internal SdcaRegressionTrainer(IHostEnvironment env, Options options, string featureColumn, string labelColumn, string weightColumn = null)
: base(env, options, TrainerUtils.MakeR4ScalarColumn(labelColumn), TrainerUtils.MakeR4ScalarWeightColumn(weightColumn))
{
Host.CheckValue(labelColumn, nameof(labelColumn));
Host.CheckValue(featureColumn, nameof(featureColumn));
_loss = options.LossFunction.CreateComponent(env);
Loss = _loss;
}
/// <summary>
/// Initializes a new instance of <see cref="OrdinaryLeastSquaresRegressionTrainer"/>
/// </summary>
internal OrdinaryLeastSquaresRegressionTrainer(IHostEnvironment env, Options options)
: base(Contracts.CheckRef(env, nameof(env)).Register(LoadNameValue), TrainerUtils.MakeR4VecFeature(options.FeatureColumnName),
TrainerUtils.MakeR4ScalarColumn(options.LabelColumnName), TrainerUtils.MakeR4ScalarWeightColumn(options.ExampleWeightColumnName))
{
Host.CheckValue(options, nameof(options));
Host.CheckUserArg(options.L2Weight >= 0, nameof(options.L2Weight), "L2 regularization term cannot be negative");
_l2Weight = options.L2Weight;
_perParameterSignificance = options.PerParameterSignificance;
}
/// <summary>
/// Initializes a new instance of <see cref="OlsTrainer"/>
/// </summary>
internal OlsTrainer(IHostEnvironment env, Options options)
: base(Contracts.CheckRef(env, nameof(env)).Register(LoadNameValue), TrainerUtils.MakeR4VecFeature(options.FeatureColumnName),
TrainerUtils.MakeR4ScalarColumn(options.LabelColumnName), TrainerUtils.MakeR4ScalarWeightColumn(options.ExampleWeightColumnName))
{
Host.CheckValue(options, nameof(options));
Host.CheckUserArg(options.L2Regularization >= 0, nameof(options.L2Regularization), "L2 regularization term cannot be negative");
_l2Weight = options.L2Regularization;
_perParameterSignificance = options.CalculateStatistics;
}
internal LbfgsTrainerBase(IHostEnvironment env,
TOptions options,
SchemaShape.Column labelColumn,
Action <TOptions> advancedSettings = null)
: base(Contracts.CheckRef(env, nameof(env)).Register(RegisterName), TrainerUtils.MakeR4VecFeature(options.FeatureColumnName),
labelColumn, TrainerUtils.MakeR4ScalarWeightColumn(options.ExampleWeightColumnName))
{
Host.CheckValue(options, nameof(options));
LbfgsTrainerOptions = options;
// Apply the advanced args, if the user supplied any.
advancedSettings?.Invoke(options);
options.FeatureColumnName = FeatureColumn.Name;
options.LabelColumnName = LabelColumn.Name;
options.ExampleWeightColumnName = WeightColumn.Name;
Host.CheckUserArg(!LbfgsTrainerOptions.UseThreads || LbfgsTrainerOptions.NumberOfThreads > 0 || LbfgsTrainerOptions.NumberOfThreads == null,
nameof(LbfgsTrainerOptions.NumberOfThreads), "Must be positive (or empty for default)");
Host.CheckUserArg(LbfgsTrainerOptions.L2Regularization >= 0, nameof(LbfgsTrainerOptions.L2Regularization), "Must be non-negative");
Host.CheckUserArg(LbfgsTrainerOptions.L1Regularization >= 0, nameof(LbfgsTrainerOptions.L1Regularization), "Must be non-negative");
Host.CheckUserArg(LbfgsTrainerOptions.OptmizationTolerance > 0, nameof(LbfgsTrainerOptions.OptmizationTolerance), "Must be positive");
Host.CheckUserArg(LbfgsTrainerOptions.IterationsToRemember > 0, nameof(LbfgsTrainerOptions.IterationsToRemember), "Must be positive");
Host.CheckUserArg(LbfgsTrainerOptions.NumberOfIterations > 0, nameof(LbfgsTrainerOptions.NumberOfIterations), "Must be positive");
