Probability of a unique variant?
Dom, 02/13/2011 - 15:05
In order to get the best possible estimate of how rare unique variants are, a lot of crafting is required in order to smooth out small random fluctuations due to noise. Depending on how low the rate is, we may need to craft several hundred haze bombs to get a good estimate.
Towards this end, I'd like to propose that we pool our data, being careful to keep track of how many items are crafted along the way. I've begun a personal project of mass-crafting 1* recipes like haze bombs (which are fairly cheap). The haven vendors sell other 1* recipes that are similarly cheap to craft, which means that we should be able to make a ton of items without critically depleting the supply of just one specific ingredient.
I will also try to note what that unique variant is. With enough data, we may even be able to determine whether good variants are less common than low bonuses.
Here are my numbers so far (to be updated):
Day 1: Made 7 haze bombs; no unique variants. (notation: Haze bombs, 0/7)
Wolver ---> dusker cap: 1/2 (moderate curse resistance at the dusker step)
Wolver--->dusker coat: 0/2
Day 2: crafted haze bombs.
1/9 (charge time reduction, low)
Total: 2/20
Dom, 02/13/2011 - 18:53
#2
Shortened notation
Wouldn't a more logical shortened notation be the other way around? (fraction of unique variants out of total crafted) Then Equinox's above statement would be "Firecracker, 0/1"
Lun, 02/14/2011 - 08:31
#3
Haze bomb: 1/4
Haze bomb: 1/4 (Damage v Gremlins: Ultra)
Dom, 02/13/2011 - 22:25
#4
Haze bombs: 3/20
Haze bombs: 3/20
Dom, 02/13/2011 - 23:43
#5
Haze Bombs: 2/18 (Increased
Haze Bombs: 2/18 (Increased Construct Ultra; Increased Gremlins Low)
Lun, 02/14/2011 - 01:39
#6
Trial 1: 1/17 (all haze
Trial 1: 1/17 (all haze bombs) Damage vs Slime low
Trial 2: 1/9 (4 haze bombs/5 heavy hatchets) Damage vs Gremlin High
Trial 3: 1/3 (all heavy hatchets) Attack Speed Increased High
so 3/29 total.
Lun, 02/14/2011 - 08:01
#7
It's too early to say for
It's too early to say for sure, but the initial data seems to be suggesting that unique variants drop at a rate of ~10% total (maybe higher- cactuscat's numbers do skew us upwards!). However, if we try to break it down further, it's still unclear whether "high" and "ultra" bonuses have a different probability than "low" bonuses. Keep that data coming, and if possible, remember to indicate the type of variant alongside the numbers!
Lun, 02/14/2011 - 23:25
#8
Made 2 haze bombs, got 1
Made 2 haze bombs, got 1 unique variant for: Attack Speed Increase: Moderate
Also made 2 wolver caps, got 1 unique variant for: Stun Resist Increase: Low
Jue, 02/17/2011 - 00:05
#9
Haze Bomb: 1/10 (Increased
Trial 1 - Haze Bomb: 1/10 (Increased damage vs Gremlins:Moderate)
Trial 2 - Haze Bomb: 2/10 (Increased damage vs Slime:Ultra, Charge time reduction:Low)
Mar, 02/15/2011 - 00:04
#10
On a side note, all of my
On a side note, all of my Haze bombs had "Moderate" damage bonuses, I forgot to mention that. Will post more later once I get some more energy.
Jue, 02/17/2011 - 09:29
#11
Haze Bombs: 2/7 (Increased
Haze Bombs: 2/7 (Increased Fiend Moderate, High)
EDIT: Sorry Pauling, typo. xD
Jue, 02/17/2011 - 09:02
#12
Wait- increased "field"???
Wait- increased "field"??? I've never heard of that bonus type before. Can anyone confirm?
Jue, 02/17/2011 - 15:34
#13
Fiend* lol sorry.
Fiend* lol sorry.
Haze Bombs: 1/10 (Increased Undead Moderate)
Vie, 03/18/2011 - 08:56
#14
So now that the range of
So now that the range of bonuses for normal equipment has changed, has anyone noticed if unique variants no longer go all the way up to ultra or whatever? Possibly for folks with extra energy, a little more crafting to test this out would be an option.
