We have shown you how to build your Pine Sports model as well as built our own NBA model on the Pine Twitch, but that doesn't scratch the surface of how deep you can dive! In this edition of the Pine Sports 'How To' series I will show you the different ways to analyze your model results so you may make even better decisions day in and day out.
Start off by going to your Pine Predict dashboard, for my examples today I will be using the NBA Moneyline model that we built together on Stream but the same concept apply to all 12 different model options (NBA, NHL, NFL, & MLB; Moneyline, Spread & Total Score).
The first, quick and easy option for some basic confidence level analysis is by clicking on the green circle next to "Model Score". This popup window is where you can look at the correct prediction rates of different confidence levels. If will show you how many games were back-tested and the correct rate from that sample as well as the full range of confidence levels. If you input a confidence level at the bottom, for example 65%, it will show just those results. For example out of the 3,084 games tested for our NBA model, 1,073 games had a 65% or higher confidence level and 753 (70.18%) of those were correctly predicted.
What we are focusing on today is a little more involved and insightful than just the confidence level win rates. A little 'Easter egg' built into Pine is the model score analysis. Click on your model score, the actual percentage listed located in the red box below, and it will take you into the Model Score Analysis Explore page specific to your model.
.png)
Here, you will see the same screens that you are accustomed to seeing in Explore with the View page open and options to Visualize, Correlate, or Select. As with Explore, all of the column headers are adjustable. This selection of data includes any backtested games from your model, and games from the current season. You will see normal columns like Home, Visitor, Final Score Spread and Total Scores as well as some new columns such as the Winner Predictions, Probability of Winner, and a final column called 'Correct' which labels whether or not the prediction was in fact correct. Let's jump straight into Visualize and start looking at different ways you can manipulate this data to find strengths or weaknesses of your model!
.png)
Now the fun begins. There are all kinds of things you can do here and I will touch on a couple of my favorite things first but use your imagination. For starters, let's look at a simple comparison of how the model reacts given specific home teams. I am going to go to the 'Bar' chart option and input the following info:
Query: Season Year == '2023'
X= Home
Y= Season Year
Aggregation = Count
Bar Sort= Season Year
Group= Correct
You will now get two charts, a 'Yes' chart and a 'No' chart with each team listed for the count of correct or incorrect predictions when that particular team was the home team. Here are my charts and then I will show you why this can be a valuable analysis.
Count of Correct Predictions when Listed Team is HOME
Count of Incorrect Predictions when Listed Team is HOME
By comparing these two charts we can find some pretty interesting things about our model. Looking at New Orleans first, as they have the most correct predictions as the Home team, we see that the model is a whopping 19-3 record in prediction New Orleans' home games. It's important to realize this does NOT mean it is predicting for New Orleans to win necessarily. It is simply stating that when New Orleans is playing at home, whoever the model predicts to win has been correct 86.4% of the time. Now, the Pelicans are 17-5 at Home this season so that certainly plays a part, but this is still a very big win rate to keep an eye on moving forward.
Compare this go Chicago, who has the fewest correct predictions when the Bulls are the listed Home team at only 6 correct predictions and 13 incorrect predictions. Chicago is 10-9 at Home this season so the correct rate really shouldn't be this bad. To me, this indicates that the Bulls Home games are very unpredictable according to the model and worth potentially avoiding unless a match-up is relatively clear.
Now let's see how we do looking only at Home Underdogs. To do this, we want to adjust a few things about or set up. You might want to just refresh your page for simplicity sake, but clear out the 'Yes' & 'No' from the Groups, and then clear out 'Correct' from Group. Now set up the following:
Query: Season Year == '2023' and Winner - Prediction == 'Home' and Home - Moneyline >= 100
X= Correct
Y= Season Year
Aggregation= Count
This input provides us with the following chart from our model predictions, which shows that whenever the Home team was a moneyline underdog and predicted to win, the model was correct 45 out of 86 times. While the margin here is small, if you wager the same amount on every play knowing that these are all plus odds plays, you can expect a small profit over the long run. Coincidentally, I plugged in >= 300 in the query as well and the model had a 2-1 record when it predicted +300 underdogs to win this season. Althought a small sample size, that's pretty impressive that this model has correctly called out a +300 or greater ML team to win more than once.
There is much more you can do in this area of Pine, for example you could look at something such as 'When Memphis is the Visitor, and <= -200 odds favorites, and Predicted to win, what is their total score trend?" From specifically our model, you would find that this specific example has only occurred 5 times this season and the Grizzlies scored 109 points twice and over 121.5 in the other 3 games. I highly encourage you to get imaginative and really play around with this data to learn further information about your model's performance so you can continue improving your decision making through the rest of the season.
As always, reach out to me here or on Discord if you have any questions!
