MLB FORUM

Comments

  • icemanyeee

    I see Derek Carty posted about THE BAT today, i’m thinking about signing up when that becomes available.
    to anyone whose used this last year, what does it consist of? do i need to purchase the lineup builder too?
    what were you’r thoughts on it? or should i look for something else

    thanks

  • DerekCarty

    ESPN, RotoGrinders - Creator of The BAT, MLB Scout School Grad, LABR Champ

    @CUTiger81 said...

    I can definitely understand your reasoning for that. I’m sure you would have some users who get off the a bad start and think it’s not worth it. For me personally, I’ve played enough and trust my process enough not to overreact to a bad month much less a bad week but I do get that most people want immediate results. Just to offer a counter to think about though – I know for me at least, especially with DFS if I get used to using a product and then don’t have it at my disposal anymore I’m likely to just go ahead and purchase it because now that I’ve used it for a week it’s become a part of my routine. I’m sure there’s some downside to it as though so I don’t think you’re crazy for not offering a trial but I know for me it would be useful.

    Do you have a screenshot of the batter and pitcher pages you could post? I would love to see what the data is that’s feeding into the model. My other question is, and I ask this because I know you’re a believer in large samples, what sort of sample are you using to feed into the outlook for a player in a given matchup? For example….if the Angels are facing Sean Manaea on Opening Day, what kind of sample size is feeding into Albert Pujols splits vs lefties? I use him as an example because his splits vs lefties have been so drastically different the last two years so is THE BAT using just the 150ish PA’s vs lefties in 2017 or is it also factoring in the PA’s from 2016 and maybe earlier?

    Thank you for the counterpoint. I like knowing how people think about these kinds of things, and so feedback like that is great. I’ll definitely keep it in mind and weigh it when considering doing a trial.

    Screenshots: there are some in the tutorial video: https://www.youtube.com/watch?v=w2RptDsJCqc I can get actual picture screenshots if you need, but I’m not sure how to post them here, so let me know if that works for you first.

    In terms of what feeds into the model, a lot of it doesn’t get displayed. It’s a lot of proprietary data and processes, and subscribers get access to the end result. I’m always happy to answer questions, though, so that people understand what goes into it and are able to make the best use of the system.

    I love your question about sample size, and the answer is more complex than most would imagine. Each factor that goes into THE BAT (umpires, catchers, platoon splits, etc.) gets its own unique projection system. It’s never as simple as just looking at the stats from this year or last year. THE BAT is always trying to distinguish performance over a particular sample from actual, underlying talent. It has career data for every player/ump/etc. for the last 50 years or so, and it runs tests to see how much data is worth using and in what proportion. The current season will always get the most weight, the one before the next most, etc. The system will test exactly how much weight and how many previous years worth of data leads to the highest level of accuracy. And it does this for each individual stat. As a basic example (just making up numbers), maybe for strikeouts the optimal number is 3 years with each year being weighed 10% less than the year before. But maybe for home runs the optimal number is 5 years with each year being weighed 50% less. The number of plate appearances also impacts the weights, so even though 2018 will receive the most weight this year, after the first game when everyone only has 4 plate appearances, it will actually receive the least weight at that point. Weighting is a combination of sample and recency.

    In your Pujols example, it’s important to note that regression is also important. Platoon splits for RHB have a lot of noise, so they tend to use data that stretches pretty far back and they tend to regress pretty heavily, because that’s what optimizes accuracy.

  • DerekCarty

    ESPN, RotoGrinders - Creator of The BAT, MLB Scout School Grad, LABR Champ

    @Shadetree42 said...

    Derek, Can you go over how much (if at all) you have changed the methodology for this season? E.g., incorporating Statcast, new weather variables, etc.

    I’m always making small, under-the-hood upgrades that should improve performance but aren’t necessarily sexy. I have been doing a lot of research into Statcast type stuff, though, and am planning on incorporating it into THE BAT. The two big things I’ve been developing are hitter projections vs. individual pitch types and pitcher “stuff” grades.

