Talk:Search Odds/Archive4

From The Urban Dead Wiki

Jump to: navigation, search
Handgreen.png Archive Page
This page is an archive page of Talk:Search Odds. Please do not add comments to it. If you wish to discuss the Search Odds page do so at Talk:Search Odds.

Search Odds Coverage, Representativeness and Accuracy

Number of Searches Made in each Location
Location Searches Accuracy
Arms 627 Fair
Auto Repair Shop 501 Fair
Cathedrals 132 Poor
Church 145 Poor
Club 136 Poor
Factory (pre 18th November) 126 n/a
Factory (post 18th November) 60 Very Poor
Fire Department (pre 18th November) 1034 n/a
Fire Department (post 18th November) 177 Poor
Fort Armoury 343 Poor
Hospital 1915 Good
Hotel/Motel 156 Poor
Junkyard 125 Poor
Library 159 Poor
Necrotech Building 2365 Good
Outside 362 Poor
Police Department 2428 Good
Power Station (pre 18th November) 135 n/a
Power Station (post 18th November} 139 Poor
Schools 530 Fair
Stadium 53 Very Poor
Warehouse 192 Poor
Mall: Generic Search (No Shopping) 467 Poor
Mall: Book Store (No Bargain Hunting) 150 Poor
Mall: Book Store (Bargain Hunting) 150 Poor
Mall: Drug Store (No Bargain Hunting) 338 Poor
Mall: Drug Store (Bargain Hunting) 866 Fair
Mall: Gun Store (No Bargain Hunting) 695 Fair
Mall: Gun Store (Bargain Hunting) 1358 Good
Mall: Hardware Store (No Bargain Hunting) (pre 18th November) 174 n/a
Mall: Hardware Store (No Bargain Hunting) (post 18th November) 98 Very Poor
Mall: Hardware Store (Bargain Hunting) (pre 18th November) 168 n/a
Mall: Hardware Store (Bargain Hunting) (post 18th November) 171 Poor
Mall: Liquor Store (No Bargain Hunting) 150 Poor
Mall: Liquor Store (Bargain Hunting) 193 Poor
Mall: Sporting Goods (No Bargain Hunting) 150 Poor
Mall: Sporting Goods (Bargain Hunting) 171 Poor
Mall: Tech Store (No Bargain Hunting) 308 Poor
Mall: Tech Store (Bargain Hunting) 350 Poor

Last updated by -- Juntzing 01:53, 1 Dec 2005 (GMT)

Approx. 17,618 Searches made and counting. Keep up the good work!

Accuracy in function of number of searches
Number of Searches in a location Accuracy
0 Nonexistent
1-99 Very Poor
100-500 Poor
501-1000 Fair
1001-2500 Good
2501-5000 Very Good
5001+ Excellent

Accuracy chart.png

Someone added 'book' to the list of items found in the |Necrotech Buildings. If this is correct, then we may have to split the data for this building. Tiarnach 13:53, 30 Nov 2005 (GMT)
Also, does anyone think that the presence of a running generator effects searching? Tiarnach 13:53, 30 Nov 2005 (GMT)
As a note on accuracy, how does the appearance of a new item (a generator, in this case) effect current data? I would think it would make current data for applicable structures invalid, since it excludes the possability of finding the new item. -- Tiarnach 15:13, 21 Nov 2005 (GMT)
It seems reasonable to believe that only the buildings which include the new item are changed, but it wouldn't hurt to mark the dates of major changes on all data pages. --Biscuit
This probably means that, as far as presentation of results is concerned, we'll have to ignore the "pre-genarator" numbers. But it's still worth keeping the old numbers around for comparison purpose. Looks like the only building affected where we had a significant amount of data is the fire department. --Barcalow 15:05, 25 Nov 2005 (GMT)
Could be a good idea to concentrate our efforts in searching Junkyards, Auto Repair shops, Factories, Mall Hardware Stores and Fire Stations to figure out the places with the best odds of finding generators and fuel cans. --Barcalow 15:05, 25 Nov 2005 (GMT)

