Class GaussianProcessFitDetectorOptions
Configuration options for the Gaussian Process Fit Detector, which analyzes model fit quality using Gaussian Process regression to detect overfitting, underfitting, and uncertainty issues.
public class GaussianProcessFitDetectorOptions- Inheritance
-
GaussianProcessFitDetectorOptions
- Inherited Members
Remarks
Gaussian Process Fit Detector uses Gaussian Process regression to analyze the residuals (differences between predicted and actual values) of a model. By examining patterns in these residuals, it can detect issues like overfitting (model captures noise rather than true patterns), underfitting (model fails to capture important patterns), and areas of high uncertainty.
For Beginners: Think of this as a quality control tool that examines how well your model's predictions match the actual data. It uses a special technique called Gaussian Process regression that's particularly good at detecting patterns and measuring uncertainty. This detector looks at the errors your model makes and determines whether they seem random (which is good) or show patterns (which suggests problems). It can tell you if your model is too complex and memorizing the data instead of learning general patterns (overfitting), or if it's too simple and missing important patterns (underfitting). It can also identify areas where your model is particularly uncertain about its predictions.
Properties
GoodFitThreshold
Gets or sets the threshold for considering model fit as good based on normalized residual patterns.
public double GoodFitThreshold { get; set; }Property Value
- double
The good fit threshold, defaulting to 0.1 (10%).
Remarks
When the normalized measure of pattern strength in residuals is below this threshold, the model is considered to have a good fit. Lower values indicate more random residuals, suggesting the model has captured the true patterns in the data without fitting to noise.
For Beginners: This setting determines when your model is considered to be "just right" - not overfitting or underfitting. With the default value of 0.1, if the detector finds that the patterns in your model's errors are less than 10% of what would be expected by chance, it considers your model to have a good fit. Think of it like checking if the mistakes your model makes appear random rather than systematic. Random errors suggest your model has learned the true patterns in the data without memorizing the noise. If you want to be more strict about what counts as a good fit, you could lower this threshold (e.g., to 0.05).
HighUncertaintyThreshold
Gets or sets the threshold for considering prediction uncertainty as high.
public double HighUncertaintyThreshold { get; set; }Property Value
- double
The high uncertainty threshold, defaulting to 0.5 (50%).
Remarks
When the normalized uncertainty measure exceeds this threshold, the model's predictions are considered to have high uncertainty. High uncertainty suggests the model lacks confidence in its predictions, which may indicate insufficient training data in certain regions or inherent complexity in the relationship being modeled.
For Beginners: This setting determines when your model's predictions are considered highly uncertain. With the default value of 0.5, if the uncertainty in predictions exceeds 50% of a reference level, the model is considered to have high uncertainty. Think of it like a weather forecast that says "temperatures between 65-90°F tomorrow" - it's giving a very wide range because it's not confident in a specific prediction. High uncertainty often occurs in regions where you have little training data or where the relationship is inherently complex and variable. If you want to be more sensitive to detecting uncertain predictions, you could lower this threshold.
LengthScale
Gets or sets the length scale parameter for the Gaussian Process kernel.
public double LengthScale { get; set; }Property Value
- double
The length scale, defaulting to 1.0.
Remarks
The length scale controls how rapidly the correlation between points decreases with distance. Smaller values make the Gaussian Process more sensitive to small-scale variations, while larger values make it focus on broader trends. This parameter affects how the detector interprets patterns in the residuals.
For Beginners: This setting controls how far the detector looks when identifying patterns in your model's errors. With the default value of 1.0, the detector uses a standard distance for determining whether nearby points should be related. Think of it like adjusting the focus on a camera - a smaller length scale (like 0.5) focuses on fine details and local patterns, while a larger length scale (like 2.0) blurs the details and focuses on broader trends. If your data has rapid changes or fine structure, a smaller length scale might be appropriate. If your data changes more gradually, a larger length scale might work better.
LowUncertaintyThreshold
Gets or sets the threshold for considering prediction uncertainty as low.
public double LowUncertaintyThreshold { get; set; }Property Value
- double
The low uncertainty threshold, defaulting to 0.1 (10%).
Remarks
When the normalized uncertainty measure is below this threshold, the model's predictions are considered to have low uncertainty. Low uncertainty suggests the model is confident in its predictions, which is desirable if the predictions are also accurate.
For Beginners: This setting determines when your model's predictions are considered highly confident. With the default value of 0.1, if the uncertainty in predictions is less than 10% of a reference level, the model is considered to have low uncertainty. Think of it like a weather forecast that confidently predicts exactly 78°F tomorrow - it's making a very specific prediction with little hedging. Low uncertainty is generally good if the predictions are also accurate, but can be problematic if the model is confidently wrong. If you want your model to be more cautious about claiming high confidence, you could lower this threshold.
NoiseVariance
Gets or sets the assumed noise variance in the data.
public double NoiseVariance { get; set; }Property Value
- double
The noise variance, defaulting to 0.1.
Remarks
This parameter represents the expected level of random noise in the data. Higher values assume more inherent randomness, making the detector more tolerant of small residuals. Lower values assume less noise, making the detector more sensitive to even small patterns in the residuals.
For Beginners: This setting tells the detector how much random noise to expect in your data. With the default value of 0.1, the detector assumes a moderate level of random fluctuation in your measurements. Think of it like listening for a conversation in a noisy room - if you expect a lot of background noise (high noise variance), you won't be concerned by small variations in what you hear. If you expect silence (low noise variance), even tiny sounds will seem significant. If your data comes from precise measurements with little random error, you might lower this value. If your data is naturally noisy with lots of random fluctuation, you might increase it.
OverfitThreshold
Gets or sets the threshold for detecting overfitting based on residual patterns.
public double OverfitThreshold { get; set; }Property Value
- double
The overfit threshold, defaulting to 0.2 (20%).
Remarks
When the normalized measure of high-frequency pattern strength in residuals exceeds this threshold, the model is considered to be overfitting. Overfitting occurs when the model captures noise in the training data rather than just the underlying patterns, leading to poor generalization.
For Beginners: This setting helps identify when your model is too complex and has started memorizing the training data instead of learning general patterns. With the default value of 0.2, if the detector finds strong high-frequency patterns in your model's errors that exceed 20% of a reference level, it flags potential overfitting. Think of it like a student who memorizes specific test questions instead of understanding the concepts - they'll do well on questions they've seen before but poorly on new questions. An overfitting model shows distinctive error patterns because it's trying too hard to fit every little fluctuation in the training data. If you want to be more sensitive to overfitting, you could lower this threshold.
UnderfitThreshold
Gets or sets the threshold for detecting underfitting based on residual patterns.
public double UnderfitThreshold { get; set; }Property Value
- double
The underfit threshold, defaulting to 0.3 (30%).
Remarks
When the normalized measure of low-frequency pattern strength in residuals exceeds this threshold, the model is considered to be underfitting. Underfitting occurs when the model is too simple to capture important patterns in the data, leading to systematic errors.
For Beginners: This setting helps identify when your model is too simple and missing important patterns in the data. With the default value of 0.3, if the detector finds strong low-frequency patterns in your model's errors that exceed 30% of a reference level, it flags potential underfitting. Think of it like using a straight line to approximate a curve - the line will systematically overestimate in some regions and underestimate in others, creating visible patterns in the errors. An underfitting model makes systematic mistakes because it lacks the complexity to capture the true relationships in the data. If you want to be more sensitive to underfitting, you could lower this threshold.