Monger charm does NOT increase Trap Loot

There is sufficient HornTracker data now to get a good estimate of if Monger charm increases loot drop rates. The tracker is an unbiased sample, while the self-reporting mechanism of this thread may have significant reporting bias.

I think this math is correct based onhttp://en.wikipedia.org/wiki/Binomia…dence_interval

Monger charm armed catches.

If we pool the 3 trap loot drops, we get 14 drops in 85 catches with Monger charm. With a statistical sampling of a normal Bernoulli distribution, that gives 95% confidence of actual drop rate to be

14/85 +/- 1.96 * sqrt(((14 / 85) * (1 – (14 / 85))) / 85)

0.164 +/- 0.079

Which is well less then the 0.32 self-reported figure. Looks like the horntracker data is about 4 sigma away from 0.32

no Monger charm, 1 or more Warpath Victories.
This gives 39 loot drops on 226 catches of Warmonger.

39/226 +/- 1.96*sqrt((39/226)*(1-(39/226))/226)

0.172 +/- 0.049 with 95% confidence interval.

IMHO, The Monger charm boosting loot myth is BUSTED. And we sure have some reporting biases! (I know I do, too depressed after missing loot on my last 2 Monger catches to bother submitting :-P)

Posted in Uncategorized | Leave a comment

Rewer’s Riposte OWNS Tactical Mice


Against Tactical Mice you get 1.75 effectiveness boost.

Assuming you have Molten Shrapnel Base,

ZFM/MSB/LGS luck boost is worth

(3-1.75)*(1.75*32)^2 = 3920 power

However, with RRP/MSB/LGS

(3-1.75)*(1.75*40)^2 = 6125 power

A Nerg Chieftan has power of 17920. For Nerg Chieftan,

ZFM/MSB/LGS catch rate = 47.61%

2010/MSB/LGS catch rate = 51.59%

RRP/MSB/LGS catch rate….

(4064*1.75+(3-1.75)*(1.75*40)^2)/(4064*1.75+17920) = 52.88%

Even better will be RRP/Magma/LGS.

Posted in Uncategorized | 2 Comments

UPDATE : Catches are not streaky, they are random and independent

UPDATE :

MORE evidence that each hunt is independent of the previous ones.
For Sweet Havarti in JoD with Clock or CLT, Magma, Molten or Spellbook…  3,674 hunts
Overall Catch Rate of JoD mice = 33.0%
Catch Rate after 1 catch in a row = 33.1%
Catch Rate after 1 fail to catch in a row = 33.0%
Catch Rate after 2 catches in a row = 34.4%
Catch Rate after 2 fail to catches in a row = 32.2%
Catch Rate after 3 catches in a row = 26.3%
Catch Rate after 3 fail to catches in a row = 32.1%
Any streaks you see are from random variation…  Not programmed streakiness.

Original Post

A great many people say that a trap setup or mouse is ‘streaky’, in that if you are on a good streak, you are more likely to continue catching mice, and if you are on a bad streak, you are more likely to continue to miss mice.  People even recommend switching traps or locations to combat that streak.

The savvy mousehunter has always assumed that each hunt is independent of the previous one, and that any streaks that occur happen in the same frequency as what you would expect from a totally random process.  However, we have never had clear evidence that this is the case. Perhaps the devs are more nefarious than we thought…

One way to demonstrate the presence or absence of unusual streakiness is to take a really long series of hunts and plot the number of times a streak of Fail to Catch / Attract occurs.  Then take that same data, and randomize the order that the catches and misses occurred in.  Imagine every hunt is a card in a playing deck. We just shuffle all the cards and then see how many streaks there are of various lengths.

Let’s shuffle the ‘deck’ 10,000 times and record the # of streaks and their length after each shuffle.  Now, if a setup is unusually ‘streaky’ we should see that the number of long streaks of FTC/FTA from the real data should be greater than the average number of long streaks from the randomly shuffled data.  Is this the case?

Above is data from a hunt log from a single player in Derr Dunes that was for 2269 hunts with the Chrome Drill Bot / Magma Base / Lucky Golden Shield.  The red circles indicate the number of FTC/FTA streaks of a given length.  The blue line indicates the average number of streaks of a given length when we shuffled this hunting ‘deck’ 10,000 times.  The errorbars indicate the 5% and 95% least and most # of streaks of a given length.

Clearly, the real data has the SAME streakiness as a totally random process.  Any perceived unusual streaks are just your brain searching for order where it doesn’t exist.  Mousehunt works just like the experts said, each hunt is independent from the previous one.  The best way to break a streak is to just keep hunting with the optimal setup in the region you are at.

The data I used for this analysis can be downloaded here DerrStreaks.

