Does anyone play Fan Duel?

[Nox]

Coors field often has rain and higher POP, so it could really hurt if you're not playing those players based on high POP and the game still gets played (hurt me in a couple line-ups last season).

 

By the way, I wanted to get your opinion with regard to what I said earlier about some projections being out of wack due to players having unforseen health or other issues affecting their performance. The players will keep showing up in your optimum projected line-ups but they will really hurt your chances. Do you have any plans to deal with these "anomolies" or you're just going to roll with the projections and such?

 

Other than obvious injury I wouldn't let your brain apply narrative. Especially when it comes to hot/cold. Not falling for hot/cold should be one of your biggest advantages over the market in fact.

[LTR]

Other than obvious injury I wouldn't let your brain apply narrative. Especially when it comes to hot/cold. Not falling for hot/cold should be one of your biggest advantages over the market in fact.

 

I'm not talking about hot and cold though. I'm talking about a players performance dropping off for undisclosed or unforeseen health reasons (or related), and being able to flag that.

[Nox]

I'm not talking about hot and cold though. I'm talking about a players performance dropping off for undisclosed or unforeseen health reasons (or related), and being able to flag that somehow.

 

If the reasons are undisclosed, how can you possibly know it's because of something tangible and not just natural random variation?

[LTR]

If the reasons are undisclosed, how can you possibly know it's because of something tangible and not just natural random variation?

 

That's exactly the question I'm posing, being able to recognize the difference will benefit you greatly in fantasy betting.

 

I hate to pull the experience card but I played 2000+ FD games last season... if I relied on projections solely I would not have finished the season with a profit.

[jays4life19]

This is a fantastic thread. Thanks guys. The insight is awesome.

[jays4life19]

That's exactly the question I'm posing, being able to recognize the difference will benefit you greatly in fantasy betting.

 

I hate to pull the experience card but I played 2000+ FD games last season... if I relied on projections solely I would not have finished the season with a profit.

 

lol. You have over 2000 wins in 1 season so i'm sure you played in a lot more than 2000 games.

[LTR]

lol. You have over 2000 wins in 1 season so i'm sure you played in a lot more than 2000 games.

 

Oops, been a while since I played... so yeah, I guess that would equate to around 3.5K games... yikes

[Nox]

That's exactly the question I'm posing, being able to recognize the difference will benefit you greatly in fantasy betting.

 

I hate to pull the experience card but I played 2000+ FD games last season... if I relied on projections solely I would not have finished the season with a profit.

 

Nobody is saying you should not update your projections/lineups based on new information/observations. That's the whole underpinning of Bayesian inference after all. However you can't apply undisclosed information to such a decision making framework.

 

"I had to pull the experience card but I played 2000+ FD games last season... if I relied on projections solely I would not have finished the season with a profit." - Not to be a complete dick but I really find this claim dubious. Did you log every time you overrode your projection system? Did you log what the change was? Did you log why you decided to make the change?

 

[LTR]

Nobody is saying you should not update your projections/lineups based on new information/observations. That's the whole underpinning of Bayesian inference after all. However you can't apply undisclosed information to such a decision making framework.

 

"I had to pull the experience card but I played 2000+ FD games last season... if I relied on projections solely I would not have finished the season with a profit." - Not to be a complete dick but I really find this claim dubious. Did you log every time you overrode your projection system? Did you log what the change was? Did you log why you decided to make the change?

 

 

That's fair, I can't validate how many more games I would have lost (or won for that matter) if I relied solely on the projected optimum line-up. It's a good experiment for this year, would need a significant sample size to have any credibility. I will say that my line-ups were roughly 95% based on projections with the other 5% being a result of "Bayesian inference".

[nmrch]

That's exactly the question I'm posing, being able to recognize the difference will benefit you greatly in fantasy betting.

 

I hate to pull the experience card but I played 2000+ FD games last season... if I relied on projections solely I would not have finished the season with a profit.

 

Or it could also lead you to consistently play suboptimal lineups when you put your finger on the model. To be fair this will be the first season i'm gonna be relying completely on my model and i'm pretty new to DFS overall, so you your experience might be saying something.

[Nox]

I will say that my line-ups were roughly 95% based on projections with the other 5% being a result of "Bayesian inference".

 

This seems like a good ratio to me.

