Base Chance... But in much repeated testing (when learning about experience counters) it seemed quite solid that I was MC'ing him about half the time. Say 45-55% of the time. This was when right next to each other. Is 83% right for your equations?
Zaimoni, re: your equation, see Explosions#Distance_from_Ground_Zero. XCOM seems to use a fairly simple approach that doesn't involve Pythagoras per se. Since it's based on a very old engine, much of its stuff is integer based. Cool deal on the deltas and other wiki symbols - I wish I knew about them sooner! Actually I know where wiki symbol tables are... just didn't bother to look up so many symbols. :P
Nobody has tackled 3D distance yet, but I may soon, in studying illumination.
Once the equations are well known (maybe they are already), I imagine some cool 3D topographical graphs of the various factors (skill and strength, you vs. enemy) could be made. Do either of you have surface graph s/w at hand? All I've got at the moment is Excel. :(
Great work! Also thanks for cleaning up the page in general, Eth.
---MikeTheRed 22:24, 2 June 2006 (PDT)
I suspect you've misremembered the details of your tests. The 49.26% success rate out of 2400 trials that you quote above would have been for your 95/16 soldier doing panics, not MCs: 44 + [95*16/50 = 30.4] - 25 = 49.4%, almost exactly the same as your 49.26%. A 95/44 doing panics would always succeed: 102.6%. A 95/44 doing MCs would succeed 82.6% of the time.
--Ethereal Cereal 11:06, 3 June 2006 (PDT)
Ah, there's that data. Thanks for reminding me. I can't remember what I put in which discussions and was going to dig it out of my db.
No, it was MCs. Indeed I relied on the lack of seeing an enemy, if I got moving too fast and couldn't remember if I successfully MC'd. Attention starts to wander with such highly repetitive testing. This was before I realized the following...
Anyone doing repetitive psi testing can readily and accurately get the number of positive results from the UNITREF.DAT Psi byte [84] if you work it right. Note the turn you start psi attempts. Then do exactly 2 or 3 psi Attempts each turn, as your TUs allow. Then note the turn you stop and get the delta. Turns x Attempts/Turn = Attempts. A failed psi attempt adds 1 to [84] and a successful attempt adds 3, so the number of successes is: ([84]-Attempts)/2. Be careful during testing marathons because the byte wraps around at 255 (=85 straight successes, or 28.3 turns at 3 successful attempts per turn). Then it starts counting again so actually even this can be dealt with by adding 255 if you're on top of it. If anybody wants an applet that shows Unitref experience values on the fly, let me know. It's great for building experience. Right now it's embedded within my mdb though... I'd have to tease it out.
I'm using XCOM DOS which I once ran XcomUtil on, if that matters.
Hmm.
---MikeTheRed 12:21, 3 June 2006 (PDT)
Distance...ok, no reason for the game to use multiple 2-d distance formulæ.
I have software set up to render 3D graphs. Is there anything on the site that jumps out as "would like to see first"? I'd rather wait until the psi formulæ are calibrated first before graphing those.
Ethereal: What I see posted is test data for ethereal MC human. Do you have test data for human MC ethereal as well? [That would test whether the game calculations are symmetrical.]
---Zaimoni 2:37, 3 June 2006 (EDT)
It is possible X-COM units get a different formula from aliens -- I only tested an alien doing psi, as it was more automated -- the thing would psi without my having to control it.
I'm not not volunteering to test x-com units for the time being -- it's a bit more laborious -- but if I had to test it, I'd do a small number of trials of 35/16, 50/51, and 100/51 soldiers trying to MC (not panic) an adjacent Muton, which should succeed 10.2%, 50% and 101% of the time.
Feel free to test it before I do. ;-P
Oh, and nice work on the flares, Mike.
--Ethereal Cereal 12:50, 3 June 2006 (PDT)
Sounds cool, Zaimoni... As for "like to see",
Once my troops have psi ability, I try to get 90+ psi strength guys. Then the question becomes "at psi str 90, at what psi skill point are most/all aliens certain to be MC'ed?" In my experience, it's somewhere around skill 40-60... Anything that might show this, or show how you fare relative to aliens at, say, STR 90, is something I've always wanted. However it might be done. In one sense you could simply plug the "worst" psi alien (Ethereal CDR?) into Eth's equations to get the final answer. But it'd be good to see the lay of the land.