Host.CheckUserArg(LbfgsTrainerOptions.StochasticGradientDescentInitilaizationTolerance >= 0, nameof(LbfgsTrainerOptions.StochasticGradientDescentInitilaizationTolerance), "Must be non-negative");
Host.CheckUserArg(LbfgsTrainerOptions.NumberOfThreads == null || LbfgsTrainerOptions.NumberOfThreads.Value >= 0, nameof(LbfgsTrainerOptions.NumberOfThreads), "Must be non-negative");
Host.CheckParam(!(LbfgsTrainerOptions.L2Regularization < 0), nameof(LbfgsTrainerOptions.L2Regularization), "Must be non-negative, if provided.");
Host.CheckParam(!(LbfgsTrainerOptions.L1Regularization < 0), nameof(LbfgsTrainerOptions.L1Regularization), "Must be non-negative, if provided");
Host.CheckParam(!(LbfgsTrainerOptions.OptmizationTolerance <= 0), nameof(LbfgsTrainerOptions.OptmizationTolerance), "Must be positive, if provided.");
Host.CheckParam(!(LbfgsTrainerOptions.IterationsToRemember <= 0), nameof(LbfgsTrainerOptions.IterationsToRemember), "Must be positive, if provided.");
L2Weight = LbfgsTrainerOptions.L2Regularization;
L1Weight = LbfgsTrainerOptions.L1Regularization;
OptTol = LbfgsTrainerOptions.OptmizationTolerance;
MemorySize = LbfgsTrainerOptions.IterationsToRemember;
MaxIterations = LbfgsTrainerOptions.NumberOfIterations;
SgdInitializationTolerance = LbfgsTrainerOptions.StochasticGradientDescentInitilaizationTolerance;
Quiet = LbfgsTrainerOptions.Quiet;
InitWtsDiameter = LbfgsTrainerOptions.InitialWeightsDiameter;
UseThreads = LbfgsTrainerOptions.UseThreads;
NumThreads = LbfgsTrainerOptions.NumberOfThreads;
DenseOptimizer = LbfgsTrainerOptions.DenseOptimizer;
EnforceNonNegativity = LbfgsTrainerOptions.EnforceNonNegativity;
if (EnforceNonNegativity && ShowTrainingStats)
{
ShowTrainingStats = false;
using (var ch = Host.Start("Initialization"))
{
ch.Warning("The training statistics cannot be computed with non-negativity constraint.");
}
}
ShowTrainingStats = false;
_srcPredictor = default;
}
internal LbfgsTrainerBase(IHostEnvironment env,
TArgs args,
SchemaShape.Column labelColumn,
Action <TArgs> advancedSettings = null)
: base(Contracts.CheckRef(env, nameof(env)).Register(RegisterName), TrainerUtils.MakeR4VecFeature(args.FeatureColumn),
labelColumn, TrainerUtils.MakeR4ScalarWeightColumn(args.WeightColumn))
{
Host.CheckValue(args, nameof(args));
Args = args;
// Apply the advanced args, if the user supplied any.
advancedSettings?.Invoke(args);
args.FeatureColumn = FeatureColumn.Name;
args.LabelColumn = LabelColumn.Name;
args.WeightColumn = WeightColumn.Name;
Host.CheckUserArg(!Args.UseThreads || Args.NumThreads > 0 || Args.NumThreads == null,
nameof(Args.NumThreads), "numThreads must be positive (or empty for default)");
Host.CheckUserArg(Args.L2Weight >= 0, nameof(Args.L2Weight), "Must be non-negative");
Host.CheckUserArg(Args.L1Weight >= 0, nameof(Args.L1Weight), "Must be non-negative");
Host.CheckUserArg(Args.OptTol > 0, nameof(Args.OptTol), "Must be positive");
Host.CheckUserArg(Args.MemorySize > 0, nameof(Args.MemorySize), "Must be positive");
Host.CheckUserArg(Args.MaxIterations > 0, nameof(Args.MaxIterations), "Must be positive");
Host.CheckUserArg(Args.SgdInitializationTolerance >= 0, nameof(Args.SgdInitializationTolerance), "Must be non-negative");
Host.CheckUserArg(Args.NumThreads == null || Args.NumThreads.Value >= 0, nameof(Args.NumThreads), "Must be non-negative");
Host.CheckParam(!(Args.L2Weight < 0), nameof(Args.L2Weight), "Must be non-negative, if provided.");
Host.CheckParam(!(Args.L1Weight < 0), nameof(Args.L1Weight), "Must be non-negative, if provided");
