1 minute ago, Brunas said:

Actually, I should be more clear - I love the way meta-wing presents data. The vast majority of users aren't interested in a billion different competing percentages and such, they just want to see what's performing well and what to expect at a tournament.

We're the weird edge case users

That's why I do it myself by now

Just now, GreenDragoon said:

That's why I do it myself by now

It's hiding in the regional thread so not as many people see it as should, because you post some scary numbers! I went to copy/paste some here, especially the Yorr win% rate, but it looks like they're gone from imgur now

I agree

Just now, Brunas said:

It's hiding in the regional thread so not as many people see it as should, because you post some scary numbers! I went to copy/paste some here, especially the Yorr win% rate, but it looks like they're gone from imgur now

Actually I have the word doc open right now (after I learned that my meeting with the boss tomorrow got cancelled...) and I'm putting it together to make a thread for it.
I removed all imgur images because they are outdated (more lists and regionals got added in the meantime) and I really don't want to mix different data sets.
It will be "from nerds for nerds" though, with way too much on methodology. But I think it's important to know where the numbers are coming from.

35 minutes ago, GreenDragoon said:

It will be "from nerds for nerds" though, with way too much on methodology. But I think it's important to know where the numbers are coming from.

The most beautiful words in the english language.

1 hour ago, GreenDragoon said:

@Tlfj200

about your point of meta diversity: I thought so too before you mentioned it. IMO the difference is not just the amount of lists, but the amount of „top lists“ and how close they are to each other. StayOnTheLeader had made a graph in one of his meta analysis threads (here's the thread) that illustrates what I mean.

edit:

this graph

IMO "there is so much more diversity" applies when the blob that clusters by itself, here called"tier 1", is gone because there are many viable squads that can actually compete. Of course we have GhostFenn, but I think it's not as far removed from the rest as other squads were in the past.

Agreed,

Though, even by that metric, the meta appears to be about as diverse as earlier metas (with ghost feen slowly moving up in numbers every week). Granted, this is nothing but post-cut date (rephrased, it only looks at lists that made the cut):



Source:
https://docs.google.com/spreadsheets/d/1kNos2DhJfYOwKfFBVZS3GbnADB4WL6NvByuCeNmLdtc/edit?usp=sharing

Just now, Tlfj200 said:

https://imgur.com/a/yeKGz

jesus christ I'm so disappointed in you

Right click -> Open image in new tab

Tab new URL, and just hit paste in here

SHAME

For the meta being open, it always is... For like a couple weeks after a new wave.

Who knows when the worlds article was written?

When @Tlfj200 says "Listener, let's have a heart to heart" I remain steadily and impassively doing whatever I was doing, but a little part of me inside is jumping up and down and clapping.

1 hour ago, GreenDragoon said:

It will be "from nerds for nerds" though, with way too much on methodology. But I think it's important to know where the numbers are coming from.

After the amount of time I've spent in science, every time I see number without methodology, I assume there's a 90% chance of bull, with the remaining 10% being an actual quality data analysis method that the person would rather not give away for free. Frequently it's the first option, masquerading as the second option.

Kudos to anyone who shows their work.

42 minutes ago, Brunas said:

jesus christ I'm so disappointed in you

Right click -> Open image in new tab

Tab new URL, and just hit paste in here

The more in-depth analysis will come post-worlds (partially because of the amount of work).

Right now, Here's what these numbers include:

1. Only Regionals that are 50+ people (arbitrary cut off, but I had to cut off somewhere, and I felt smaller tournaments skew away from a real 'meta').
2. The current regional data (2018) starts after January 22, 2018 (the trajectory simulatory/genius nerf) through February 25, 2018.
3. The 2016 data (Deadeye nerf) goes from wave 8 release to the nerf, and includes gencon and US nationals.
4. 2017 data is entirely inclusive and ignores all FAQs (simply because the post-faq data is so small it's irrelevant).

Edit: *sigh* I accidentally quoted Chris instead of myself...

@Tlfj200 - Forgive me, I slept through part of the lesson, I think. What is the difference between percentage and cumulative percentage? I'm making assumptions based on only the headers since I forgot to download the podcast before going to work today.

My assumption is that percentage represents how often a single row (category/ship(s)?) show up in lists, and cumulative is based on making the cut? No, that seems... borked... Please let this be wrong.

3 minutes ago, LagJanson said:

@Tlfj200 - Forgive me, I slept through part of the lesson, I think. What is the difference between percentage and cumulative percentage? I'm making assumptions based on only the headers since I forgot to download the podcast before going to work today.

My assumption is that percentage represents how often a single row (category/ship(s)?) show up in lists, and cumulative is based on making the cut? No, that seems... borked... Please let this be wrong.

No Sweat.

The cumulative percentage is just tracking how much all the lists add up together.

So if the highest list is 15%, and the 2nd list is 10%, together they make up 25% of the meta. So I track that stat to see how many lists it takes to get to 50%, 75%, and 90% of the entire meta (highlighted), to get a feel for how "diverse" the meta is.

The more lists it takes to reach each benchmark, the more diverse the meta is.

3 minutes ago, Tlfj200 said:

The cumulative percentage is just tracking how much all the lists add up together.