We have shown you how to build your Pine Sports model as well as built our own NBA model on the Pine Twitch, but that doesn't scratch the surface of how deep you can dive! In this edition of the Pine Sports 'How To' series I will show you the different ways to analyze your model results so you may make even better decisions day in and day out.
Start off by going to your Pine Predict dashboard, for my examples today I will be using the NBA Moneyline model that we built together on Stream but the same concept apply to all 12 different model options (NBA, NHL, NFL, & MLB; Moneyline, Spread & Total Score).
The first, quick and easy option for some basic confidence level analysis is by clicking on the green circle next to "Model Score". This popup window is where you can look at the correct prediction rates of different confidence levels. If will show you how many games were back-tested and the correct rate from that sample as well as the full range of confidence levels. If you input a confidence level at the bottom, for example 65%, it will show just those results. For example out of the 3,084 games tested for our NBA model, 1,073 games had a 65% or higher confidence level and 753 (70.18%) of those were correctly predicted.
What we are focusing on today is a little more involved and insightful than just the confidence level win rates. A little 'Easter egg' built into Pine is the model score analysis. Click on your model score, the actual percentage listed located in the red box below, and it will take you into the Model Score Analysis Explore page specific to your model.
Here, you will see the same screens that you are accustomed to seeing in Explore with the View page open and options to Visualize, Correlate, or Select. As with Explore, all of the column headers are adjustable. This selection of data includes any backtested games from your model, and games from the current season. You will see normal columns like Home, Visitor, Final Score Spread and Total Scores as well as some new columns such as the Winner Predictions, Probability of Winner, and a final column called 'Correct' which labels whether or not the prediction was in fact correct. Let's jump straight into Visualize and start looking at different ways you can manipulate this data to find strengths or weaknesses of your model!
Now the fun begins. There are all kinds of things you can do here and I will touch on a couple of my favorite things first but use your imagination. For starters, let's look at a simple comparison of how the model reacts given specific home teams. I am going to go to the 'Bar' chart option and input the following info:
Query: Season Year == '2023'
X= Home
Y= Season Year
Aggregation = Count
Bar Sort= Season Year
Group= Correct
You will now get two charts, a 'Yes' chart and a 'No' chart with each team listed for the count of correct or incorrect predictions when that particular team was the home team. Here are my charts and then I will show you why this can be a valuable analysis.
Count of Correct Predictions when Listed Team is HOME
Count of Incorrect Predictions when Listed Team is HOME
By comparing these two charts we can find some pretty interesting things about our model. Looking at New Orleans first, as they have the most correct predictions as the Home team, we see that the model is a whopping 19-3 record in prediction New Orleans' home games. It's important to realize this does NOT mean it is predicting for New Orleans to win necessarily. It is simply stating that when New Orleans is playing at home, whoever the model predicts to win has been correct 86.4% of the time. Now, the Pelicans are 17-5 at Home this season so that certainly plays a part, but this is still a very big win rate to keep an eye on moving forward.
Compare this go Chicago, who has the fewest correct predictions when the Bulls are the listed Home team at only 6 correct predictions and 13 incorrect predictions. Chicago is 10-9 at Home this season so the correct rate really shouldn't be this bad. To me, this indicates that the Bulls Home games are very unpredictable according to the model and worth potentially avoiding unless a match-up is relatively clear.
Now let's see how we do looking only at Home Underdogs. To do this, we want to adjust a few things about or set up. You might want to just refresh your page for simplicity sake, but clear out the 'Yes' & 'No' from the Groups, and then clear out 'Correct' from Group. Now set up the following:
Query: Season Year == '2023' and Winner - Prediction == 'Home' and Home - Moneyline >= 100
X= Correct
Y= Season Year
Aggregation= Count
This input provides us with the following chart from our model predictions, which shows that whenever the Home team was a moneyline underdog and predicted to win, the model was correct 45 out of 86 times. While the margin here is small, if you wager the same amount on every play knowing that these are all plus odds plays, you can expect a small profit over the long run. Coincidentally, I plugged in >= 300 in the query as well and the model had a 2-1 record when it predicted +300 underdogs to win this season. Althought a small sample size, that's pretty impressive that this model has correctly called out a +300 or greater ML team to win more than once.
There is much more you can do in this area of Pine, for example you could look at something such as 'When Memphis is the Visitor, and <= -200 odds favorites, and Predicted to win, what is their total score trend?" From specifically our model, you would find that this specific example has only occurred 5 times this season and the Grizzlies scored 109 points twice and over 121.5 in the other 3 games. I highly encourage you to get imaginative and really play around with this data to learn further information about your model's performance so you can continue improving your decision making through the rest of the season.
As always, reach out to me here or on Discord if you have any questions!
Your Sports IQ + AI = Pine Sports