Microsoft.ML.Trainers Namespace

Arguments class for averaged linear trainers.

Base class for averaged linear trainers.

The IEstimator to predict a target using a linear binary classification model trained with the averaged perceptron.

Options for the AveragedPerceptronTrainer as used in AveragedPerceptron(BinaryClassificationCatalog+BinaryClassificationTrainers, AveragedPerceptronTrainer+Options).

Represents a coefficient statistics object containing statistics about the calculated model parameters.

Computes the standard deviation matrix of each of the non-zero training weights, needed to calculate further the standard deviation, p-value and z-Score. Use this class' implementation in the Microsoft.ML.Mkl.Components package which uses Intel Math Kernel Library. Due to the existence of regularization, an approximation is used to compute the variances of the trained linear coefficients.

Exponential Loss, commonly used in classification tasks.

This class implements Exponential Learning rate decay. Implemented from the tensorflow documentation. Source: https://www.tensorflow.org/api_docs/python/tf/compat/v1/train/exponential_decay Default values and implementation of learning rate is from Tensorflow Slim model tests. Source : https://github.com/tensorflow/models/blob/master/research/slim/train_image_classifier.py

Support for feature contribution calculation.

Model parameters for FieldAwareFactorizationMachineTrainer.

The IEstimator to predict a target using a field-aware factorization machine model trained using a stochastic gradient method.

FieldAwareFactorizationMachineTrainer.Options for FieldAwareFactorizationMachineTrainer as used in FieldAwareFactorizationMachine(BinaryClassificationCatalog+BinaryClassificationTrainers, FieldAwareFactorizationMachineTrainer+Options).

Hinge Loss, commonly used in classification tasks.

The IEstimator for training a KMeans clusterer

Options for the KMeansTrainer as used in KMeansTrainer(Options).

The IEstimator to predict a target using a linear logistic regression model trained with L-BFGS method.

Options for the LbfgsLogisticRegressionBinaryTrainer as used in LbfgsLogisticRegression(BinaryClassificationCatalog+BinaryClassificationTrainers, LbfgsLogisticRegressionBinaryTrainer+Options)

The IEstimator to predict a target using a maximum entropy multiclass classifier trained with L-BFGS method.

LbfgsMaximumEntropyMulticlassTrainer.Options for LbfgsMaximumEntropyMulticlassTrainer as used in LbfgsMaximumEntropy(MulticlassClassificationCatalog+MulticlassClassificationTrainers, LbfgsMaximumEntropyMulticlassTrainer+Options).

The IEstimator for training a Poisson regression model.

Options for the LbfgsPoissonRegressionTrainer as used in LbfgsPoissonRegression(Options).

Base options class for trainer estimators deriving fromLbfgsTrainerBase.

Base class for L-BFGS-based trainers.

The IEstimator to predict a target using a non-linear binary classification model trained with Local Deep SVM.

This abstract class defines a learning rate scheduler.

The model parameters class for linear binary trainer estimators.

Base class for linear model parameters.

Statistics for linear model parameters.

Linear model of multiclass classifiers. It outputs raw scores of all its linear models, and no probablistic output is provided.

Common linear model of multiclass classifiers. LinearMulticlassModelParameters contains a single linear model per class.

Model parameters for linear regression.

The IEstimator to predict a target using a linear binary classification model trained with Linear SVM.

Options for the LinearSvmTrainer as used in LinearSvm(BinaryClassificationCatalog+BinaryClassificationTrainers, LinearSvmTrainer+Options).

The Log Loss, also known as the Cross Entropy Loss. It is commonly used in classification tasks.

This class implements linear scaling rule and LR decay. Implementation adopted from RESNET-CIFAR benchmark test in Tensorflow slim. https://github.com/tensorflow/models/blob/b974c3f95a37acedcc3c58566834c78fcae4b214/official/vision/image_classification/resnet_cifar_main.py

The IEstimator to predict elements in a matrix using matrix factorization (also known as a type of collaborative filtering).

Options for the MatrixFactorizationTrainer as used in MatrixFactorization(Options).

Linear maximum entropy model of multiclass classifiers. It outputs classes probabilities. This model is also known as multinomial logistic regression. Please see https://en.wikipedia.org/wiki/Multinomial_logistic_regression for details.

Generic base class for all model parameters.

Statistics for linear model parameters.

Model parameters for NaiveBayesMulticlassTrainer.

The IEstimator for training a multiclass Naive Bayes model that supports binary feature values.

Model parameters for OlsTrainer.

The IEstimator for training a linear regression model using ordinary least squares (OLS) for estimating the parameters of the linear regression model.

Options for the OlsTrainer as used in Ols(Options)

Model parameters for OneVersusAllTrainer.

The IEstimator for training a one-versus-all multi-class classifier that uses the specified binary classifier.

The IEstimator for training a linear regression model using Online Gradient Descent (OGD) for estimating the parameters of the linear regression model.

Options for the OnlineGradientDescentTrainer as used in OnlineGradientDescent(Options).

Arguments class for online linear trainers.

Base class for online linear trainers. Online trainers can be updated incrementally with additional data.

