Fishing Skill - An Adventure in Bad Maths

[Toothsplit 2]

Hello Wurminites! Veteran newbie here, very exciting, coming back to the game after a ~4yr break. Much has changed!

 

This post is riddled with possible game mechanic spoilers, you have been warned!

 

Pleasantries out of the way, I've come across something interesting while futzing around with my skill log. I had initially set out to derive a rough formula for skill gain in different skills; being a newbie, my skills all started at 1, so I would have access to a pretty good data set for describing the rate of change in skill gain, given that the most interesting (rapidly changing) part of any plot like this would be at the front end of the set. IANA statistician, mathematician, scientist, etc., so please forgive any mistakes or poor practice!

 

tl;dr

3 unexplained phenomena in skill gain, Toothsplit is bad at math.

 

The setup

What I expected, going in, was to find a rational or power function solution in the "number of actions" domain. This would make that

   ____  / /|

type of plot that we'd expect from decreasing skill gain as skill gets higher (more accumulated actions), with a slant-ish asymptote. Yes, a log could look like that as well, but (IIRC, IANA programmer) logs are computationally more expensive than arithmetic operations, so I thought the simpler solution best. My set would be chopped off a bit through, but the solution should still fit and not change greatly with more data. I'm using good ol' grep to get data out of the logs, and LibreOffice Calc to plot it and fit it. All data thus far is from a single new character on Xanadu, using one tool for the whole log. I'm only plotting fishing and mining since those took place in roughly the same circumstances for the entirety of the set, fishing being sat in one spot on the shore for a few hours, mining being sat in two spots for a few hours, initially on a low-QL iron vein and later just tunnelling out rock.

 

The adventure begins

At first, I just grabbed this month's log, thinking that should be plenty for a simple fit. I subtracted the initial value of the skills so that the sets would start at zero, thinking that would help the trend-line-inator. Mistake #1 right there, but the plot still shows something interesting despite being incorrect. You'll notice, looking at the red (fishing) plot, that it has several 'hikes' in it where skill gain seems to speed up, most easily seen at ~175 actions where it bounces above the blue (mining) plot, and at ~500 actions where there's a clear cusp. Keep that in mind as I fix the issues in my analysis.

Next, I figure out it'd be a good idea to go ahead and splice the logs together, getting a full data set from 1 skill, rather than "subtract 16 and hope for the best," despite a pretty good R² for mining.

And here, after fiddling with the built-in trend-line-inator for longer than I care to admit, I gave up on the prospect of finding a function to describe the mining data. The general form of the plot matches our expectations (log/root-looking start, slant-looking asymptote), but I lack the math and the tools to make a meaningful function out of it. However, that doesn't mean there isn't still something to be learned here! Let's revisit that weird periodicity in the fishing plot.

 

Changing gears

My initial suspicion, relating between the #actions domain and the time domain, was a little far-fetched, but reasonable enough. I suspected that, as in real life, fishing would be tied to the time of day, and fishing skill would be tied to the quality or type of fish caught at different times of day. So! Back to the logs I went to fetch the time data and hook it up to the skill gain ticks so we can more precisely see if certain times of day give higher skill per tick. This would explain the periodic variation in shape in the raw skill plot in the #actions domain, the ticks being the rate of change in skill in either domain.

 

After some futzing with the software (time formatting is painful to a casual user), we get this hot mess:

which, once you understand what you're seeing, blows that hypothesis out of the water (pun points!). The x-axis is the time of day, with 0 being whatever time it was at my first data point, 0.5 being 12 hours beyond that, and 1 being just before that first time, all in Wurm time. There are some obvious breaks in the set where I afk'd a bit or whatever, but there are also a few clear pieces of associated data, such as that sparse line all the way across the middle. All of these associated sets appear to trend downwards. Of note, the sets that are lower on the y-axis are denser than those higher, and the whole set comprises just over 4 in-game days, breaks included, which makes this plot a little murkier than it looks (pun points!).

 

What we should have seen, if our hypothesis was correct, is four overlain parabolic/siney shapes, likely chopped in half or something, but a clear maxima and minima somewhere. They should also lean a little to the right, as skill gain per tick decreases, as we saw above, with increasing number of actions. Instead, we have a mess of associated sets with no clear time-of-day dependence, and importantly much more than four sets.

 

A pair of hypotheses

Two things are likely going on here. In the time-of-day plot, as any fishing veteran has been shouting at the screen, an apparently associated set probably shows us different skill gain for different types of fish. In the periodic-looking raw skill plot, as any veteran at all has been shouting at the screen, the abrupt changes in slope could represent a gradual decrease in tool QL, interrupted after a number of actions by a sudden increase, i.e. I was using a newbie-crafted low-QL rod and repaired it frequently. We can test both of these by further dredging my logs!