Dom, 03/27/2011 - 21:38
#15
Since the wipe is coming, I
Since the wipe is coming, I decided this was a good time to test. I'm out of gel drops, shards, and patience, so I'll let someone else continue from here.
For all weapons: a total of 16 variants in 174 attempts. (9.2% probability)
In 73 attempts, I got another 8 unique variants (a rate ~ 1 in 9).
The list includes:
Damge bonus vs fiend; medium
damge bonus vs construct: low
damage bonus vs undead: low
damage bonus vs beast: medium
attack speed increased : medium
damage bonus vs beast: high
attack speed increased: high
damage bonus vs fiend: high
EDIT: I then proceeded to try the same trick with thwack hammers: 5 variants out of 74 attempts. Note that "very high" variants are still possible!
damge bonus vs construct: medium
damage bonus vs fiend: medium
damage bonus vs gremlin: low
damage bonus vs slime: very high
attack speed increased : low
And the frost gun: 1/13 unique variants
Attack speed increased: medium
Circuit breaker helm: 2/14 unique variants
increased fire resistance: low
increased poison resistance, high
Dom, 03/27/2011 - 21:46
#16
Pauling... I thank you for
Pauling... I thank you for your research. In the Name of Science you are spectacular.
(I think the unique chance is approximately 10%)
Dom, 03/27/2011 - 21:52
#17
Summing up all the posts in
Summing up all the posts in this thread, the cumulative rate is 35 unique variants in 306 attempts, a rate of 11.44% (1 in 9). This could of course be skewed a bit by several really lucky trials posted, but does anyone reading this have the statistics background to say for sure?
Lun, 03/28/2011 - 10:31
#18
On Statistics
Don't really need a background in statistics, but I know enough of the math involved, that I could suggest some things:
It might be that the odds of a unique variant are different across stars, item group (helm, shield, armor, sword, gun, bomb), or even by the specific item type. So, the "total" lines probably won't be useful data, but if the odds are the same for all crafting, then calculating the probabilities across all star/group will support that. Meaning: post what you make, and the UV/Normal numbers, not just "I made 100 things today, 9 were UV".
Also, the certainty of these calculations will increase as more of the same thing are made. So, when one person makes 20 Haze Bombs, it will still be useful to collect data by making more. In the voodoo of statistics, there are things like the normal distribution, sample size, confidence level, blah blah blah- The larger our sample size, the better the confidence in the result. And I think a reasonable sample size starts at 100.
Finally, the list of resulting UV effects is also useful, since maybe not all UV effects appear on all items, and we can additionally correlate these effects with the item types they appear on.
My contribution to this is very little data, but here goes:
I made 1/0 Silvermail, came out UV with dmg vs. undead: low
and 1/0 Winmillion, UV with charge time reduction: very high!
(so sad these are gone now)
Dom, 04/17/2011 - 13:50
#19
I would say that, at least
I would say that, at least initially, you should assume different UV rates (and different quality of UV granted) for each star level of gear. So avoid mixing data on * with data on ** tests.
I would say it ought to be a safe assumption that the UV rate/quality variance would only be based on * level, not on specific recipe as well.
Lun, 04/18/2011 - 01:56
#20
Wolver coat: 1/3 (Got
Wolver coat: 1/3 (Got moderate shadow defence)
Wolver cap: 0/4
Blast bomb: 0/3
Super stun gun: 0/3
Dusker cap: 0/1
Firo driver: 0/1
Lun, 04/18/2011 - 08:58
#21
160 Haze Bombs
Here's a list of my UV's... just on haze bombs.
Damage vs Fiend: Low
Damage vs Fiend: Medium
Damage vs Fiend: Low
Damage vs Undead: Medium
Damage vs Undead: High
Damage vs Slime: High
Damage vs Construct: Low
Charge Time Reduction: Low
Charge Time Reduction: Low
Damage vs Beast: Low
Damage vs Slime: Low
Damage vs Slime: Low
Damage vs Undead: Low
Damage vs Fiend: Low
Damage vs Beast: Medium
Damage vs Construct: Low
Damage vs Construct: Low
17/160, all made in one continuous go.
