Opening Day is upon us, and it's time to dust off your MLB Predict model (or build your first one) to get ready for over 2,000 games to gamble on throughout the season. This article will walk you through building that first model to hit the ground running. There are things to consider early in the season that will not be reflected well in the AI predictions, such as offseason trades and how the new rules will affect the game such as the new 'shot clock' for pitchers and different sized bases. Outside of that, Pine offers you 4 seasons worth of data for training and testing your model based off the criteria that YOU choose to include. Let's jump right in!
The first step is to log in to your free Pine-Sports account and navigate to the pinned post on the home page. Here, you will click on the 'Predict Winners, ATS & Totals' button followed by 'MLB'. This process is the same for all sports.
.png)
Now your Predict Dashboard for MLB will show up. Once you have models built, this is where you will navigate to either your Moneyline, Spread, or Game Total model but since we are building our first one, we want to click on 'Build a New Model'
.png)
When building your model, you will see a few fields that you need to fill out. There are no right or wrong answers for the most part but let's look at what I am choosing and why.
.png)
1) What do you want to name your model?
- This can be anything. If you ever accidentally replace a model that you want to recover, just send us a message in support chat with your Pine username and what the model's name was and we can attempt to recover it for you.
2) What do you want to predict?
- Again, this can be any one of the three options (Winner, Spread Winner, Total Score) but for this tutorial we will be using the 'Winner' option.
3) How many season back do you want to go to teach your model? And how many games back would you lie to look when comparing teams' stats?
- Once again, no right or wrong answer here. I like to use the most data I can. This allows the AI to learn as much as it can about general tendencies in the MLB that lead to a team winning the game and use that knowledge in generating it's set of 'rules' for predicting future games. For the number of games back, I like to use a number divisible by three specifically for MLB. It likely has to do with the fact that most series in MLB are three-game series, but for some reason it performs slightly better using a games-back number divisible by three.
4) Lastly, Choose the stats you would like to include in your model.
- And again, no right or wrong answer here (sort of?). I used the correlate feature found in 'Explore' to come up with these 9 stats for my model. If you are unfamiliar with correlate, be sure to take a look at our Pine-Sports User Manual by following this link: https://docs.google.com/document/d/1hgQDHcUufSnNW6ppoDYITQVdtnMKJO68/edit?usp=sharing&ouid=101378810671298694336&rtpof=true&sd=true
- Additionally, I advise to never include factors such as opening moneyline, implied odds, or anything relating to what the books are setting as favorite/underdog. The reason for this is that the model will always side with a books favorite and you will see prediction percentages that do no deviate much from implied odds. That means that your predictions are not being based primarily off of game stats. I only use game stats in my choices.
Now that you have input everything, BUILD YOUR MODEL! Here is how ours turned out.
First off, a 57.52% model score is not bad and not the best. However that does not mean a whole lot. If the outcome of 55% predictions and 75% predictions ALL result in a 57.5% win rate than it isn't really helping you, however that is not the case for ours. Click on the little green circle next to model score to dig a little deeper.
This screen shows us different win rates at different confidence levels. We know that overall it gets 57.5% of predictions correct, but let's learn how it does in different segments. If we put in 60% on the confidence level option, our win rate increases to 64.82%, correctly predicting 643 of the 992 games that had a confidence level of at least 60%. Then if we put in 70% for confidence level, we see a win rate increase to 69.7%. These clear breaks are one of the more important factors for the model, because now if I see a confidence level of 70% throughout the season, I can trust that it is a 70% win rate and as long as the implied odds are lower than that then I will have a positive expected value.
Hopefully this helps get you started for Opening Day tomorrow! In order to view upcoming predictions, just navigate to your MLB Predict Dashboard, then click 'Winner', and then 'Get Latest Predictions' at the top! A new window will pop up with all the upcoming games, starting pitchers when available, and your prediction percentages. It's highly encouraged that you play around with games back, or the stats you are including. The best MLB winner model on Pine-Sports is currently running a 60.65% Model Score so see if you can beat it and share those predictions on Pine!