    I can’t guarantee it will be there by Opening Day, but that is the plan. People love these new toys like Statcast, but a lot of times people get so excited that they forget to study them and make sure they are actually useful, or they forget to consider all the ways they can be misleading. For example, people are using pitch type data for hitters semi-frequently now, but nobody has studied whether this data is actually predictive, how much noise they have, etc. Platoon splits have a ton of noise, and lots of people use them incorrectly, so I’ve always been a bit skeptical of this pitch-type data. This winter I dug in, though, and they actually have much less noise than I expected. You still want to account for it and regress, but the regression isn’t too heavy. That being said, there are a ton of other issues that can make the raw data misleading, so I’ve been working on accounting for all of them to truly isolate the hitter’s talent against each pitch. A few of the issues:

    - Scale: It’s not as simple as just going to FanGraphs and seeing which pitch types a guy has a higher wOBA against, though. For one thing, each type has it’s own unique scale. Sliders are the toughest pitch to hit, and so a .350 wOBA against a slider is massively more impressive than a .350 wOBA or even a .400+ wOBA against a fastball. A guy with a .350 wOBA against a slider is a better slider hitter than the guy with a .350 wOBA against a fastball is a fastball hitter.

    - Count: A home run on 0-2 is much more impressive than a home run on 3-1, even for the exact same pitch, and the hitter deserves more credit for it.

    - Plate Discipline: A patient hitter will see more hitters’ counts, more fastballs, and may have his fastball numbers inflated because of it. He’s not elite against fastballs, he just finds himself in favorable counts where he gets easier ones to hit. If you’re looking at the raw stats, they don’t account for this.

    - Handedness: A guy who is bad against sliders and sinkers may not necessarily be bad against sliders and sinkers. Those two pitches have wide platoon splits, so it may just be that this RHB struggles against LHP arm slots. Against a RHP who throws a slider or sinker, he may be completely fine.

    - Parks: Certain pitches, like curveballs, are easier to hit at high altitude. So Coors hitters will have inflated vs. curve numbers using the raw data. If they face a curveball pitcher on the road, we’ll overestimate how well they’ll do.

    - MOST IMPORTANTLY, Game Theory and Advanced Scouting: if a hitters is great against sliders but awful against change-ups, the opposing pitcher will know this. He’ll throw him fewer sliders and more change-ups accordingly, until it reaches a kind of equilibrium. Some pitchers won’t have that option, but a lot will. So even if we’re absolutely certain that this hitter mashes a particular pitch type, it may wind up not even mattering because he’ll see that pitch proportionally less.

    This is my long-winded way of saying that, yes, I’m incorporating new things into THE BAT, but I always want to do things the right way. And for a lot of these new toys that aren’t yet incorporated into THE BAT, doing them the right way can be very complicated, so it takes some time. But that’s also why I believe THE BAT is far and away the best system on the market, because all of the things that already go into it have been handled with the same type of care and thought that others just aren’t doing or aren’t even thinking of.

  • CUTiger81

    Great responses, I appreciate you taking the time to respond thoroughly.

  • Larrioto

    • 142

      RG Overall Ranking

    • Ranked #14

      RG Tiered Ranking

    @DerekCarty said...

    Thank you for the counterpoint. I like knowing how people think about these kinds of things, and so feedback like that is great. I’ll definitely keep it in mind and weigh it when considering doing a trial.

    Screenshots: there are some in the tutorial video: https://www.youtube.com/watch?v=w2RptDsJCqc I can get actual picture screenshots if you need, but I’m not sure how to post them here, so let me know if that works for you first.

    In terms of what feeds into the model, a lot of it doesn’t get displayed. It’s a lot of proprietary data and processes, and subscribers get access to the end result. I’m always happy to answer questions, though, so that people understand what goes into it and are able to make the best use of the system.

    I love your question about sample size, and the answer is more complex than most would imagine. Each factor that goes into THE BAT (umpires, catchers, platoon splits, etc.) gets its own unique projection system. It’s never as simple as just looking at the stats from this year or last year. THE BAT is always trying to distinguish performance over a particular sample from actual, underlying talent. It has career data for every player/ump/etc. for the last 50 years or so, and it runs tests to see how much data is worth using and in what proportion. The current season will always get the most weight, the one before the next most, etc. The system will test exactly how much weight and how many previous years worth of data leads to the highest level of accuracy. And it does this for each individual stat. As a basic example (just making up numbers), maybe for strikeouts the optimal number is 3 years with each year being weighed 10% less than the year before. But maybe for home runs the optimal number is 5 years with each year being weighed 50% less. The number of plate appearances also impacts the weights, so even though 2018 will receive the most weight this year, after the first game when everyone only has 4 plate appearances, it will actually receive the least weight at that point. Weighting is a combination of sample and recency.

    In your Pujols example, it’s important to note that regression is also important. Platoon splits for RHB have a lot of noise, so they tend to use data that stretches pretty far back and they tend to regress pretty heavily, because that’s what optimizes accuracy.

    Thanks for these long and clear answers, Derek.