Ah, I see it now. Thanks

14 minutes ago, Tlfj200 said:

The more in-depth analysis will come post-worlds (partially because of the amount of work).

Right now, Here's what these numbers include:

1. Only Regionals that are 50+ people (arbitrary cut off, but I had to cut off somewhere, and I felt smaller tournaments skew away from a real 'meta')
2. The current regional data (2018) starts after January 22, 2018 (the trajectory simulatory/genius nerf).
3. The 2016 data (Deadeye nerf) goes from wave 8 release to the nerf, and includes gencon and US nationals
4. 2017 data is entirely inclusive and ignores all FAQs (simply because the post-faq data is so small it's irrelevant).

Edit: *sigh* I accidentally quoted Chris instead of myself...

Oh, and 2018 data is specifically:
1/22/2018 - 2/25/2018

26 minutes ago, Biophysical said:

After the amount of time I've spent in science, every time I see number without methodology, I assume there's a 90% chance of bull, with the remaining 10% being an actual quality data analysis method that the person would rather not give away for free. Frequently it's the first option, masquerading as the second option.

Kudos to anyone who shows their work.

Care to show where your 90/10 came from? ;P

15 minutes ago, LagJanson said:

@Tlfj200 - Forgive me, I slept through part of the lesson, I think. What is the difference between percentage and cumulative percentage? I'm making assumptions based on only the headers since I forgot to download the podcast before going to work today.

My assumption is that percentage represents how often a single row (category/ship(s)?) show up in lists, and cumulative is based on making the cut? No, that seems... borked... Please let this be wrong.

Don't feel bad, I also missed it, though i am notably bad at listening

Just now, catachanninja said:

Care to show where your 90/10 came from? ;P

I shared that the value was based on experience and assumption, certainly not the strongest sources, but you know where it came from. The value you place on that assumption is up to you.

14 minutes ago, Tlfj200 said:

Oh, and 2018 data is specifically:
1/22/2018 - 2/25/2018

So... 312 lists from events with 50+ attendees within a one month time frame. Just curious, were the sources complete for all these events? Typically the large ones have significant holes further down the ranks. These admittedly should have a low impact to your meta numbers. Just a curiosity. Slow day at work so I'm playing with the numbers for fun.

Just now, LagJanson said:

So... 312 lists from events with 50+ attendees within a one month time frame. Just curious, were the sources complete for all these events? Typically the large ones have significant holes further down the ranks. These admittedly should have a low impact to your meta numbers. Just a curiosity. Slow day at work so I'm playing with the numbers for fun.

Yeah he cuts were complete, but no, not the entire event.

So ‘field’ data is more skewed due to blanks.

there is still stuff to learn, because the field data is ‘too heavy’ since people are more likely to fill out lists if they did better, which is still useful, but we won’t get the universe of lists/breakdowns

17 minutes ago, LagJanson said:

So... 312 lists from events with 50+ attendees within a one month time frame. Just curious, were the sources complete for all these events? Typically the large ones have significant holes further down the ranks. These admittedly should have a low impact to your meta numbers. Just a curiosity. Slow day at work so I'm playing with the numbers for fun.

Also, admittedly, I just don't have the answers yet, because it's a WIP.

Immediate first thing was getting all the functions right, and parsing the most relevant data (cuts); next step is expanding things as we ask questions, and broadening to the entire "field" (while finding ways to account for "holes" in the field data)

5 minutes ago, Tlfj200 said:

because it's a WIP progress.

Do you work for the Department of Redundancy Department?

15 minutes ago, Tlfj200 said:

Yeah he cuts were complete, but no, not the entire event.

So ‘field’ data is more skewed due to blanks.

there is still stuff to learn, because the field data is ‘too heavy’ since people are more likely to fill out lists if they did better, which is still useful, but we won’t get the universe of lists/breakdowns

There will be a meta presence in the bottom as well, but not to the same extent since some of these lists have a pretty high floor. I don't forget expect it would skew much - maybe some of the close top lists will trade, the percentages overall will be smaller, but I think you will indeed see the same type of lists on top.

Wasn't questioning your work. Just getting an understanding of the data sources to know how to interpret the numbers. It's actually kinda scary though... Of 312 lists in that one month, 50 were Ghost/Fenn. Yes, it's a lower than previous year's monsters... but those took time to get up to speed. Has Ghost/Fenn hit that yet, or is it still building up steam?

Edit: When I typo, I typo whole wrong words...

4 minutes ago, LagJanson said:

There will be a meta presence in the bottom as well, but not to the same extent since some of these lists have a pretty high floor. I don't forget expect it would skew much - maybe some of the close top lists will trade, the percentages overall will be smaller, but I think you will indeed see the same type of lists on top.

Wasn't questioning your work. Just getting an understanding of the data sources to know how to interpret the numbers. It's actually kinda scary though... Of 312 lists in that one month, 50 were Ghost/Fenn. Yes, it's a lower than previous year's monsters... but those took time to get up to speed. Has Ghost/Fenn hit that yet, or is it still building up steam?

Edit: When I typo, I typo whole wrong words...

Agreed.

(andit's cool to question the work. It didn't sound like questioning in a bad way.)

So, to clarify, of the 312 lists that made it to the elimination cut (top 8/16/32/64), 50 were ghost fenn.

Edit:
It's lower than previous year's monsters... so far.