 

First, let's see if the dips and cusps in the raw skill plot line up with repair actions. The lazy best way to see this would be plotting repair actions on the same dang ol' chart where we first noticed periodicity, raw skill in the number-of-actions domain.

The dips to 0 are the repair events; I'm sure there's a better way to denote that, but couldn't find it quickly. Pretty clear there isn't a correlation, particularly if you look at ~175 actions, ~250 actions, and especially at ~500 actions. We should expect each of those repairs to coincide with an increase in the slope of the curve on the right side of it, but there exist both changes in slope disconnected from repair events as well as repair events which do not seem to affect slope. I think the repair hypothesis is busted (pun points!).

 

On to our second problem, the disjointed skill gain ticks in a given time set; the hypothesis is that these differences might correspond to different types of fish. Again dredging the event log, we can line up the times for a given tick with a given fish. Plotting these related values should show us the amount of skilled gained when a specific fish type is caught.

For clarity's sake, I omitted the first 10 catches, which were some 50% higher than the rest of the set. Also, I only included the very last tick before the fish was caught. To my eyes, the chart very slightly suggests larger skill gain for roach, and possibly lower for perch, but given the spread, the differences are likely insignificant. No pun here, but I don't think this hypothesis works either. Interestingly, there seem to be two 'groups' of ticks for each fish, one high and one low. Not sure what to make of them, however.

 

The conclusion

I have three phenomena, looking at these plots, that I cannot explain. Firstly and most obviously, I'm unable to figure a good solution to total skill as a function of number of actions; this will likely be different for every skill, as evidenced by the different curves of fishing and mining, but is likely just that I lack the math skill or am not considering more variables. Second is the periodic change in the slope of the skill chart for fishing, as well as periodic change in per-tick skill gain. We can see it's not easily explained by tool QL or in-game time of day, but I lack any other explanations. Third is the existence of more than four associated sets in the plot of per-tick gain over four in-game days, which cannot be easily explained by different fish types.

What I desperately hope someone will swoop in (on their white horse, wearing shining armor) and do is offer some kind of explanation for these. I'd hoped to be able to add some information to Wurmpedia's Tips for Skills article, but I ended up raising more questions than answers. Thanks for reading, and HELP if you can!

[Edge 141]

Different types of actions (iron ore vs rock) and tools (quality) give vastly different skillgain, at different skill levels, It's not possible to put it into a function. A tool or action (type of fish, iron vs rock) can give almost no skill at 1 fishing, but much more skill once you reach 20 fishing. Typicall at low skill level, high quality tools give vastly more skillgain than low ql tools, and it slowly changes to the opposite as you get higher skill. Depending on your repairing skill, repairing in- or decreases the quality of the tool, affecting ticks too.

 

The only usefull way to collect data is with a fixed quality tool, a fixed type of action (mining on iron, rock,..also for fishing the location matters as to what type of fish you catch) at a fixed skill level. As soon as you change any of these, the skillgain curve against number of actions or against time will change completely. You can get a skillgain curve that looks vaguely log2 if you do the ideal action, with the ideal tool at every skill level. 

 

And yes it's very different for different skills.

Edited August 4, 2015 by Edge

[Nicrolis 980]

I have messed with fishing and noticed that ql of the rod seems to affect the frequency of ticks. Not sure how much this helps, but it wasn't stated in here yet

[Kyleshandra 205]

My first thought was that there be some action at work like in mining. Skill gain only comes when the shard is >1 and <40 ql. Perhaps something similar is going on with the fishing. Likewise, when making planks, you make a plank with every action but sometimes it is a fail of 1.0 ql which gives little or no skill.

[Nicrolis 980]

My first thought was that there be some action at work like in mining. Skill gain only comes when the shard is >1 and <40 ql. Perhaps something similar is going on with the fishing. Likewise, when making planks, you make a plank with every action but sometimes it is a fail of 1.0 ql which gives little or no skill.

[Chiqa 742]

Too much thinking, just go fishing

[Zachariah 736]

Too much thinking, just go fishing

 

agreed there was a lot of thought and math put into this investigation. Just activate the rod and click fish and see what happnes

[Nicrolis 980]

Well for some one at 99+ fishing, this math helps out

[Kianga 219]

Maybe not look at skill-tick when related to time of day, for fishing, but rather look at number of fish caught?