Recommend
 
 Thumb up
 Hide
10 Posts

MLB Showdown» Forums » General

Subject: Pat, 2014 Set? rss

Your Tags: Add tags
Popular Tags: [View All]
Fall Jester
United States
flag msg tools
Pat, any chance you made a 2014 set already? If not, can you post the spreadsheet that you use for creating new cards based off of projections? I was re-reading some of your posts on card creation and i am interested in seeing it in action.
 
 Thumb up
 tip
 Hide
  • [+] Dice rolls
Pat Senechal
Canada
Toronto
Ontario
flag msg tools
mb
Good timing, I was looking at getting that running again, but I learned a few things from last year, so there are some things to tweak.

Last year, I was building from the 00/01 style, but that had issues with pitchers, making it difficult to make pitchers stand out. Here, I'm basing myself off the 04/05 style, which should help there.

I've got a quick baseline done, and I'll try using that on the 2004 season, and see what comes out. If it works, I'll release that, and then try running on 2013 or 2014 projections.

We'll see what happens, but I've got family in town this weekend, and overtime at work during the week, so it might be a bit slow going, but I'll do what I can. It would be nice if I can make the system so that others can help out, too.
 
 Thumb up
 tip
 Hide
  • [+] Dice rolls
John Jurena
United States
Unspecified
Unspecified
flag msg tools
mbmbmbmbmb
Pat - you are not basing hitters on the 2004/2005 style, are you?
The power creep was way too crazy for hitters those years.
 
 Thumb up
 tip
 Hide
  • [+] Dice rolls
Pat Senechal
Canada
Toronto
Ontario
flag msg tools
mb
When I say 00/01 style, or 04/05 style, I mean how the cards tend to be arranged. A 00/01 hitter was more like, OB 8, 3 out, 16+ HR, while an 04/05 hitter might be OB 13, 6 out, 19+ HR. There's more than one way to do a card, and ties go to the card that looks the most like a card of that style.

But no matter what the cards tend to look like, the main goal is the ballpark of accuracy. A .300 hitter will hit .300 against average pitchers, not .300 against Glavine. And vice versa for pitchers. This means games should average out to like, 4-5 runs a game, as you'd expect.

I'll try something this week, and post the results when I have it.

 
 Thumb up
 tip
 Hide
  • [+] Dice rolls
Fall Jester
United States
flag msg tools
So i was looking through your 2013 card set and while I think the batters are done pretty well, I wasnt really happy with the treatment of the pitchers. 3 main things:

1. All pitchers have a 1 PU? Was there data for this, or did you assign it. This is also not representative of how cards looked back when they were in print.
2. No pitcher has a value for HR? same question as above, but more important, imo. A HR of 1 for a crappy pitcher would allow him to have a higher control, actually becoming useful in certain situations, altho a definite gamble.
3. pitching stats are way too similar. Nearly all pitchers have a similar "curve" (if you plot their values in a line). Better pitchers seem to just be scaled versions of the general formula. I'm not sure if this is due to the uniformity of the datasets, but it is a shame to have put in so much work for so many bland pitchers.

I really like your methodology of comparing each player against the league "average" player, I had not thought of that direction. This entire puzzle of creating new cards has interested me for years, including a project I did for an MBA class a few years ago. In that project I used historical data for 1999 (not projected, which I now think was a mistake) and compared those numbers to the 2000 set. The idea being that the designers used the 1999 data to design the set. I set up a neural network program which made connections between the 1999 data and the 2000 cards, then applied those rules to a 2012 season of data to produce new players. It seemed to work pretty well, but I abandoned the project after I finished the class. Maybe it's time I get back into this...

Below is the summary video that I made for the class:
https://www.youtube.com/watch?v=t-gY2DYF8Yo
 
 Thumb up
 tip
 Hide
  • [+] Dice rolls
Pat Senechal
Canada
Toronto
Ontario
flag msg tools
mb
I know. When I worked on them last year, people were talking about the 00/01 style sets, so that's what I tried, to see what would happen. You managed to summarize quite nicely the issues I ran into.