Data pulled from http://www.hsuke.com/mhunt/Data_View.php

Hunt Log #508

Posted in Uncategorized | Leave a comment

Mousehunt Catch Rates 3.0

Aaron Milstein and SethK appear to have cracked the Mousehunt 3.0 Catch Rate equation. Sam Siebert’s equation based on Mousehunt 2.0 data was pretty far off in some regions after the release of Longtail (Mousehunt 3.0).  These data also draw on Paul Humphreys’s updated mouse power and effectiveness estimates.

A summary of these updates can be seen on this spreadsheet.

The catch rate of a mouse is dependent on Trap Power (T), Trap Effectiveness (Eff), Luck, and Mouse Power (M).  In the new equation for the Longtail release of Mousehunt (version 3.0), the catch rate can be estimated by this equation :

Why should you believe me?

Check out the plots Seth has provided, they have me convinced.

100%: PHYSICAL--Chillbot/magma/shield vs various Indigenous, Derr and other mice that should be 100% eff

150%: ARCANE--ACRo/MSB/shield vs Balack's Banished.

175%: TACTICAL--2010/Magma/shield vs Nerg, Furoma, and a few other mice 175% susceptible:

200%: ARCANE--ACRo/MSB/shield vs Bristle woods forgotten (note this should be unchanged from MH2 as (3-eff) in this case = 1

I am pretty much sold on this equation.  Seth K however, does mention a few caveats

1) Data compiled by COMBINING results from Nick A and Nathan Y’s databases (submitted between 8/1/2010 and today)–yes we are aware there are people who post logs to BOTH, and therefore there is some data duplication, which means the error bars will be a bit larger than represented in the graphs…
2) Two anomalous mice have been noted (neither shown in these graphs). The first is the Squeaker claws which ought to be a 100% catch but isn’t–we are not sure exactly why–it is the ONLY anomalous event mouse. The other is the Swarm of Pygmy–it is acting in terms of CR like a 3000-power shadow mouse currently (it was 3000 power according to Paul Humphreys in MH2, but he raised it to 4000 just recently for MH3–I personally believe this is an error by Paul, because the in-game power order displayed on the mice page of the application supports 3000 but not 4000.

3) Ideally I would like to have much more data for smaller error bars like we had at the time we were working on the MH2 equation that Sam Siebert ultimately solved, but we just don’t at this point. We have to make do with what we have.

Very impressive work guys.  I hope that I was able to help out in sparking a reexamination of the equation, but I certainly wasn’t thinking it would come out like this.

Posted in Uncategorized | 11 Comments

Catch Rate Estimates Revisited

UPDATE #4 :

A new formula has been found!  This page is out of date. GO HERE

 

UPDATE #3 :

Double random variable model in blue, Derr Dunes

UPDATE #2 :

New Model : The Head to Head model.

Player has two chances to catch a mouse.

First Chance :

Player rolls a die from 1 to Trap Power * Trap Effectiveness
Mouse rolls a die from 1 to Mouse Power
If player roll > mouse roll, catch!
If not, go to the second chance

Second Chance :
Player rolls a die from 1 to (Luck*Trap Effectiveness)^2.4
Mouse rolls a die from 1 to Mouse Power
If player roll > mouse roll, catch!
If not, fail to catch.

The model never has a value above 100% and has a nice S-shaped curve at low mouse powers, similar to what is seen in actual data. So far I haven’t been able to distill it down to an analytical formula, I’m just simulating 100,000 rolls per condition.

Currently I am comparing this to other regions, and its looking pretty good across the board.  For now, let’s just look at how it compares to the Standard Model.

Models vs. Actual Catch Rate for Derr Dunes

Actual catch rates are in magenta boxes.
Standard Model predictions are black dots.
Head to Head Model predictions are green dots.

Nerg for 2010/Magma/LGS using 1.5x effectiveness

UPDATE#1 :

As noted in a previous post, the current standard model for estimating catch rates from Trap Power, Luck, Effectiveness and Mouse Power works in many cases but there are some clear examples (Derr Dunes) in the Longtail release of Mousehunt where the equation fails to match real catch rates.  I don’t have an answer to WHY yet, nobody does, though a dedicated group of mousehunters are trying to figure it out.

Standard Model of Catch Rate Estimates

One common complaint from people is that the catch rate estimate equation doesn’t look like it treats luck as a second chance, or roll, to catch the mouse. The developers have clearly stated that luck acts as a second roll. Well, about that…

Tonight, I was building  alternative catch rate models in Matlab to see predicted catch rates, and one of the models gave the exact same predictions as the current model. That model was set up as two rolls! Just like how the devs say the catches work. Working through the fractions shows that the current catch rate model is mathematically equivalent to a system where there is :

Roll 1 - Catch if a random number from 0-1 is lower than
Eff * Trap Power / (Eff Trap Power + Mouse Power). If higher, then go to Roll 2.