    I was building a model on my own and will probably buy the BAT to compare with my projections day to day.
    While building my model however, I encountered some problems that I wonder how THE BAT goes about solving. Also included here general questions that I’d like to discuss:

    • Innings pitched: How do you go about projecting that statistic. Do you use some sort of pitch count average and adjust for games where the pitcher can be pulled early, if he’s about to bat and close to his average pitch count (in non-DH games), etc.? Also, do you factor in the distribution of pitches resulting from a strikeout, hit, BB, etc.? (for example, will a pitcher that draws a lot of first contact go deeper into the game than one that does not, assuming they both have an average pitch count of 100).

    • Bullpen: When the pitcher is pulled, do you assume the replacing “pitcher” is the entire aggregate bullpen stats of the opposing team? Also, do you factor in % of righties/lefties in the bullpen when reweighting the predicted values for 1B rate, 2B rate, etc. of batters vs. the bullpen pitcher.

    • Umpire factors: Do you only use umpire data on obvious statistic affected by umpires such as strikeout and BB or use them on all? (Since for HR for example, if hitters have favorable counts they are more likely to hit a HR). Do you also make an adjustment in your regressions to take into account the park mix of umpires? (If an empire was to ump 50% of his game in Coors Fields, his HR boost would obviously be very high, but not because of his umpire skill, but rather because of the park).

    • General correlations: In continuity with my last point, do you also adjust for covariances between inputs in your model. For example, Chase Field is a hitters park in part because it is in Arizona and we know high temperatures favor batters (up to a certain point where the roof would be closed anyways). So, in your model, if you were to treat them independently, there would be some form of double counting of the temperature effect (explicitly in your temperature factor but also implicitly in your park factor).

    • Technical: If you don’t mind me asking, what kind of program do you use to run your simulations? Mines are built in R, but it takes roughly 20 minutes to run on my PC so it leaves a very small amount of leeway to make changes based on new lineup information. I see that in your YouTube video you say that there are new projections every 5 minutes in THE BAT so I guess you must have solved that operational problem.

    • Correlation coefficient: What kind of correlation coefficient do you achieve with your model? For pitchers I get to roughly 0.4 in my backtesting sample (19 july – 5 august 2017), but for batters I have a hard time reaching 0.2 which would make sense since there is more variance for batters…

    Thank you, looking forward to your answers.

  • DerekCarty

    ESPN, RotoGrinders - Creator of The BAT, MLB Scout School Grad, LABR Champ

    @Larrioto said...

    Thanks for these long and clear answers, Derek.

    I was building a model on my own and will probably buy the BAT to compare with my projections day to day.
    While building my model however, I encountered some problems that I wonder how THE BAT goes about solving. Also included here general questions that I’d like to discuss:

    • Innings pitched: How do you go about projecting that statistic. Do you use some sort of pitch count average and adjust for games where the pitcher can be pulled early, if he’s about to bat and close to his average pitch count (in non-DH games), etc.? Also, do you factor in the distribution of pitches resulting from a strikeout, hit, BB, etc.? (for example, will a pitcher that draws a lot of first contact go deeper into the game than one that does not, assuming they both have an average pitch count of 100).

    • Bullpen: When the pitcher is pulled, do you assume the replacing “pitcher” is the entire aggregate bullpen stats of the opposing team? Also, do you factor in % of righties/lefties in the bullpen when reweighting the predicted values for 1B rate, 2B rate, etc. of batters vs. the bullpen pitcher.

    • Umpire factors: Do you only use umpire data on obvious statistic affected by umpires such as strikeout and BB or use them on all? (Since for HR for example, if hitters have favorable counts they are more likely to hit a HR). Do you also make an adjustment in your regressions to take into account the park mix of umpires? (If an empire was to ump 50% of his game in Coors Fields, his HR boost would obviously be very high, but not because of his umpire skill, but rather because of the park).

    • General correlations: In continuity with my last point, do you also adjust for covariances between inputs in your model. For example, Chase Field is a hitters park in part because it is in Arizona and we know high temperatures favor batters (up to a certain point where the roof would be closed anyways). So, in your model, if you were to treat them independently, there would be some form of double counting of the temperature effect (explicitly in your temperature factor but also implicitly in your park factor).

    • Technical: If you don’t mind me asking, what kind of program do you use to run your simulations? Mines are built in R, but it takes roughly 20 minutes to run on my PC so it leaves a very small amount of leeway to make changes based on new lineup information. I see that in your YouTube video you say that there are new projections every 5 minutes in THE BAT so I guess you must have solved that operational problem.