Because pitchers get advantage more often, changes to their card leads to bigger swings, so it makes it harder to be precise. The high OB of 04-05 is like 2 to 1, but the low OB of 00-01 is like, 3 to 1. That explains your item 3.

As for PU, I might not have had G/F rates at the time, since it was a WIP. As for HR, that's a pretty major swing that can lead to some pretty big rounding errors. I set it up so that pitchers could increase 2B or HR, and try to match Slg, and it picked 2B in all cases. Any pitcher bad enough to outright HR on 20 was likely cut from the set.

Since I'm going back to a 04/05 style, these should mitigate those factors a bit.

I'm working on a tool to rate the cards, and I'll try running them on 2013 or the projections. I can let you in when I try, and we can see if the new baselines work any better.

We need to settle on a baseline, but once that's set, we can run it on anything.
 
 Thumb up
 tip
 Hide
  • [+] Dice rolls
Pat Senechal
Canada
Toronto
Ontario
flag msg tools
mb
Just to follow up...

I have started writing the app to create card ratings from stats, but I'll be travelling for work, so I'll move things to the laptop, in case I can get productive. I have a few things to note:

I put more thought on the raw stats vs projections, and I agree with what you're saying, we need to use projections.

Also, I keep talking about baselines. I've got some proposed ones, but the best way to test those is to take a season, run all the cards, and see if they look right. The issue with that is that scoring now is way lower than we've seen in 2000-05. The long ball period is over, and because OBP is lower overall, either outs go up, or pitchers need to win advantage more.

Ultimately what needs to happen is to take the 2005 and 2014 projections, generate a full set from it, and see what comes out, and see what we want to tweak. Remember, my formulas will make the players behave, to within rounding error (and as we saw with 00-01 pitchers, rounding can make things boring), but we need to double check the constraints, since like any model, extreme values will introduce error.

Hopefully I can find some time in the hotel to get a good example out.
 
 Thumb up
 tip
 Hide
  • [+] Dice rolls
Pat Senechal
Canada
Toronto
Ontario
flag msg tools
mb
Ok, a few updates. I wanted to get this out earlier, but between overtime, travel and too much heat, it's hard to get things done.

I've written a utility that will take a batting line (per 400 PA), and methodically try fitting it to a bunch of swing charts and try to find the best one. However, the first example I ran gave me a chart with 9 outs, so I'll need to give it some rules to penalize cards that are too far from baseline.

Tomorrow, I'll plug some preseason projections through the tool and see what comes out. I'm sure I'll spot things to change, but I'll put it out there, and see how we adjust.

Also, since batters are mostly done, modifying it to do pitchers will be easy, and once I pretty it up a bit, I can post it so we can all use it.

I've also been toying with a new way to do pitcher stamina, but we'll see how it goes. I really want to remove the strict Starter/Reliever barrier, because that's not how it works in the majors.
 
 Thumb up
 tip
 Hide
  • [+] Dice rolls
Bruce Gazdecki
United States
Lindsey
Ohio
flag msg tools
badge
mbmbmbmbmb
Sorry to be late to the party. Where can I find the 2013 edition? Or is this limited to certain people? Thanks.

Bruce
 
 Thumb up
 tip
 Hide
  • [+] Dice rolls
Pat Senechal
Canada
Toronto
Ontario
flag msg tools
mb
The only 2013 edition was a prototype I had made in the offseason, the results are on the board, you can find them here: http://boardgamegeek.com/article/11281139#11281139
 
 Thumb up
 tip
 Hide
  • [+] Dice rolls
Front Page | Welcome | Contact | Privacy Policy | Terms of Service | Advertise | Support BGG | Feeds RSS
Geekdo, BoardGameGeek, the Geekdo logo, and the BoardGameGeek logo are trademarks of BoardGameGeek, LLC.