If you make the simplifying assumption that many of your targets are psi skill 0, there's one less variable. Additional graphs can be done later just for the psi-capable aliens... first things first though...
Taking the above into consideration, then, XCOM Psi Skill (Str=90) vs. Alien Psi Strength (Skill=0) vs. Success rate is one 3D graph that could be made. If you can do a 4th dimension as color, you could replace the Success rate axis with XCOM Psi Str, then have color indicate success rate, from primary red for "no way" through yellow for "sometimes" to primary green for "always". Or something like that. It'd also be nice but not critical to have annotations for where various aliens lie along their dimension. I can supply this info if you want. Which reminds me: Eth, Zombie and others found some errors in Aztec's table (which came from the OSG). Just a few percent of them, but they're definitely there (I can't remember where). I've hacked alien stats straight out of GEOSCAPE... see the alien stats page. I believe Zombie has found these to be correct, but he has not yet given the final word.
Zai, as far as I'm concerned, you can model using EC's equation already, if you have the time... I'm sure his equations are very close, if not spot on, and even if not, it may show key places to test and otherwise support further testing. E.g., both aliens and XCOM can readily have their relevant values hacked to test for boundary conditions of 0% and 100% success. Plus once it's set up it shouldn't be hard to tweak the equation and voila, the graphs update. But if you don't have the time - eh, it's all volunteer work anyway.
Eth, I suppose it's possible aliens vs. humans is different. They are known to have some differences in functionality, such as night vision. I don't know if I'll be able to test, but it definitely intrigues me. BTW I said somewhere else that I was looking into automating the XCOM interface in order to do psi testing. I have concluded that I can't do it. At least not the way I was hoping to. Oh well. Still, my belated realization that one can use Unitref[84] to count psi successes is a real boon to repetitive testing. So we'll see. Thanks for suggestions on what to test if I do. By the way, did I do that 83% correct? I just want to make sure I'm doing the math right.
Thanks re: the flares!
---MikeTheRed 21:08, 3 June 2006 (PDT)
The influence of distance still needs to be tested, although it doesn't seem to be an enormous factor. Rather than a complicated graph, I think a few charts with distance = 1, 10 and 20 would probably tell the story more clearly. The distance formula might even prove to be something really simple, like +1 distance = -1% chance.
The hardest aliens to MC are Sectopods and Cyberdiscs: look at the Psionics#Summary_of_Alien_Psionic_Resistance section I added. At distance 1 on Superhuman you'll need 93 to 192 Attack Strength to achieve 1% to 100% chance. Assuming Psi Str 90, that's Skill 52 to 107.
Now vs. the strongest attacker: Superhuman Ethereal Commander, attack str 88. You'll need a Resistance of 132 to be totally immune to panics from one adjacent to you -- that's str 100, skill 160, or str 90, skill 210. Against MC, it's str 100/skill 60 or str 90/skill 110. In practice you won't be right next to them, so you can probably go a bit lower. They'll focus on the weakest soldiers anyhow, which in practice means a couple of decoys in with some supertroopers is a good strategy.
82.6%, yes.
--Ethereal Cereal 21:51, 3 June 2006 (PDT)
OpenOffice.org is CPU-bound when converting the graph from 2D to 3D. [XCOM Psi Str/Sk 90/0-107, alien Psi Str/Sk 25-116/0; I'm on a 1GHz Pentium III/512MB] Not sure how ColdFusion single-IP demo would do yet (would have to reinstall), but the UI options look better in OpenOffice.org.
I'll attempt the transition later today (when I know I'll be away from the system and positively not working). I just need to be able to take a screenshot.
You probably already noticed this: even if a bug permitted a 0 Psi Skill operative to use a Psi Amp, the success chance of MC is still negative.
Also (thinking about the testing anomaly MikeTheRed had): The floating-point change in success rate from an increase of 1 in Psi Skill is 1.9 at 95 Psi Str. Integer truncation would make this 1...leaving the floating-point formula overestimating by 39.6%. At the precision of reporting, this works.
[Edit: but does *not* work against the initial regression. So, no good.]
---Zaimoni 9:51, 4 June 2006 (CDT)