#Moowell's Research Thread
fickle flame
I ran a simulation of 50000+ weeks to see what workshop crafts were used over time.
After 375000+ data points, only 1646 crafts ever counted as the highest paying craft for their day
These were the top 100 crafts
timber trout
The fact that 44448 ever hit the highest is astounding and curious
fickle flame
I'm not sure what you mean
timber trout
That said, I like research, but is there any chance you can take a look at the 100 datamined pop charts and run the simulations based off of those?
timber trout
I'm not sure what you mean
Pattern # 1646, for example, is 4hr-4hr-4hr-4hr-8hr for crafting agenda.
No wait sorry
I'm stupid ignore me, essential draught is 6, not 8
fickle flame
Oh, gotcha. And sure. I mentioned in the Guide thread, but I didn't realize datamined charts had already been found
timber trout
needs more coffee
fickle flame
Where can I get those? Google?
fickle flame
Alright. Yep, I can use this instead of purely random ones. I'll post what I find this afternoon
timber trout
Sounds good! Looking forward to it. Just curious though, what are you using to determine the supply peaks? Just randomly throwing them into the day-to-day peak caps?
fickle flame
I'm using rand() to generate numbers for each craft, then ranking them. Depending on their rank, they get a 1, 1.3, or 1.6 multiplier over 6 days
I'd love a better way to do it if you have one. My experience with spreadsheets isn't the greatest, but I'll eventually get where I want to go
static crane
I'm using rand() to generate numbers for each craft, then ranking them. Dependin...
There are 12 patterns for supply peaks, a weak and strong peak for cycles 2 through 7. There are 4 items with each pattern every Season, except for the 2 Cycle7 Patterns, of which there are 5.
Each item starts at 0 supply at the start of a season, and then gets -4 or -8 supply the day before they peak, followed by -4 or -7 supply on the day of their peak, then it adds 10 or 15 to their supply the day after their peak.
So each item is only really worth making on 2 days each season.
plain barn
specifically the high value items are only worth making on 2 cycle each season.. low value items are required regardless to gain access to the high value item
fickle flame
The rank system I'm using sorts each craft into their cycles, adding the 1.3 and 1.6 multipliers only on their specific days
The basic system is I assign a random number to each craft, then rank them from 1 to 50. 1 through 4 is cycle 2 strong, 5 through 8 is cycle 2 weak, etc.
Then on a separate sheet, I have all of the crafts copied over from secret stuff in the guide. I do a VLOOKUP to get the values for all the crafts, add up each set, then determine which set makes the most cowries and increments a ticker. This happens for every day of the week
fickle flame
After 1500-ish weeks of testing, I've got 12122 data points. Only 519 sets of crafts were the strongest on any given day even once. To speed things up, I've removed all the sets with 0 days
static crane
Did you try comparing it against the output from my 1k weeks to see if they were similar?
fickle flame
I can. Just finished 100k weeks
Where can I find it?
Here's the totals after 100k weeks for the sets that survived the initial culling: https://docs.google.com/spreadsheets/d/1asxOFwOQoToue51re4fDCFyW9rc1MuuSJk39fOWkMUY/edit?usp=sharing
fickle flame
Material Frequency over the course of 100k weeks
static crane
Where can I find it?
https://docs.google.com/spreadsheets/d/1rYmTXW7M_jRKMd_BA5K4BovQ0RA1_DT3hC0CmClU_00/edit?usp=sharing
Although I had a big writeup about it in the Granary Expedition Research thread.
There's a lot of varience here, could be coming from a lot of things. For one, I suspect that you didn't multiple the number of crafts by 3 (for the number of workshops making them) Cause an average use of 12 for palm logs seems REALLY low. But on the other hand, an average usage of 36 seems really high, so... Shrug
Is probably the difference in methodology for calculating which things would see use.
fickle flame
It's true, I did not multiply by 3. It's not really needed to understand which craft is best on a given day. Did you cull under-performing crafts? I chopped out all but 520 of them to speed up iteration time
That might have skewed numbers by a small margin
static crane
Yes/No.
The solver I used (Which is 99% Mienna's work, I just did some tweaks to make it function on the dataset I was using) Does only consider a certain set of crafts, but there's still like 65k of them. It's basically any/everything that could realistically chain together.