Mar, 04/19/2011 - 03:30
#22
Rigorous analysis
So, this format will be really difficult to mine for data if we later realize that there were factors we neglected. Can the Wiki include a table that people can edit? If so, columns of |date/time (probably irrelevant, but better safe than sorry): item type: item level : item name: UV : any variation in starting mat (for upgrades)| covers every variable I can think of off the top of my head.
It certainly looks like a straight 10ish% chance of a UV, and then rolling on some property table that doesn't seem all that heavily weighted towards lower end stuff (which was my assumption before reviewing the data). But with a good data set, it'd be pretty easy to put together a full probability table (a little more work from the experimenters would result in a lot less work for analysts). And it'd be pretty easy to do a naive bayesian analysis (analysis of the influence a variable has on the outcome, sort of) on any variable that's well documented in a table.
Anyway, the analysis is probably an evening's work, once the data's there. I'd be delighted to do it in a couple of weeks (when my school semester's over), but I'm not likely to go through and assemble a table from this thread.
Dom, 05/01/2011 - 17:45
#23
Guys, include heat level as
Guys, include heat level as well. My own personal experience seems to strongly suggest that odds are better if you upgrade a higher-heat item.
I'm thinking it's probably something like
UV Drop % = (Base 10%) + (heat level)*(5%)
Dom, 05/01/2011 - 19:28
#24
I sat down to a lucky
I sat down to a lucky brandish streak, 4/8. All of them were low vs. type though.
Also 2/5 skelly shields, one medium and one low.
I didn't get a variant when upgrading any of my 4* items, and I've made 8 of those. (4 weapons, helm/armor/shield, 2 sold sunset pieces). Since 4* items are made from level 5 items minimum, your suggested rate would be 35% UV chance. The probability of that guess being correct is therefore .65^8, or ~2.07%
Dom, 05/01/2011 - 20:22
#25
I'm hoping that people will
I'm hoping that people will post here, but just in case anything gets lost, am posting a link to a general discussion thread so that it doesn't get lost/ forgotten about:
http://forums.spiralknights.com/en/node/6361
Mar, 05/03/2011 - 14:38
#26
Hi, UV probabilities are not
removed..
Mar, 05/03/2011 - 14:19
#27
Netmask... what is your
Netmask... what is your source for this? We can't put something on the wiki until it's possible to independently confirm the results. (there's a lot of wild speculation out there)
Mar, 05/03/2011 - 14:38
#28
Removed
removed..
Mar, 05/03/2011 - 14:37
#29
Anything that you see in-game
Anything that you see in-game is what you work with. It's not encouraged to use data from inside the game files stored on your computer...
Mar, 05/03/2011 - 14:48
#30
Fine, removed the info. 3R
removed..
Mar, 05/03/2011 - 23:49
#31
UVs
I've gotten 8 unique variants so far, but they are all "low". Am i doing anything wrong, or am i just unlucky? All but 1 were on 2 star items.
Vie, 05/06/2011 - 19:12
#32
I'll post some of my results
I'll post some of my results if this topic is even still being watched:
Item: Wolver Cap
Total Crafts: 25
Total UV Obtained: 7
28% of my crafts came out as a UV
Detailed Results:
Crafted 2 Normal
Crafted UV Shadow Def Low
Crafted UV Poison Res Low
Crafted 1 Normal
Crafted UV Sleep Res Low
Crafted 4 Normal
Crafted UV Piercing Res Low
Crafted 4 Normal
Crafted UV Poison Res Low
Crafted 3 Normal
Crafted UV Freeze Res Low
Crafted 3 Normal
Crafted UV Sleep Res Low
Crafted 1 Normal
Dom, 06/05/2011 - 09:35
#33
Change since the 5/17 Update?
I may just have had a really lucky streak before, or am just really unlucky now, but has anyone else noticed a drop in the number of UVs they get while crafting since the big Auction House/Change update?
Dom, 06/05/2011 - 18:10
#34
Hm
http://forums.spiralknights.com/en/node/9126
Why isn't this link here?
Calculated by a forumite:
After 10% chance of getting any uv :
Low: 65.74% (+/- 8.95%)
Medium: 21.30% (+/- 7.72%)
High: 7.41% (+/- 4.94%)
Very High: 5.56% (+/- 4.32%)
Sáb, 06/11/2011 - 10:49
#35
horrid luck
haze bombs
0/13
:|
We could probably just shorten the notation.
Firecracker 1/0