riding with you rsigley

I don't like the georgiev, riding him in cash specifically for save equity

After some thought working on the rest of the hockey props including goalie saves and Yahoo DFS cash game rosters because competition makes everyone better

TomG you can make money in NHL DFS once you put the equation together. All the major sites (e.g., daily roto, fantasy labs) have horrible coverage for these sports. It's probably the biggest sport that isn't destroyed by optimizers and 17 people doing a podcast before lock with all the obvious picks.

Georgiev was looking good until he got 6 saves in 30 seconds with 2 minutes left in the 3rd

How do you model for that

Still got it, barely though!

crawford holds thru OT!!!

lucky too since Koskinen is gonna have 30 saves in the 1st period

100% why I chose to aggressively target hockey

I can disagree with Sigs, be very wrong and yet still have a fantastic slate because everyone else is just clicking buttons

TomG as much as I hate sharing 3 of the payout slots in the 6 man groups with you, welcome back

P.P.S. why not fantasy draft and their no rake?

BimmyLee Goalie Saves

Koskinen u29.5 LOSS -1.15
Quick o25.5 LOSS -1.15
Gibson o27.5 CANCELLED (Didn't start)
Crawford u29.5 WIN +1
Holtby o26.5 WIN +1
Raanta u30.5 WIN +1
Georgiev u33.5 WIN +1

Yesterday: 4-2 +1.7u
Total: 9-10 -2.35u on 21.85u risked (-10.75% ROI)

little positive today, maybe this is our turn around moment.

Any else like Gary Woodland at +285 to win the Hero World Challenge? I have fair as ~+223. Only got a one stroke lead, but everything looks good.

I've been working on a model for LoL (LCS) DK, and was wondering if anyone else has dipped into esports predictions/modeling?

http://jmcauley.ucsd.edu/cse255/projects/fa15/018.pdf

This is pretty useless. Obv DOTA/LoL are modeled identically. The problem is drafting is such a crapshoot. Pre-match you're basically power ranking players and adjusting for recent form.

Poogs,

Why would I take up a prop bet with someone who I think is a scumbag? Doesn't really make any sense. Aside from the fact that I'm clearly stating that you tried to cheat that guy in that thread. Which you've admitted to multiple times... So not sure what we disagree about?

You're really just mad for the handful of things that are true that I make fun of you about:
1) You tried to stiff a dude on a prop bet.
2) You couldn't beat Greek openers on a hockey prop and frequently bet players who didn't even play.
3) You are a hypocrite who seems to enjoy ridiculing me over pics/vids, but find it weird and creepy when others do it to you.

Thank you for chatting with your magic sky person about me.

rick,

Thanks. That was super informative. Maybe some of the high stakes players have rakeback deals? Seems to make sense that some would use their popularity to run an affiliate and end up in situations where they can afford to take on seemingly 0EV positions due to exogenous factors. (I guess the prizing and freerolls for the biggest winners would also factor in this? No clue if that still exists. Also if you don't actually attend it doesn't have value: RIP Brettfarve444)

Haha funny you bring up that paper since I had actually tried to build a DotA model off that paper earlier and it was totally useless. I guess I also feel like past performance (especially kills which are very important for both Draftkings scoring and match prediction) capture a bit of the draft itself, but maybe not.

Right now I've scraped a bunch of data from OraclesElixir and am trying to predict kills in a game off of a bunch of their statistics (along with some self made ones). It hasnt been overly successful via any sort of regression (tried Lasso and then Ridge) or random forest type stuff (Random Forest and xgBoost). My latest run through a normal random forest was somewhat decent but I think it overfit and the error for the test data was awful.

Also I'm surprised at how many people are playing NHL dfs in here, its my favorite sport to fire at. I assume everyones playing higher stakes than I am but I wrote a script that combines lineup data with analytics data and gives a nice printout of HDCF/60 data (yesterdays below) for the slate if anyone wants it.

JS,

Weird. It is like I googled the subject at some time in the past and actually read this stuff and kinda expect others to have passing familiarity!

I have played a smattering of MOBAs (like an hour of DOTA2 basically and was pretty bad at HOTS, just no drive to sink the 2-3hrs a day to rank like a normal human being) and don't really know the scoring so take it for what its worth, but I think people who play hard carries are gonna be your best bet for most things. I'd shy away from teams that try to experimental strategies. Take a look at how Corsi is derived. He used a kitchen sink approach to ML but ultimately built something that was good enough to win. Sigs has written about some of the major issues that it has, but they're all pretty clear. (Why are you running a bunch of random **** against random data sets, then bagging the whole thing?!?)