It would be nice if, when you change an existing data point, that you make a note of is somewhere. I propose adding your signature in small font below that of the submitter so the date is shown as well. That way you can tell at a glance if anything in the existing data has changed. I ran into this during the last update of my spreadsheets and someone changed a data point in the Police Department set... it took me forever to find out why my total wasn't matching the one on this page! Tiarnach 12:27, 28 Nov 2005 (GMT)

<span id="Empty Buildings"/>Empty Buildings: how does this work? I just searched 30 times in a Building, does that mean we're now sure it's still empty? I also searched 10 times in a bank, how many more before we're sure it's still the same? --dayfat 05:52, 30 Nov 2005 (GMT)

I would say that 30 searches turning up nada is a pretty good indication that there is nothing there, or you are extremely unlucky. 10 searches seems inconclusive. Tiarnach 13:53, 30 Nov 2005 (GMT)
However much you want to read into it, it's more evidence backing up the most likely scenario - that is, that nothing has changed in those locations. As a result, I'm now convinced there still isn't anything to be found in the bank. I'll update accordingly (and add your numbers to the Other Buildings section). --Biscuit 19:34, 30 Nov 2005 (GMT)
Okay, I meant if the "(60 more searches)" meant 60 searches more until we're 100% sure it really is empty and not "there's been 60 more searches." It was a little ambiguous. --dayfat 07:56, 1 Dec 2005 (GMT)
This is a really interesting question. I have used probability density functions and the beta distribution as discussed in the Confidence Intervals section below to compute the probability of still finding something, given that we have some number of searches turning up nothing. We can never be '100% sure' that a given building type is empty, but we can decide that we are satisfied with a 95% confidence in our measure of 'sure' and go ahead and compute the chance of still finding something. The plot of this information looks like this: Empty probability.png
For example, if we searched a particular building type 10 times and came up with nothing, we could be 74% sure that it was empty with a confidence of 95%. Another way to think about it is: IF there something in the building that has a chance of being found of 26% or more, and IF we made 10 searches repeatedly, many, many times, 95% of the time we would have found the thing at least once. Here is a table of the likelihood of things being found, given a number of empty searches:
Number of Empty Searches Remaining chance of being found
10 26%
30 10%
60 5%
90 3.3%
100 3.0%
300 1%
600 0.5%
900 0.33%
1000 0.30%
So the gut feeling expressed by Tiarnach (above) that 30 empty searches should be enough, would be good enough to find items that have a chance of being found of 10% or more. To find very rare items, those with a 1% chance of being found (if such items even exist in UrbanDead), we would have to search a building 300 times to be 'sure'. --Stuckinkiel 14:01, 19 Jan 2006 (GMT)

It looks like what I always feared would happen has happened - in an effort to make us all stop spending our time analyzing his game, Kevan has made an arbitrary unannounced change to a major building. I mean he has a point, there are better ways to kill time, but in a sense it's downright mean. I guess we have to restart counting in my favorite building now. --Biscuit 05:07, 1 Dec 2005 (GMT)

Well, while you're at it, can people make sure to check whether the generator's running while you're searching? When we get enough data we can do a proper analysis to determine whether it actually makes a difference or not. -- Odd Starter 05:32, 1 Dec 2005 (GMT)

Yep, we got confirmed Generators in Power Stations. I've already split off the pre-18th November entries from the post-18th November entries.

Current Theories

Here is list of conclusions that are not yet well supported enough to put on the main page:

  • With bargain hunting in a mall, you have a 30% chance of finding something --Barcalow 16:52, 24 Oct 2005 (BST)
Note: I find this strange since the skills page indicates that bargain hunting makes you "25% more likely to find something when searching a mall." According to this, you should have a 1 in 4 chance (25% ) of finding something with the shopping skill, not 30%. --Barcalow 16:52, 24 Oct 2005 (BST)
Main page now says it's 35% chance. I personally suspect it's really a 1/3 chance, and 1/5 without bargain hunting. Fractional rates simply seem to fit the results (and KISS) better. - Dashiva 12:09, 4 Dec 2005 (GMT)
Applied to just the Drug store data, it seems to test well compared to 35/65 (62.88% probability compared to 48.52%). Throwing all Bargain Hunting data in, it compares even better, 72.47%, compared with a mere 6.76% probability of a 35/65 split. So, you seem to be justified, currently anyway. -- Odd Starter talk | Mod 13:49, 18 Dec 2005 (GMT)
Hi there, I didn't mean to cause trouble, but we've had a month and a half to gather more data, so this conclusion (1/3 instead of 35%) should be far more robust now... And in conjunction with this, I was wondering if you could verify that Bargainless Mall Shopping has an 80% chance of failure, instead of the KISS 77.78%. If Bargain Hunting is 3 in 9, perhaps normal Mall Searching is 2 in 9 -- instead of the 20% found elsewhere. Note that this would mean that Bargain Hunting is exactly +50%. Thanks for all the work! --Tycho44 08:33, 8 Feb 2006 (GMT)
  • In NecroTech buildings, you have equal chances of finding Syringes, Extractors or GPS. --Barcalow 16:52, 24 Oct 2005 (BST)
Is there any data on whether the chances of finding Syringes in NecroTech Buildings is actually dropping, as is implied in the item description of the Syringes? --Ethan Frome 16:43, 14 Nov 2005 (GMT)
The item description is referring to the change of syringe effects. In an in character perspective, chances are that there would be some Mark I syringes hidden in the dark corners of NecroTech Buildings. As far as I can tell, the only way for Mark I syringes to still exist are if they're held by people who are holding onto them for souvenirs. I haven't noticed any decrease in the syringe find rate, although it's still worth investigating. --Kleptonis 23:49, 16 Dec 2005 (GMT)
  • It appears that the proportion of successful searches resulting in ammunition remains constant regardless of where you shop. % Found for pistol clips seems to remain constant, and % Found for shotgun shells appear to be moving towards each other, suggestion that they are in reality equal as well. My suspicion is that the probabilities that would normally be assigned to Flak Jackets and Flare Guns is redistributed to the firearms themselves, not evenly distributed across the board as I originally believed. -- Odd Starter 06:06, 3 Nov 2005 (GMT)
  • Even though the Fire Axe can be found in other buildings (Auto Repair Shop, Factory, Hardware Store), the best place to look for it is the Fire Department. It's one of the rare case where it's easier to find an item in a building rather than in the mall with shopping/bargain hunting. You've got about 6.50% chance of finding a Fire Axe in the Fire Department when searching. It will take you an average of 15 AP to find it. --Barcalow 15:18, 1 Nov 2005 (GMT)
  • While data still needs to be processed for more areas, it seems very likely that there is a base probability of 20% to have a sucessful search, followed by an independent event that determines what you found. So far, the Arms, Fire Department, Necrotech Building and Police Department all seem to indicate that you will find nothing ~80% of the time. As someone experienced in programming, this makes sense: First I check if the search is successful, and if not I can skip the rest of the math (80% of the time, in fact). I'd only have to worry about picking an item and adding it to the inventory 20% of the time. The notable exception to this, of course, is the Library. --Tiarnach 00:45, 18 Nov 2005 (GMT)
  • After playing with chi squares for a bit, I believe that Police Station search probabilities are most likely to be as follows:
Result  % Search  % Found
Pistol 2.00% 10.00%
Clip 7.00% 35.00%
Shotgun 2.00% 10.00%
Shell 6.00% 30.00%
Flare Gun 1.50% 7.50%
Flak Jacket 1.50% 7.50%
Nothing 80.00%
For this particular combination, the chi square result is 5.16, which appears to have a 52.34% probability of the results we have being a chance result of that combination, which is a lot better than any other combination I've tried. -- Odd Starter 08:46, 1 Dec 2005 (GMT)
How far off is that from adjusting to fractions: Clip/Shell 6/18, Shotgun/Pistol 2/18, Flare/Flak 1/18? I don't know how the server handles drops, but it seems like a much simpler way of maintaining it. If each item just has a weight, you first check for drop/not, if successful each item drops with probability weight/sigma(weights). - Dashiva 11:59, 4 Dec 2005 (GMT)
Very far, unfortunately. Performing the Chi square on your proportions, we get a result of 24.91. Consulting my chi square probability script, the probability that our data comes from that distribution is around 0.03%. -- Odd Starter 12:15, 4 Dec 2005 (GMT)
But note that the miss on the Chi square test is due to the uneven Clip/Shell ratio. Trying Flare-Flak-Pistol-Shotgun 1/13, Shell 4/13, Clip 5/13, or perhaps Flare 1/12 Flak 1/12 Pistol 1/12 Shotgun 1/12 and Shell 8/27 Clip 10/27, retains the Clip's 25% higher appearance than the Shotgun Shell. It is possible that the Pistol is also e.g. 25% more likely than the Shotgun, although there doesn't appear to be enough data to distinguish. --Tycho44 08:43, 8 Feb 2006 (GMT)

