Talk:Parseable Area Statistics

The values for the degrassi knoll are inconsistent -- they add to 109.3%. (There are eight 9.4% encounters making 75.2%; the gearhead is given as 14.1% and the two non-combats 10% a piece). What's up with that?--DirkDiggler 20:37, 10 October 2006 (CDT)

Some testing suggests very strongly that the Toga party should be at 5% - 113 adventures only produced 5 Toga parties - that's well outside the 80% confidence for that sample size and the listed value -- the other ones are in line. NateTG November 7 2006.

• Try a larger sample size first; Yiab is usually very good about this.--Dehstil (t|c) 23:05, 6 November 2006 (CST)

• I spent a total of 35,543 adventures at Degrassi Knoll and, in the end, Gnollish toga party made up 3,994 of those. Of course, these adventures include both those before removal of the gnollish sorceress and bodybuilder and those after their removal, but since I don't believe the combat chance changed during that modification, these numbers are still likely applicable (in either case the listed combat chance was calculated only using post-change numbers). Basically, to get the 10% figure, I rounded off the combat chance at degrassi knoll to 80%, then noted that the two noncombat adventures are equally likely (3,994 gnollish toga parties and 3903 gnollish workshops), so I divided the 20% chance of noncombat adventure in two for each of them. Of course, if I rounded in the wrong direction (i.e. if the combat chance is actually 75%), the chance for each noncombat encounter would go up to 12.5%, which is further away from your result, NateTG. If people desire, I can write up tables for the encounter rates in each area like I did for the drop rates in monster talk pages, though I thought these numbers were relatively self-explanatory from the collected data of the monster talk pages.--Yiab 01:09, 7 November 2006 (CST)

The numbers that are currently on the page for the knoll are clearly incorrect since they add up to 110%, and for 35,000 adventures, the 95% confidence interval is roughly plus minus half a percent assuming the adventures are independant, so the 10% error is well outside any reasonable rounding errors. In fact, your numbers strongly call the hypothesis of 20 or 25% combat frequecy into the question. Extrapolating from the existing pattern of fractionalitysuggests that the non-combats occur at 1/9 of the time each (11.1%), the Gearhead occurs 16/90 (17.7%) of the time, and the remaining combats occur 8.8% of the time each. For 35543 trials, that would lead to expectations of: 3949 plus-minus 177 for the noncombats 6318 plus-minus 177 for the gearhead 3159 plus-minus 177 for the other combats How well does that match your test results? --NateTG 13:32 PST, 7 November 2006

I've looked at Yiab's data, and there is a pretty clear break in the non-combat rate. Using the last ~6000 adventures from his data set, I've come up with the following rates:

• 75% base combat rate
• 18.75% Gnollish Gearhead
• 9.375% Other Combats (No Knollish Piebaker?!)
• 12.5% non-combats

Is there a 'standard' sample size?

--NateTG 16:41, 8 November 2006 (CST)

Unfortunately, there seems to have been some error in transmission of my data here. In addition, I calculate the percentages that go to this page by hand after seeing a summary of the data on my screen, so any errors are likely to have been introduced by my slight tendancy to dyslex. Regardless, here are the total numbers of encounters I recorded during my last recording run from Aug. 24, 2006 to Aug 30, 2006.
Gnollish Crossdresser: 1312
Gnollish Flyslayer: 1311
Gnollish Gearhead: 2000
Gnollish Piebaker: 1328
Gnollish Plungermaster: 1302
Gnollish Tirejuggler: 1281
Gnollish War Chef: 1328
Guard Bugbear: 1279
one-eyed Gnoll: 1313
Gnollish Toga Party: 1526
Gnollish Workshop: 1556
Please, feel free to correct any calculation errors I've made.--Yiab 18:14, 8 November 2006 (CST)

• I've now corrected the problem of the numbers not adding up to 100%. I thought I'd corrected it after DirkDiggler pointed it out last month, but it appears to have been the exact error he mentioned.--Yiab 18:20, 8 November 2006 (CST)

The Haunted Pantry

The Haunted Pantry rates are all jaked up:

 the haunted pantry:fiendish can of asparagus=28.6|
   possessed can of tomatoes=28.6|undead elbow macaroni=28.6|14.3|
   the baker's dilemma=5|oh no, hobo=5|the singing tree=5|tresspasser=5||

there is an ill-formed segment following 'undead elbow macaroni' and the meat blobs and 'Cut Down in his Prime' are missing. Also, the percents sum to more than 100.--DirkDiggler 22:08, 30 January 2007 (CST)

Cut Down in His Prime appears to be once-per-ascension or once-per-day (I got one per character during a single day of adventuring) and it seems that I somehow borked up putting the meat blob's name in. I think that the adding up to more than 100% is that when I was calculating chances of combat encounters I forgot to multiply by the combat chance... oops.--Yiab 14:57, 31 January 2007 (CST)

Mine

The data for the mine lumps both types of 7-foot dwarves in together; this doesn't matter all that much but is probably worth mentioning. --Starwed 10:18, 3 March 2007 (CST)

Errors

For some reason I'm unable to edit this page (and as far as I can tell it's just this one) so I'll post the desert's chances here for the time being: the arid, extra-dry desert (ultrahydrated):cactuary=20|giant giant giant centipede=20|swarm of fire ants=20|plaque of locusts=20|rock scorpion=20|| --Yiab 19:26, 26 July 2007 (CDT)

Several zones invalidated by NS13?