It also considers things like mid-day breakpoints based on shifting supply (due to your own crafts) which, if you didn't account for, would be the biggest source of deviation for sure.
fickle flame
I didn't account for mid-day breakpoints, no
static crane
It does cull down to only consider the top 15 items for each day as well
Although I don't claim to FULLY understand how that part works.
But yea, if you don't account for breakpoints then high value items will disproportionately outvalue some of the lower value crafts. Because even the day before a peak, things like Brush+Spruce are probably going to beat out most other crafts.
fickle flame
Do you know how your solver accounts for break points? It would be tough with my current model to determine whether something needed to two-slot vs. three slot
Much less swap between overused 4-hour parts
static crane
I'm 99% sure it doesn't suggest crafting different things in different workshop slots. Always the same thing in all 3 workshops.
As for how it calculates... I don't 100% know. I have some theories, but I don't claim to know Mienna's code well enough to know for sure, and I don't want to send you down the wrong path.
Also Mienna's code is... well... Mienna's. It ain't mine, and I know it's not publicly available, so I don't think I should be sharing the secret sauce without her blessing. I don't THINK She'd mind, but I'm more of a "Ask permission instead of forgiveness" kinda person.
fickle flame
No worries
static crane
When I was planning on writing my own solver before I lucked into getting to use hers, My plan had just been to simulate the entire week though, keeping track of supply per item and everything. I couldn't think of a more clever way of doing it. Brute force is not the most elegant of algorithms, but it DOES get the job done...
fickle flame
My model is geared toward seeing which single craft is best on any given day for a single workshop. It doesn't look at second place or whether materials have already been used in the week.
And yeah, brute force is effective if you can cut iteration times down.
static crane
Still might write that solver for fun. I'm in a "Python is the god-king of programming languages" phase at the moment and this is a fun puzzle to model/solve.
fickle flame
I'm definitely not a programmer beyond excel xD
static crane
Oh wait your approach is done entirely in excel!?
fickle flame
Yeah
static crane
I'm surprised/impressed you got that far!!!
fickle flame
Thanks ^_^;;
static crane
Leiton's the one to talk to then. I know that there are some scripts in the google sheet, but I'm pretty sure they only exist to pull down supply/demand data. I think it calculates breakpoints in sheet somehow. Couldn't tell you how though.
I love Sheets, While I think the meme is overplayed, I used to play Eve Online and had... a LOT Of google sheets I used for a variety of purposes.
I don't understand HALF of the various tricks Leiton is using in the comprehensive guide. I fear them.
fickle flame
Hah xD I started on sheets, but iteration time was really slow, so I moved to excel. At the moment, I'm using the model to get a sense of what a typical week will look like for a workshop.
fickle flame
I took a sample of 1000 weeks using my existing model. Assuming you get the standard of 9 groove per day, your average cowries per day are:
Day 1: 1427
Day 2: 1551
Day 3: 1672
Day 4: 1791
Day 5: 1822
Passing the lowest 20% values requires:
Day 1: 1300
Day 2: 1413
Day 3: 1517
Day 4: 1619
Day 5: 1649
These are all for a single work shop, so multiplying them by three gets closer to the true daily value
This averages out to be 8,263 cowries per week in a single shop, or 24,789 cowries per week
Which is really close to the 24,727 you got from your solver
And barely passing the 20% line gets you 22,494 in a week
fickle flame
Totaling up the days by always taking cycles 2-5, then taking the better of cycles 6 and 7, got an average cowries per week of 25107, a max of 28494, and a min of 22185. Only 7 weeks out of 1000 fell below the 20% line
fickle flame
I updated the model, culling workshop setups after 10,000 weeks instead of only 1500. This brought up the number of workshop setups with at least 1 day where they were the highest earning setup from 520 to 1000. I also updated the calculations for the most potent crafts to skip days right before the peak in an attempt to make setups fall better in line with what people will actually run in a week, rather than allowing the craft to be done on both the peak and the day before.
Cowries per day, average (bottom 20%):
1: 1384 (1229)
2: 1502 (1333)
3: 1626 (1438)
4: 1732 (1538)
5: 1800 (1570)
I then calculated what each of the 1000 weeks earns by removing the first day to fall below the 20% line in cycles 2-4. If none of those cycles fell below the 20% line, I removed the lowest earning day from cycles 5-7. The lowest earning week earned 21664 cowries, the highest earned 28870 cowries, and the average was 24831 cowries.