    • Correlation coefficient: What kind of correlation coefficient do you achieve with your model? For pitchers I get to roughly 0.4 in my backtesting sample (19 july – 5 august 2017), but for batters I have a hard time reaching 0.2 which would make sense since there is more variance for batters…

    Thank you, looking forward to your answers.

    Lots of stuff to address here. Some I’ll be a little vague about, because I don’t want to give away too much of the secret sauce recipe for certain things, but I’ll answer what I can.

    The way I project innings is one of the things that I think really sets THE BAT apart, so I can’t give too much detail. I project pitches, and I examine how each projected outcome impacts pitches (walks drive up pitch counts, etc.), managerial tendencies (pulling certain guys early, letting others go deep), and how the pitcher’s projected rate production will impact his leash.

    Aggregate bullpen stats by handedness of pitcher and batter, weighed by recent usage. Eventually I’d like to try and incorporate leverage index based on projected game outcomes, but that’s not there yet.

    Just K/BB for umpires. Actual HR/hit/etc. stats are extremely volatile for umpires, zero predictiveness on the at-bat level. I could approach it a different way and assign ball/strike values to a variety of stats, but I haven’t and don’t think it’s completely necessary. Every factor in my system gets its own projection system, so past data gets neutralized. If an umpire calls 50% of his games in Coors, his projected strikeout rate will likely be higher than his actual strikeout rate. Ump data also gets neutralized for hitter/pitcher/catcher/weather/etc.

    As with umpires, park factors get neutralized for weather, so they become completely independent factors and I can then combine them in the daily layer without any covariance.

    I use SQL

  • idontluvdemhos

    Derek, I always like to purchase the product of people that use their own projections. I notice that you don’t have any screennames in your profile. Do you play DFS yourself? If so, what sites and names? Do you play mostly cash games or tournaments and do you use the BAT or do you use a different projection system? I love the golf product on RG because I see those guys in the higher stakes tournaments and am looking to branch out to MLB and NASCAR. Thanks in advance for your reply.

  • DerekCarty

    ESPN, RotoGrinders - Creator of The BAT, MLB Scout School Grad, LABR Champ

    @idontluvdemhos said...

    Derek, I always like to purchase the product of people that use their own projections. I notice that you don’t have any screennames in your profile. Do you play DFS yourself? If so, what sites and names? Do you play mostly cash games or tournaments and do you use the BAT or do you use a different projection system? I love the golf product on RG because I see those guys in the higher stakes tournaments and am looking to branch out to MLB and NASCAR. Thanks in advance for your reply.

    Yup, I play on DraftKings under the name “DerekCarty”. It’s not hard to find me lol. I’m strictly a cash game guy, so I don’t have my screennames linked because the rankings are much more about GPP play, which isn’t my game. As users of THE BAT know, I use my system as much as anyone in the industry. You’ll usually see me run out the exact optimal lineup for my own play, sometimes with an extra lock or exclude based on a scouting take I have, but I’ll usually talk about when I’m doing this on my pods and in the chat room for subscribers. Hope this helps!

  • divusjulius

    • Blogger of the Month

    this thread makes me very impatient for carty’s weekly articles :)

    those new tweaks seem very very interesting. tempting to see how much of an edge they give over other models

  • trap4

    Derek, I’m a 2-yr subscriber to the BAT, and very much looking forward to see what you’ve got in store for the upcoming season. Thanks for all your hard work. Quick question since you’ve made multiple references to the fact that you play almost exclusively on DK: Do you have any data you can share on the BAT’s baseline performance across the different DFS platforms? Thanks again.

  • DerekCarty

    ESPN, RotoGrinders - Creator of The BAT, MLB Scout School Grad, LABR Champ

    @trap4 said...

    Derek, I’m a 2-yr subscriber to the BAT, and very much looking forward to see what you’ve got in store for the upcoming season. Thanks for all your hard work. Quick question since you’ve made multiple references to the fact that you play almost exclusively on DK: Do you have any data you can share on the BAT’s baseline performance across the different DFS platforms? Thanks again.

    Thanks, I appreciate it :) I don’t really have data like that since there’s no true “right” way to use THE BAT. It’s a projection system that gives an expectation for every player taking the field that day. How DFS players decide to use that information to build their lineups is very much a subjective process, so I’m not even sure how I’d go about measuring that, really.

  • PETERFT73

    Derek would you be willing to post the projections from last season? Then we could interpret the results for ourselves I imagine it would generate a few transactions for you. Even if you just put out August that would be some excellent advertising and give us a chance to (back) test out the service.

  • durbinjag

    I haven’t used that BAT in the past. How well has it done in cash games?

  • sotnohc

    • 653

      RG Overall Ranking

    When is THE BAT available for purchase for this year? Any preseason discount like last year?