I've just noticed that for the last set of lab data, which has been collected Post-NS13, the rate of Mad Scientists has gone up to around 71% (consistently for all 3 datasets of this era). All the datasets that I've checked for zones with unequal rates which were taken before NS13 match the rate given here extremely closely. Thus I am quite sure that the encounter rate for the lab has changed and assume the same is true for all zones with unequal rates. Unfortunately there seems to be no second data point to confirm or reject this theory atm. (Though there is some spading of HitS rates going on for On-The-Trail-spading AFAIK) --O 3 141592 16:17, 28 February 2008 (CST)

What dataset are you talking about?.  :) --Starwed 16:49, 28 February 2008 (CST)

Whoops, should have mentioned that. ^^° I'm talking about the encounter numbers on the monster talk pages. The data not fitting is from 14-Aug-07 on Talk:Knob Goblin Mad Scientist and Talk:Knob Goblin Very Mad Scientist. --O 3 141592 00:45, 29 February 2008 (CST)

It certainly could be the case that these numbers have changed. I'm fairly certain that it was 2/3 vs 1/3 early on, but that's only because Jick used to create unequal encounter rates by putting multiple identical monsters in the zone, thus I concluded 2 mad scientist to 1 very mad scientist. Since then, of course, he and Xenophobe have created a much more generalized system for tweaking encounter probabilities. Chances are, when I redid the drop rates I didn't notice a glaring difference in encounter rates so I just didn't change it. Feel free to edit the chances to match the actual numbers, though - I'm good at data collection but less proficient with statistics. For the other zones you hinted at, go ahead and match the encounter rates to the numbers if the encounter rates are wrong, and if you have any areas you'd like me to take another pass through feel free to mention it and I'll put it on my list.--Yiab 10:51, 1 March 2008 (CST)

There's also the issue with the bloody Q, which probably messes with the observed rates. (It might be that the real rate is something like 75%/25%, for instance.) --Starwed 12:17, 2 March 2008 (CST)

I think I found an explanation (which basically is, what you're suggesting), see Talk:On the Trail. I'll wait a few days to see if my theory holds up and if it does I'll change the data on this page accordingly, asking Yiab for additional data where OtT-spading doesn't yield enough. --O 3 141592 07:13, 3 March 2008 (CST)

I've posted my data on Talk:On the Trail, but the Knob Goblin Kitchens are definitely 66/33, not 60/40. I tried this for 966 adventures. I'll change the data shortly. --Andylicious 08:44, 4 March 2008 (CST)

After thinking about the Q for a while, I decided that it's almost impossible to come up with an equation which takes into account all the effects you'd see with multiple monsters and uneven distributions. (The chance of seeing a monster depends on whether it's in the Q, but of course whether it's in the Q is determined by how often you see it...) So, I wrote a script to simulate the Q. It predicts that a base rate of 75%/25% would turn into 71%/29% with a Q length of 5, but 33%/67% with a l=3. That's consistent with calculations done at Talk:On the Trail, as well as the observed data, so I'm pretty confident that it works.

I've uploaded the script+XUL interface here, though it'll only work in Firefox, since I'm lazy... :) (The Q length and number of turns simulated are also hardcoded, for the same reason.) --Starwed 00:33, 6 March 2008 (CST)

Good to know that you get the same results using this completely different approach from mine. Now we can both be quite certain that our implementations are correct and any failures that might occur are rather due to errors in our model of the Q. --O 3 141592 08:25, 6 March 2008 (CST)

Predictions of the simulator:

40/40/20 ⇒ 38/38/24

75/25 ⇒ 71/29

70/30 ⇒ 67/33

These seem to match well with the known knob/peak values. --Starwed 00:59, 6 March 2008 (CST)

I'd just like to point out that I have been working for quite some time on an Area Simulator for KoL. It's open source, so if you want to edit stuff, be my guest. I have to admit though, that I've been really slacking as far as the documentation of the simulator goes. But if nothing else, I can at least confirm the above results. --Flolle 22:39, 7 March 2008 (CST)

I should probably chime in here and say that I haven't done any remodeling on the data I've posted here and I haven't taken any queue effects into account - all of this is observed frequencies without reworking. In each area, I estimate the combat frequency, look over the encounters and estimate the ratios at which they occur (I usually assume that they are integer weights of 1 to 3 per monster - an assumption which probably is causing most of the reworking now needed) and divide the frequencies up by estimated weight. For example:

Name	Encountered	Actual Frequency	Estimated Weight	Estimated Frequency
100% Legal	228	5.26%	1 (nc)	5%
7-Foot Dwarf	2029	46.81%	2 (c)	50%
7-Foot Dwarf Foreman	1077	24.84%	1 (c)	25%
A Flat Miner	240	5.54%	1 (nc)	5%
Grilling the Foreman	231	5.33%	1 (nc)	5%
See You Next Fall	231	5.33%	1 (nc)	5%
Speak, Frat Boy, and Enter	236	5.44%	1 (nc)	5%

(Estimating the combat frequency in the area as 75%, from an observed rate of 72.7%). Notice, please, that this data is out of date and does not include some of the area's encounters. Of course, this works very well on the old system where differing chances of encounter actually were multiple identical copies of a monster, but it may not work so well in the new system. Does anybody have any suggestions how I can go about correcting this in future data?--Yiab 09:20, 8 March 2008 (CST)

I think it would probably be best if those total numbers are added to the respective location(probably best on the discussion page), so everyone who wants to take a look at those exact numbers has the chance to do so. You can leave the numbers on this page the way that they are, maybe only add a note about them only being estimates and that the queue might have some influences on them. Maybe this isn't the best way to go, but I can't come up with something better currently. --Flolle 11:17, 10 March 2008 (CDT)