Table of Contents

Class NormalizationInfo<T, TInput, TOutput>

Namespace
AiDotNet.Models
Assembly
AiDotNet.dll

Represents information about how data normalization is performed for a model, including the normalizer and parameters.

public class NormalizationInfo<T, TInput, TOutput>

Type Parameters

T

The numeric type used for calculations, typically float or double.

TInput
TOutput
Inheritance
NormalizationInfo<T, TInput, TOutput>
Inherited Members

Remarks

This class encapsulates all the information needed to normalize input features and denormalize predictions for a machine learning model. Normalization is a preprocessing technique that scales data to a standard range, which can improve the performance and stability of many machine learning algorithms. This class stores the normalizer object that performs the actual normalization operations, as well as the parameters that describe how each feature and the target variable were normalized during training.

For Beginners: This class stores information about how data is scaled before and after prediction.

When working with machine learning models:

  • Input data often needs to be scaled to a standard range (like 0-1)
  • After prediction, the results may need to be scaled back to the original range
  • You need to store how this scaling was done during training

This class stores all that information, including:

  • The normalizer object that performs the actual scaling operations
  • Parameters for how each input feature should be normalized
  • Parameters for how the target variable (what you're predicting) should be denormalized

This ensures that new data is processed consistently with how the model was trained, which is essential for accurate predictions.

Properties

Normalizer

Gets or sets the normalizer used to normalize input data and denormalize predictions.

public INormalizer<T, TInput, TOutput>? Normalizer { get; set; }

Property Value

INormalizer<T, TInput, TOutput>

An implementation of INormalizer<T> that performs normalization operations.

Remarks

This property contains the normalizer object that performs the actual normalization and denormalization operations. The normalizer implements the INormalizer<T> interface, which provides methods for normalizing matrices and vectors and denormalizing the results. Different implementations of this interface might use different normalization techniques, such as min-max scaling, z-score normalization, or robust scaling. The property is nullable because it might not be initialized immediately, but it must be set before normalization operations can be performed.

For Beginners: This is the object that actually performs the scaling operations.

The normalizer:

  • Implements the INormalizer interface
  • Contains methods to scale data to a standard range (normalize)
  • Contains methods to convert scaled data back to original values (denormalize)
  • Could use different techniques like min-max scaling or z-score normalization

This property is marked as nullable (with the ? symbol) because:

  • It might not be initialized immediately
  • When deserializing from storage, it might be populated later

However, it must be set before you can normalize or denormalize data, or you'll get an error when trying to use it.

XParams

Gets or sets the normalization parameters for each input feature.

public List<NormalizationParameters<T>> XParams { get; set; }

Property Value

List<NormalizationParameters<T>>

A list of NormalizationParameters<T> objects, one for each input feature.

Remarks

This property contains a list of normalization parameters, one for each input feature in the model. Each NormalizationParameters<T> object contains the information needed to normalize a single feature, such as the minimum and maximum values, mean and standard deviation, or other statistics depending on the normalization method. These parameters are typically calculated during training based on the training data and are then used to normalize new data in the same way. The order of the parameters in the list corresponds to the order of the features in the input data.

For Beginners: This contains the scaling information for each input feature.

The X parameters list:

  • Contains one set of parameters for each input feature
  • Stores information like minimum/maximum values or mean/standard deviation
  • Is used to ensure new data is scaled the same way as training data

For example, if your model has three input features:

  • The first element might contain parameters for normalizing "square_footage"
  • The second element might contain parameters for normalizing "num_bedrooms"
  • The third element might contain parameters for normalizing "location_score"

This information is essential because:

  • Each feature might have a different range of values
  • Each feature needs to be normalized consistently
  • The normalizer uses these parameters to perform the scaling

YParams

Gets or sets the normalization parameters for the target variable.

public NormalizationParameters<T> YParams { get; set; }

Property Value

NormalizationParameters<T>

A NormalizationParameters<T> object for the target variable.

Remarks

This property contains the normalization parameters for the target variable (the variable being predicted). The NormalizationParameters<T> object contains the information needed to normalize the target variable during training and to denormalize predictions to obtain the final results in the original scale. These parameters are typically calculated during training based on the training data and are then used to denormalize predictions made on new data.

For Beginners: This contains the scaling information for the target variable (what you're predicting).

The Y parameters:

  • Store information about how the target variable was normalized during training
  • Are used to convert predictions back to their original scale
  • Contain values like minimum/maximum or mean/standard deviation

For example, if your model predicts house prices:

  • The YParams might store that prices ranged from $100,000 to $1,500,000
  • This information is used to convert a normalized prediction like 0.25 back to $450,000

This is important because:

  • The model works with normalized values internally
  • But users need predictions in the original, meaningful units
  • Consistent denormalization ensures accurate final predictions

Methods

DeepCopy()

Creates a deep copy of this NormalizationInfo instance.

public NormalizationInfo<T, TInput, TOutput> DeepCopy()

Returns

NormalizationInfo<T, TInput, TOutput>

A new NormalizationInfo with copied values.

WithParameters(Vector<T>)

Creates a new NormalizationInfo instance. Since normalization parameters are independent of model parameters, this returns a copy with the same normalization settings.

public NormalizationInfo<T, TInput, TOutput> WithParameters(Vector<T> parameters)

Parameters

parameters Vector<T>

The model parameters (not used for normalization info).

Returns

NormalizationInfo<T, TInput, TOutput>

A new NormalizationInfo with the same normalization settings.

Edit this page