I didn't mean anything by it m8 I just thought it was funny that I had worked off of that exact same paper and we both thought it was useless. I've also played a ton of mobas and when I was in college I hit mid diamond (not that that matters) in LoL so I'm def familiar with the game itself.

In terms of the second part of what youve said you're right that carries produce most of the DK fantasy points (distributions: https://imgur.com/a/WAelzB2), but salary is restrictive and each player on the team (except from top lane) is highly correlated with eachother - so it seems like you predict a team that will get a ton of kills/win and try to mash 3-4 players from that team onto a roster.

Also not sure what you mean by random data sets, the data I have is match log data for the entire last season of LCS - and I've tried to filter out the pieces that aren't predictive of the number of kills a team gets. The best approach I've done so far is very similar to corsi, in that the feature set contains differences important things like gold/min, damage/min, towers/game between the two teams. Thats where I'm at right now in terms of the model.

JS, you're on the right track there but unless you're doing MME 150x entries then I would spend less time on the teams you'll never play and more time on fluidity of the lines you would play

if it's all automated and such then whatever, but one thing in dfs is people spend way too much time trying to produce projections for the 2nd string TE when they should be asking themselves "am I ever going to play a second string TE if the first stringer is healthy"

Yeah the NHL thing is automated and I'm not currently building anything off of that, it was more of an aside since I saw a lot of people playing NHL in here. I do imagine putting in a lot of LoL dfs volume if I can produce a decent model though.

Counterpoint:

Daily: 4-6-0, -1.74u
YTD: 152-151-1, -12.1u, 365.61u risked, -3.3% ROI

Early look

OTT Anthony Duclair u2.5 -111
VAN Jake Virtanen u1.5 +132
TBL Victor Hedman u2.5 -167
CBJ Boone Jenner u2.5 -169
TOR Morgan Rielly o2.5 +119

is the DAL/NYI prop market totally weird? it's like the perfect storm of two defensive teams who neither allow many shots nor take them

JS,

I was ranting about something and I was confused. No worries.

Here is a paper I'm reading about penalizing the wiggiliness of splines: https://hal.archives-ouvertes.fr/hal-01853459/document

Maybe useful for you!

Woodland +4 through 4. Classic.

Yeah but even with rakeback, I mean I've actually modeled it assigning equity to people in 10, 6, and 3 man's to figure out at what point it became feasible - don't recall the specifics but my ultimate conclusion was that in order to just beat rake I needed a massive edge over the field or there had to be just terrible players.

Problem with DFS is any terrible player is still going to button mash 30% wins in 50/50s and most competent guys will get more than 40% and if someone even gets 44% against you then you lose on rake

Now that I've improved my math skills a bit I think I should revisit this question and try to get a clearer answer

The real thing about DFS is that there's going to be 5 guys who max reg to every contest so even if your playing for $1 in a 10 man group then you're facing several of your best cash game players in the world - this is why do many people get drained because they're sitting down at plo2 but half the table is there same crushers doing the Phil Galfond challenge

Game selection is huge and thus why in basketball, baseball and football I don't join a group with less than 50 players because only then do I feel there's enough breathing space to make a clear pathway to cashing

any nba pastebin today rsigley?

sure. it's assists and rebounds been killing it this year

assists 17.4% roi over 497 bets
rebounds 12.42% over 674 bets
turnovers and blocks+steal been bad (-4.6% and -2.22%), but luckily those are harder to find and lower limits.

https://pastebin.com/uuVS8VwC

also can i throw my hat into the NHL SoG ring? Maybe by the end of the year we could figure this sport out.

here's my SOG PRojections

https://pastebin.com/cCQf8z1L

and here's what I bet

Olofsson (BUF) o2.5 +100
Aho (CAR) o2.5 -115
Kucherov (TBA) o2.5 -115
MacKinnon (COL) u3.5 +105
Hoffman (FL) o2.5 -105
Zuccarello (MIN) o1.5 -105
Couture (SJS) o2.5 +100
LeTang (PIT) o3 -115
Eichel (BUF) o3.5 -115
Malkin (PIT) u3 +100
Matthew (TOR) u3.5 -110
Athanasiou (DET) o2.5 +115

So model spits out mostly under for goalies and overs for SoG. Makes sense.

When is that ever not the case tbh?