Show Me and Get Started
Select the "Show Me" button to expand more information on the Baseball ML Workbench functionality:
Select and Search Baseball Players
You may select a player using the dropdown or click-into the control and type in your desired player. Note: Players are sorted by years played. Therefore, players with longer careers will appear higher in the results.
Performing What-If Analysis (Rules Engine)
Navigate to the What-If Analysis (Rules) area. Move the slider left or right to change the number of seasons played. Compare this to the actual player's stats & categorization (static categorization above). Note: Any player with less than 500 HRs will be categorized as not making the HOF. Conversely, any player over 500HRs will be categorized as a HOF inductee.
Performing What-If Analysis (Single Machine Learning Model)
Navigate to the What-If Analysis (Single Model) area. Move the slider left or right to change the number of seasons played. Compare this to the actual player's stats & prediction (static prediction above) Note: As you move the slider, you are rceiving a probability rather than a binary categorization. This allows you to quantify the uncertainty of the selected player's HOF induction by picking a 'decision threshhold'
Performing What-If Analysis (Multiple Machine Learning Models)
Navigate to the What-If Analysis (Multiple Models) area. Move the slider left or right to change the number of seasons played. Compare this to the actual player's stats & prediction (static prediction above) Note: There are two models surfaced returning probabilities. As you move the slider left to right, the probablity of being on the ballot should light up green first.
Comparing Multiple Baseball Players
The Baseball ML Workbench is a stateless application. You can launch multiple browser windows side-by-side to compare players. Note: This is not limited to two, you can launch as many windows as necessary.
How can I understand the machine learning models?
In the workbench, only single probabilities were returned from the machine learning models. Many more constructs can be returned giving deeper explanations of your models. Navigate to https://github.com/bartczernicki/BaseballHOFPredictionWithMlrAndDALEX for a further understanding on how you can breakdown Machine Leanring models. Such as prediction breakdowns, most influential features, impact of features as data oscillates etc.