[LTR]

Or it could also lead you to consistently play suboptimal lineups when you put your finger on the model. To be fair this will be the first season i'm gonna be relying completely on my model and i'm pretty new to DFS overall, so you your experience might be saying something.

 

I started out thinking the same way. I wanted to rely on the projection system alone and have a well oiled machine working for me but it just didn't work that way. But good luck, I'm really enthused to see so many people playing and will be curious to see how everyone does, especially with a few of you using your own custom projection systems.

 

Also would be fun to face off with everybody, doesn't even have to be money games, can just play for bragging rights (though anyone can win in small samples).

[jays4life19]

I started out thinking the same way. I wanted to rely on the projection system alone and have a well oiled machine working for me but it just didn't work that way. But good luck, I'm really enthused to see so many people playing and will be curious to see how everyone does, especially with a few of you using your own custom projection systems.

 

Also would be fun to face off with everybody, doesn't even have to be money games, can just play for bragging rights (though anyone can win in small samples).

 

Yeah it's going to be awesome. It kinda sucked last year that it was really hard to get more than me + you into games last year. Looks like it's going to be easy to get BJMB games this year.

[nmrch]

I started out thinking the same way. I wanted to rely on the projection system alone and have a well oiled machine working for me but it just didn't work that way. But good luck, I'm really enthused to see so many people playing and will be curious to see how everyone does, especially with a few of you using your own custom projection systems.

 

Also would be fun to face off with everybody, doesn't even have to be money games, can just play for bragging rights (though anyone can win in small samples).

 

I'd be in for facing any of you guys for a dollar everynight, the rest of my meager bankroll i'll use it in the public leagues. Playing just for bragging rights would be great too, we can use this thread to post lineups (after the lineup deadline for those of us who are not as generous as Jfas) and keep track of results.

 

On that note, i'm thinking of using the three most optimal lineups(with different %es of bankroll invested based on the most efficient to the third most efficient) instead of just one to reduce variance. For all the veterans out there, that makes sense right? I'll find my true win rate faster in exchange for slightly fewer wins, if my model is crap i'll find out sooner and then i can either fix it or get out of the game.

[NorthOf49]

Yeah it's going to be awesome. It kinda sucked last year that it was really hard to get more than me + you into games last year. Looks like it's going to be easy to get BJMB games this year.

I've played in a couple basketball 50/50s and lost both, one badly. Think I'm going to wait for baseball to start

[nmrch]

Here's an interesting graph i came up with poking around on Excel trying to figure out how to properly project runs. I didn't think speed has this much of an effect on the rates at which players score runs. There's probably an even bigger effect if you take out doubles

 

http://s4.postimg.org/swug6uxwd/Untitled.jpg

Edited March 17, 2015 by nmrch

[LTR]

Fangraphs doesn't include pitcher batting projections (NL games). So my starting pitcher projections are a bit lacking if I can't fully evaluate the opposing batting line-up. Right now I'm just rounding off the pitcher spot with league average offensive stat for pitchers but this doesn't seem ideal (pitchers are usually pretty negligible offensively but there are some that are decent).

 

Any suggestions?

[Nox]

With more time I will project pitcher hitting myself.

 

I feel like actually projecting individual pitchers should be pretty low on the priority list. What you have now seems good to me.

[LTR]

Looking for some SQL advice with my line-up optimizer.

 

With up to 15 games / 30 players per position, there are 1,237,940,039,285,380,000,000,000,000 (8^30) possible line-up permutations (my pitcher spot is done seperately). This number gets squashed when I use a baseline for total points (greater than) and total dollar value (less than or equal to).

 

The problem I’m having is I can only run the query on the top 10-12 or so players from each position… after that the SQL engine breaks.

 

Any tips?

[LTR]

My SQL optimizer takes under 2 seconds.

 

First, remove all players that have a player with more points than them and less salary at their position. For OF change this to 3 players with more points and less salary. For example for DK today that leaves me with 6 1B, 4 2B, 6 3B, 6 SS, 5 C, 18 OF, 10 SP (need to have less than or equal to 2 better and cheaper than them). A player that is more expensive and worse than another option will never show up in an optimal.

 

Only have one join per query. Nesting queries is much much faster. Group by points at the OF stage to eliminate all the abc cba acb bca bac cab possibilities.