Roll 2 – Catch if a random number from 0-1 is lower than
(Eff * Luck)^2 / Mouse Power. If higher, miss.

So lets name these elements shorter names

T = Eff*Trap Power
L = (Eff*Luck)^2
M = Mouse Power

Standard equation is :

CRE = (T+L) / (T+M)

Let’s now look at the 2 Roll hypothesis.

Since the odds of failing to catch on roll 1 are (1-(T / T+M)), The 2 roll equation is

CRE = T / (T+M) + (1-(T / T+M)) * L / M

Multiply out the 2nd roll

CRE = T/(T+M) + L/M – L*T/(M(T+M))

Now we need to get the denominators all the same, so multiply by M/M for first term, (T+M)/(T+M) for the second term. We then get

= (T*M + L*(T+M) – L*T) /(M(T+M))

Multiply out the middle term

= (T*M + L*T +L*M – L*T )/(M(T+M))

Cancel out terms

= (T*M + L*M) / (M(T+M))

Remove the Ms

CRE = (T + L) / (T + M) The SAME as the Standard Equation

So the Standard Model PERFECTLY fits the description the devs gave regarding luck being treated as a second roll!

Now what to do about those pesky deviations from the expected catch rates?

A few possibilities…

1) The second roll is actually Eff Luck^2 / (Eff Trap Power + Mouse Power)

This will result in very similar catch rates to standard model for mice whose power is << Trap Power. As Mouse Power increases, the catch odds from the 2nd luck roll is decreased. This would seem to fit the data for Dragon. However, maybe not for Derr.

2) The Trap Power and Luck effectiveness multipliers are different.

That’s all I’ve got for now, will investigate these possiblities and any other suggestions and will update this page soon.

Posted in Uncategorized | Leave a comment

Double Diamond Adventure Catch Rate

DDA Predicted Catch Rate

DDA / MSB / LGS

Protector 52.7%
Champion 46.5%
Elub Chieftain 40.4%
Hydra 42.3%
Silth 15.9%

Double Diamond Adventure is ever so slightly better than Kraken Chaos and should be used by everyone with a LGS for catching all Hydro mice except for Silth.

Silth should use Heat Bath > DDA > Kraken > ASG > ETR.

 

Posted in Uncategorized | Leave a comment

Catch Rate Estimates

UPDATE :

A new formula has been found!  This page is out of date. GO HERE

In Mousehunt, everyone wants to know what the best trap to use in their situation.  Ask on the forums and you are bound to get a variety of conflicting advice.  How can you decide who is correct? How do we properly evaluate the relative strengths of traps?  We need to have an estimate for the catch rate of a mouse based on trap properties.

The catch rate of a mouse is dependent on Trap Power (T), Trap Effectiveness (Eff), Luck, and Mouse Power (M).   For most locations, this catch rate appears to follow this formula.

Using this equation, we can visualize the effect of Trap Power, Effectiveness and Luck for a bunch of different mouse powers.

In the following plots, Trap Power is on the x-axis, and Catch Rate is on the y-axis.  The colors indicate the total Luck of the trap (including base and lucky golden shield), increasing from 0 (blue) to 45 (red). Each panel indicates an effectiveness rating which is dependent on mouse and trap class (Shadow, Physical, Tactical, etc).  Normal effectiveness is 100%.  The curves show the estimated catch rate for a given Trap Power, Mouse Power, Luck and Effectiveness.

The first panel is for a mouse power of 100, this is around the power of a Wiggler mouse, stronger than a White or Brown mouse, and weaker than a Dwarf.

The second panel is for mouse power of 300, around the power of a Mutated White, a bit stronger than a Dwarf, and weaker than a Steel mouse.

The third panel is for mouse power of 1000, around the power of a Diamond, a bit stronger than a Bionic, and weaker than a Elven Princess mouse.

The fourth panel is for mouse power of 3000, around the power of a Wordsmith, a bit stronger than a Zombie, and weaker than a Keeper’s Assistant mouse.

The fifth panel is for mouse power of 10000, around the power of a Renegade, a bit stronger than a Furoma Student, and weaker than a Furoma Master mouse.

The sixth panel is for mouse power of 30000, around the power of a Master of the Dojo, a bit stronger than a Magma Carrier, and weaker than a Dragon mouse.

This equation was not discovered by me, I just made the pretty plots. For much more info on mouse powers, the development of this equation, and even some places where it seems to be a bit off, check out the links on the MouseHunt Data 3.1 spreadsheet.

Below is a table of mouse powers I compiled from data from Paul H, Geoff O and probably others.  Click for high res.

If you would like to play around with the code used to generate these plots, a Matlab m-file is below

catchRate.m

Posted in Uncategorized | 4 Comments