Opening Day is upon us, and it's time to dust off your MLB Predict model (or build your first one) to get ready for over 2,000 games to gamble on throughout the season. This article will walk you through building that first model to hit the ground running. There are things to consider early in the season that will not be reflected well in the AI predictions, such as offseason trades and how the new rules will affect the game such as the new 'shot clock' for pitchers and different sized bases. Outside of that, Pine offers you 4 seasons worth of data for training and testing your model based off the criteria that YOU choose to include. Let's jump right in!
The first step is to log in to your free Pine-Sports account and navigate to the pinned post on the home page. Here, you will click on the 'Predict Winners, ATS & Totals' button followed by 'MLB'. This process is the same for all sports.
Now your Predict Dashboard for MLB will show up. Once you have models built, this is where you will navigate to either your Moneyline, Spread, or Game Total model but since we are building our first one, we want to click on 'Build a New Model'
When building your model, you will see a few fields that you need to fill out. There are no right or wrong answers for the most part but let's look at what I am choosing and why.
1) What do you want to name your model?
- This can be anything. If you ever accidentally replace a model that you want to recover, just send us a message in support chat with your Pine username and what the model's name was and we can attempt to recover it for you.
2) What do you want to predict?
- Again, this can be any one of the three options (Winner, Spread Winner, Total Score) but for this tutorial we will be using the 'Winner' option.
3) How many season back do you want to go to teach your model? And how many games back would you lie to look when comparing teams' stats?
- Once again, no right or wrong answer here. I like to use the most data I can. This allows the AI to learn as much as it can about general tendencies in the MLB that lead to a team winning the game and use that knowledge in generating it's set of 'rules' for predicting future games. For the number of games back, I like to use a number divisible by three specifically for MLB. It likely has to do with the fact that most series in MLB are three-game series, but for some reason it performs slightly better using a games-back number divisible by three.
4) Lastly, Choose the stats you would like to include in your model.
- And again, no right or wrong answer here (sort of?). I used the correlate feature found in 'Explore' to come up with these 9 stats for my model. If you are unfamiliar with correlate, be sure to take a look at our Pine-Sports User Manual by following this link: https://docs.google.com/document/d/1hgQDHcUufSnNW6ppoDYITQVdtnMKJO68/edit?usp=sharing&ouid=101378810671298694336&rtpof=true&sd=true
- Additionally, I advise to never include factors such as opening moneyline, implied odds, or anything relating to what the books are setting as favorite/underdog. The reason for this is that the model will always side with a books favorite and you will see prediction percentages that do no deviate much from implied odds. That means that your predictions are not being based primarily off of game stats. I only use game stats in my choices.
Now that you have input everything, BUILD YOUR MODEL! Here is how ours turned out.
First off, a 57.52% model score is not bad and not the best. However that does not mean a whole lot. If the outcome of 55% predictions and 75% predictions ALL result in a 57.5% win rate than it isn't really helping you, however that is not the case for ours. Click on the little green circle next to model score to dig a little deeper.
This screen shows us different win rates at different confidence levels. We know that overall it gets 57.5% of predictions correct, but let's learn how it does in different segments. If we put in 60% on the confidence level option, our win rate increases to 64.82%, correctly predicting 643 of the 992 games that had a confidence level of at least 60%. Then if we put in 70% for confidence level, we see a win rate increase to 69.7%. These clear breaks are one of the more important factors for the model, because now if I see a confidence level of 70% throughout the season, I can trust that it is a 70% win rate and as long as the implied odds are lower than that then I will have a positive expected value.
Hopefully this helps get you started for Opening Day tomorrow! In order to view upcoming predictions, just navigate to your MLB Predict Dashboard, then click 'Winner', and then 'Get Latest Predictions' at the top! A new window will pop up with all the upcoming games, starting pitchers when available, and your prediction percentages. It's highly encouraged that you play around with games back, or the stats you are including. The best MLB winner model on Pine-Sports is currently running a 60.65% Model Score so see if you can beat it and share those predictions on Pine!
Win your way.
Your AI model belongs to you.