Class BayesianFitDetectorOptions
Configuration options for the Bayesian model fit detector, which evaluates how well a model fits the data.
public class BayesianFitDetectorOptions- Inheritance
-
BayesianFitDetectorOptions
- Inherited Members
Remarks
This class provides threshold values used to interpret Bayesian Information Criterion (BIC) or similar Bayesian metrics that assess model fit. These thresholds help determine if a model is a good fit, overfit (too complex), or underfit (too simple) for the given data.
For Beginners: When building AI models, it's important to know if your model is "just right" for your data. Think of it like Goldilocks choosing a bed - one can be too soft (overfit), one too hard (underfit), and one just right (good fit). This class helps set the thresholds for determining which category your model falls into.
An overfit model is like memorizing exam answers without understanding the concepts - it works perfectly for the practice questions but fails on the actual exam. An underfit model is too simple, like using a straight line to predict stock prices that go up and down. A good fit balances complexity and generalization, capturing the important patterns without getting distracted by random noise in the data.
Properties
GoodFitThreshold
Gets or sets the threshold for determining a good model fit.
public double GoodFitThreshold { get; set; }Property Value
- double
The good fit threshold value, defaulting to 5.
Remarks
This threshold is used to identify models that have an appropriate balance between complexity and fit to the data. Models with Bayesian metric values around this threshold are considered to have a good fit.
For Beginners: This value (default: 5) represents the sweet spot for your model's performance. When your model's score is close to this number, it suggests your model has found the right balance - it's capturing the important patterns in your data without being either too simple or too complex. Think of it like the "just right" porridge temperature in Goldilocks - not too hot, not too cold. Models with scores near this threshold are likely to perform well not just on your training data but also on new, unseen data.
OverfitThreshold
Gets or sets the threshold for detecting model overfitting.
public double OverfitThreshold { get; set; }Property Value
- double
The overfit threshold value, defaulting to 10.
Remarks
Bayesian metric values above this threshold suggest that the model may be overfitting the data, meaning it is too complex and may be capturing noise rather than true patterns.
For Beginners: When your model's score exceeds this value (default: 10), it's a warning sign that your model might be "memorizing" your training data instead of learning general patterns. This is called overfitting. It's like a student who memorizes specific test questions but can't solve similar problems on the actual exam. An overfit model performs extremely well on the data it was trained on but fails when given new data. If your model exceeds this threshold, consider simplifying it by reducing features or using regularization techniques.
UnderfitThreshold
Gets or sets the threshold for detecting model underfitting.
public double UnderfitThreshold { get; set; }Property Value
- double
The underfit threshold value, defaulting to 2.
Remarks
Bayesian metric values below this threshold suggest that the model may be underfitting the data, meaning it is too simple and fails to capture important patterns.
For Beginners: When your model's score falls below this value (default: 2), it suggests your model is too simple to capture the important patterns in your data. This is called underfitting. Imagine trying to predict house prices using only the number of bedrooms, while ignoring location, size, age, and other important factors. An underfit model makes oversimplified predictions that miss key relationships in the data. If your model falls below this threshold, try adding more features, using a more complex model type, or reducing regularization.