Model parameters for PairwiseCouplingTrainer.

The IEstimator for training a pairwise coupling multi-class classifier that uses the specified binary classifier.

Model parameters for RandomizedPcaTrainer.

Poisson Loss function for Poisson Regression.

Model parameters for Poisson Regression.

This class implements polynomial Learning rate decay. Implemented from the tensorflow documentation. Source: https://www.tensorflow.org/api_docs/python/tf/compat/v1/train/polynomial_decay Default values and implementation of learning rate is from Tensorflow Slim model tests. Source : https://github.com/tensorflow/models/blob/master/research/slim/train_image_classifier.py

Model parameters for PriorTrainer.

The IEstimator for predicting a target using a binary classification model.

The IEstimator for training an approximate PCA using Randomized SVD algorithm.

Options for the RandomizedPcaTrainer as used in RandomizedPca(Options).

Model parameters for regression.

Options for SdcaBinaryTrainerBase.

SDCA is a general training algorithm for (generalized) linear models such as support vector machine, linear regression, logistic regression, and so on. SDCA binary classification trainer family includes several sealed members: (1) SdcaNonCalibratedBinaryTrainer supports general loss functions and returns LinearBinaryModelParameters. (2) SdcaLogisticRegressionBinaryTrainer essentially trains a regularized logistic regression model. Because logistic regression naturally provide probability output, this generated model's type is CalibratedModelParametersBase. where TSubModel is LinearBinaryModelParameters and TCalibrator is PlattCalibrator.

The IEstimator for training a binary logistic regression classification model using the stochastic dual coordinate ascent method. The trained model is calibrated and can produce probability by feeding the output value of the linear function to a PlattCalibrator.

Options for the SdcaLogisticRegressionBinaryTrainer as used in SdcaLogisticRegression(Options).

The IEstimator to predict a target using a maximum entropy multiclass classifier. The trained model MaximumEntropyModelParameters produces probabilities of classes.

SdcaMaximumEntropyMulticlassTrainer.Options for SdcaMaximumEntropyMulticlassTrainer as used in SdcaMaximumEntropy(MulticlassClassificationCatalog+MulticlassClassificationTrainers, String, String, String, Nullable, Nullable, Nullable)

Options for the SdcaMulticlassTrainerBase.

The IEstimator to predict a target using a linear multiclass classifier model trained with a coordinate descent method. Depending on the used loss function, the trained model can be, for example, maximum entropy classifier or multi-class support vector machine.

The IEstimator for training a binary logistic regression classification model using the stochastic dual coordinate ascent method.

Options for the SdcaNonCalibratedBinaryTrainer.

TheIEstimator to predict a target using a linear multiclass classifier. The trained model LinearMulticlassModelParameters produces probabilities of classes.

SdcaNonCalibratedMulticlassTrainer.Options for SdcaNonCalibratedMulticlassTrainer as used in SdcaNonCalibrated(MulticlassClassificationCatalog+MulticlassClassificationTrainers, String, String, String, ISupportSdcaClassificationLoss, Nullable, Nullable, Nullable).

The IEstimator for training a regression model using the stochastic dual coordinate ascent method.

Options for the SdcaRegressionTrainer.

Options for the SDCA-based trainers.

The IEstimator for training logistic regression using a parallel stochastic gradient method. The trained model is calibrated and can produce probability by feeding the output value of the linear function to a PlattCalibrator.

Options for the SgdCalibratedTrainer as used in SgdCalibrated(Options).

The IEstimator for training logistic regression using a parallel stochastic gradient method.

Options for the SgdNonCalibratedTrainer as used in SgdNonCalibrated(Options).

A smooth version of the HingeLoss function, commonly used in classification tasks.

The Squared Loss, commonly used in regression tasks.

The IEstimator to predict a target using a linear binary classification model trained with the symbolic stochastic gradient descent.

Options for the SymbolicSgdLogisticRegressionBinaryTrainer as used in SymbolicSgdLogisticRegression(BinaryClassificationCatalog+BinaryClassificationTrainers, SymbolicSgdLogisticRegressionBinaryTrainer+Options).

This represents a basic class for 'simple trainer'. A 'simple trainer' accepts one feature column and one label column, also optionally a weight column. It produces a 'prediction transformer'.

This represents a basic class for 'simple trainer'. A 'simple trainer' accepts one feature column and one label column, also optionally a weight column. It produces a 'prediction transformer'.

The base class for all trainer inputs.

The base class for all trainer inputs that support a group column.

The base class for all trainer inputs that support a Label column.

The base class for all trainer inputs that support a weight column.

Tweedie loss, based on the log-likelihood of the Tweedie distribution. This loss function is used in Tweedie regression.

The base class for all unsupervised trainer inputs that support a weight column.

This structure represents a learning rate scheduler item type

Allows support for feature contribution calculation by model parameters.

The loss function may know the close-form solution to the optimal dual update Ref: Sec(6.2) of http://jmlr.org/papers/volume14/shalev-shwartz13a/shalev-shwartz13a.pdf

Interface for the Trainer Estimator.

Type of loss function.