Feel free to add to the list or correct/comment on existing statements.

Are there any thoughts or tests on race conditions or other load issues when searching? I tried searching during the rush hours last night, clicking search 10 times or so without waiting for the page to load, and the results were very disappointing. I wasn't recording, just wanted to spend the AP before sleep, but ~30 AP in drug store with bargain hunting netted ~3 FAKs. I'm fully aware it's a tiny sample size, so a easily possible scenario, and I don't even have the exact results, so I'm not making any claims. Just wondering if anyone else has experience to relate. - Dashiva 10:35, 20 Dec 2005 (GMT)

I find the possibility fairly unlikely. The only possible segment of the procedings that could be a race condition would be the database update segment, and this would likely occur after the search had been resolved. And hey, it's a database update, so I expect that the database software itself would have mutexes and stuff to stop race conditions in that case. -- Odd Starter talk | Mod 21:30, 20 Dec 2005 (GMT)
Yeah, I have no doubts the database can hold its own. However, even the seemingly absurd is possible with the right server scripts, cf. Groove Theory. I don't want to dismiss it outright just because it shouldn't happen. That, and I'm bored enough to spend the next week or so entertaining my curiosity. :) - Dashiva 22:54, 20 Dec 2005 (GMT)

Inaccurate Data

Location Contributor Time of entry Scheduled for deletion on* Deleted/Corrected
Police Department F1r3br4nd 13:14, 5 Oct 2005 (BST) 12 Oct 2005 Deleted
Police Department Juntzing 14:41, 13 Oct 2005 (BST) 21 Oct 2005 Deleted
Necrotech Building Gibso 09:40, 26 Oct 2005 (BST) 2 Nov 2005 Deleted

* Unless corrected by user before deletion

Other Buildings

The following buildings are currently believed to be empty. The number following each building is the number of unsuccessful searches that have been recorded there.

  • Bank (30 searches)
  • Building (159 searches)
  • Cinema (40 searches)
  • Mansion (85 searches)
  • Museum (59 searches)
  • Railway Station (40 searches)
  • I think a friend of mine started the game (I think he gaave up after that - he found it boring) and he was searching around. Anyway, I think he searched a railway station (not sure in in or out) and found a Portable Generator. Is this possible? Jon Kril 19:07, 7 March 2006 (GMT)
  • Tower (40 searches)

Hit Limit Proposal

Hey all, I made a bunch of searching alts (and filled their info in appropriately to ensure they're not "cheating") and I was wondering if anyone else would like to make use of them for gathering data? The names follow a simple short pattern - so01, so02, so03 etc and all use the same password. Along these same lines, I was considering donating some funds to get them freed from the ip hit limit. I would of course provide those who wanted to contribute with their names and password. Is there any need or interest in this?

If you donate money for one of these characters, you're able to do all the searches yourself. If you pm the password to another data-gatherer instead, then the search-character doesn't cost ip hits on your computer and doesn't need to donate. Once several people use that searcher, whoever goes to a coffeeshop or school on that day can perform the 50 searches without loading their ip at all. Keep them at 0xp with a detailed profile explaining their mission and things would probably work out. --Tycho44 11:00, 25 February 2006 (GMT)
Personal tools