  • anirudh

    • Blogger of the Month

    Hey Carty love the BAT, last year was first time I used it. Made a huge difference in my cash builds and my ROI in cash games jumped up. When do you expect the BAT to be available this year?

  • cordscords

    FYI- THE BAT season long projections are now available on fangraphs.

  • bigez952

    @sotnohc said...

    When is THE BAT available for purchase for this year? Any preseason discount like last year?

    The BAT is available to purchase and there is a preseason discount just like last year now until the 28th.

  • poppadom182

    Anyone know how much the preseason discount is?

  • bigez952

    @poppadom182 said...

    Anyone know how much the preseason discount is?

    Last year the full price went up to $175 and the pre season discount was the same at $149 so I would guess it is going to be the same and be about a $25 savings to get it this week.

  • sotnohc

    • 653

      RG Overall Ranking

    @bigez952 said...

    The BAT is available to purchase and there is a preseason discount just like last year now until the 28th.

    Thanks!

  • poppadom182

    Cheers

  • thirdeyevest

    • x2

      Blogger of the Month

    If you are a seasoned veteran or just dipping your toes into MLB DFS, THE BAT is a must! I met Derek at the RotoGrinders party and he sat and talked to me for quite some time about THE BAT, It is truly his baby. The amount of analytics and theory behind the projections is very impressive. The best part, at least to me is Derek’s willingness to help others. If you have a question about something, ask. If you have ideas for ways to improve THE BAT, bring them up. He is very approachable and never acts as if a question is a dumb question.

    In regards to the projections themselves, I find them to be more of CASH focused but I used the projections in conjunction with my own for GPP. Last year was my first time playing MLB daily and I would have never continued without the help of THE BAT. The Pitcher projections were money! I ran correlations throughout the year of various MLB projections sources, including my own; THE BAT was close to .800 while others were lucky to break .650. I’ll take an 80% chance of nailing my Pitcher any day of the week, especially on FD with one pitcher.

    Projections aside, you will learn so much about MLB DFS by purchasing THE BAT.

  • zpruitt3

    how frequently is the bat updated? are the updates automated (ex. weather, lineups, bullpen), so they will always be up to date as the day goes on?

  • bigez952

    @zpruitt3 said...

    how frequently is the bat updated? are the updates automated (ex. weather, lineups, bullpen), so they will always be up to date as the day goes on?

    It is updated automatically on a 5 minute loop I believe as lineups come out and umps are announced. Weather is taking into account except for when it comes to possible rain outs. The BAT will always assume the game is going to play so unless it is officially postponed you still need to watch weather to decide if you want to wait out possibly rain delays or not.

  • timgodd375

    @thirdeyevest said...

    If you are a seasoned veteran or just dipping your toes into MLB DFS, THE BAT is a must! I met Derek at the RotoGrinders party and he sat and talked to me for quite some time about THE BAT, It is truly his baby. The amount of analytics and theory behind the projections is very impressive. The best part, at least to me is Derek’s willingness to help others. If you have a question about something, ask. If you have ideas for ways to improve THE BAT, bring them up. He is very approachable and never acts as if a question is a dumb question.

    In regards to the projections themselves, I find them to be more of CASH focused but I used the projections in conjunction with my own for GPP. Last year was my first time playing MLB daily and I would have never continued without the help of THE BAT. The Pitcher projections were money! I ran correlations throughout the year of various MLB projections sources, including my own; THE BAT was close to .800 while others were lucky to break .650. I’ll take an 80% chance of nailing my Pitcher any day of the week, especially on FD with one pitcher.

    Projections aside, you will learn so much about MLB DFS by purchasing THE BAT.

    Was that correlation only on the pitchers? If so, did you run one for the hitters as well?

  • thirdeyevest

    • x2

      Blogger of the Month

    I did run correlations on hitters but variance on that aspect of the game made the numbers hard to trust. On FD, your Pitcher is so important; that is why i focused on that. THE BAT pointed me in the right direction in terms of “Stacks” when it came to hitters that is for certain.

  • X Unread Thread
  • X Thread with New Replies*
  • *Jumps to your first unread reply

Use our links to sign up and deposit on sites listed in this thread to get these bonuses:

Subforum Index

RotoGrinders.com is the home of the daily fantasy sports community. Our content, rankings, member blogs, promotions and forum discussion all cater to the players that like to create a new fantasy team every day of the week. Our goal is to help all of our members make more money playing daily fantasy sports!

Bet with your head, not over it!
Gambling Problem? Call 1-800-Gambler