 

Damn, seems so obvious now... thanks a ton mate, this will go a long way.

[Nox]

Damn, seems so obvious now... thanks a ton mate, this will go a long way.

 

What JFas said.

 

Or look to do something in the realm of non-deterministic optimization. This will take a bit longer to converge (10 seconds+ depending on specific code and hardware) but alot of these techniques will give you more flexibility when you're not just trying to maximize mean points.

[LTR]

JFaS. I compared our optimal line-ups yesterday and today... yesterday was basically identical (might have been one difference), but today leaves some question marks. I selected the same pitcher for sanity testing purposes.

 

Here's my optimal line-up for today (top 2):

 

http://i.imgur.com/xIOsMx4.gif

 

I'm using a combination of ZiPS and Steamer (basically the same as FGDC) by the way.

 

Namely three differences:

 

- 2B: Dan Uggla (vs. Scooter Gennett)

 

This one is weird because Dan Uggla f***ing sucks BUT even the most basic projections * PF * Opposing Pitcher have Dan Uggla as the far and away best value for dollar. Therefore, I don't see how he's not in your line-up unless you used some non-standard method here. Also Scooter Gennett in terms of value came up totally middle of the pack for me so I don't get it (he was facing Jeff Samardzija today). And this goes back to what I was talking about in terms of projections - would you really ever play Dan Uggla knowing what you know about him, regardless of projections/value? I would not.

 

- OF Hanley Ramirez (vs. Jay Bruce)

 

Jay Bruce would have been the next in line so no real concern here but I'm wondering if you're taking your fielding positions from Fanduel itself since Hanley is not normally an OF?

 

- C: John Jaso (vs. Alex Avila)

 

Don't see how you got Avila here, too lazy to comment further.

[LTR]

OK, I kind of figured out what happened with Uggla.

 

He's not showing up in the main Steamer projections or the FGDC projections, therefore I was only using ZiPS (and you might not have had him in your FGDC projections but I'm still surprised he wasn't in your ZiPS one). ZiPS is also favoring him more on his player page compared to Steamer (my system falls back to one projection system if both are not available).

[LTR]

Don't have the data anymore, since it's doing tomorrow now. I'm not storing it yet either. One simple explanation might be lineup position and the guys around them. Gennett had Lucroy and Braun behind him. Uggla surrounded by s***. Uggla isn't very split either, while Gennett is.

 

Thanks. I haven't applied that method to my projections yet (easy enough to do but still giving it some thought).

[LTR]

I'm finding FanGraphs Depth Charts unusable for my purpose. They combine ZiPS + Steamer but remove any records that don't exist in both. Which might makes sense from their perspective, but I have 20+ players with projections unaccounted for in yesterdays matchups (none unaccounted for if I'm merging both Steamer and ZiPS myself). Edited March 26, 2015 by LTR

[LTR]

That's what Im doing, I'm jus calling it FGDC though. It's the same thing just all inclusive.

 

Word. Would just be easier to have to import one table instead of two (or two instead of four, since pitchers and batters).

[nmrch]

I'm finding FanGraphs Depth Charts unusable for my purpose. They combine ZiPS + Steamer but remove any records that don't exist in both. Which might makes sense from their perspective, but I have 20+ players with projections unaccounted for in yesterdays matchups (none unaccounted for if I'm merging both Steamer and ZiPS myself).

 

But how many players on MLB lineups won't have Depth Charts projections though?

[LTR]

But how many players on MLB lineups won't have Depth Charts projections though?

 

Too many it seems.

[nmrch]

JFas, what kind of thresholds are you using to regress platoon splits on things like hitter K%, hitter ISO, pitcher K%, pitcher BB%, pitcher GB% etc

 

My overall regressed splits look good(wOBA for hitters and xFIP for pitchers) but i'm just wondering what thresholds to use for the sub components.

[GD]

JFas, what kind of thresholds are you using to regress platoon splits on things like hitter K%, hitter ISO, pitcher K%, pitcher BB%, pitcher GB% etc

 

My overall regressed splits look good(wOBA for hitters and xFIP for pitchers) but i'm just wondering what thresholds to use for the sub components.

 

Is this helpful? http://www.fangraphs.com/library/principles/sample-size/