Host.CheckParam(!(Args.OptTol <= 0), nameof(Args.OptTol), "Must be positive, if provided.");
Host.CheckParam(!(Args.MemorySize <= 0), nameof(Args.MemorySize), "Must be positive, if provided.");
L2Weight = Args.L2Weight;
L1Weight = Args.L1Weight;
OptTol = Args.OptTol;
MemorySize = Args.MemorySize;
MaxIterations = Args.MaxIterations;
SgdInitializationTolerance = Args.SgdInitializationTolerance;
Quiet = Args.Quiet;
InitWtsDiameter = Args.InitWtsDiameter;
UseThreads = Args.UseThreads;
NumThreads = Args.NumThreads;
DenseOptimizer = Args.DenseOptimizer;
EnforceNonNegativity = Args.EnforceNonNegativity;
if (EnforceNonNegativity && ShowTrainingStats)
{
ShowTrainingStats = false;
using (var ch = Host.Start("Initialization"))
{
ch.Warning("The training statistics cannot be computed with non-negativity constraint.");
}
}
ShowTrainingStats = false;
_srcPredictor = default;
}
internal SdcaMulticlassClassificationTrainerBase(IHostEnvironment env, MulticlassOptions options,
string featureColumn, string labelColumn, string weightColumn = null)
: base(env, options, TrainerUtils.MakeU4ScalarColumn(labelColumn), TrainerUtils.MakeR4ScalarWeightColumn(weightColumn))
{
Host.CheckValue(labelColumn, nameof(labelColumn));
Host.CheckValue(featureColumn, nameof(featureColumn));
_loss = options.InternalLoss ?? options.LossFunctionFactory.CreateComponent(env);
Loss = _loss;
}
internal SdcaMultiClassTrainer(IHostEnvironment env, Arguments args,
string featureColumn, string labelColumn, string weightColumn = null)
: base(env, args, TrainerUtils.MakeU4ScalarColumn(labelColumn), TrainerUtils.MakeR4ScalarWeightColumn(weightColumn))
{
Host.CheckValue(labelColumn, nameof(labelColumn));
Host.CheckValue(featureColumn, nameof(featureColumn));
_loss = args.LossFunction.CreateComponent(env);
Loss = _loss;
}
internal LdSvmTrainer(IHostEnvironment env, Options options)
: base(Contracts.CheckRef(env, nameof(env)).Register(LoadNameValue),
TrainerUtils.MakeR4VecFeature(options.FeatureColumnName),
TrainerUtils.MakeBoolScalarLabel(options.LabelColumnName),
TrainerUtils.MakeR4ScalarWeightColumn(options.ExampleWeightColumnName))
{
Host.CheckValue(options, nameof(options));
CheckOptions(Host, options);
_options = options;
}
/// <summary>
/// Initializes a new instance of <see cref="SdcaMultiClassTrainer"/>
/// </summary>
/// <param name="env">The environment to use.</param>
/// <param name="labelColumn">The label, or dependent variable.</param>
/// <param name="featureColumn">The features, or independent variables.</param>
/// <param name="weights">The optional example weights.</param>
/// <param name="loss">The custom loss.</param>
/// <param name="l2Const">The L2 regularization hyperparameter.</param>
/// <param name="l1Threshold">The L1 regularization hyperparameter. Higher values will tend to lead to more sparse model.</param>
/// <param name="maxIterations">The maximum number of passes to perform over the data.</param>
internal SdcaMultiClassTrainer(IHostEnvironment env,
string labelColumn = DefaultColumnNames.Label,
string featureColumn = DefaultColumnNames.Features,
string weights = null,
ISupportSdcaClassificationLoss loss = null,
float?l2Const = null,
float?l1Threshold = null,
int?maxIterations = null)
: base(env, featureColumn, TrainerUtils.MakeU4ScalarColumn(labelColumn), TrainerUtils.MakeR4ScalarWeightColumn(weights),
l2Const, l1Threshold, maxIterations)
{
Host.CheckNonEmpty(featureColumn, nameof(featureColumn));
Host.CheckNonEmpty(labelColumn, nameof(labelColumn));
_loss = loss ?? SdcaTrainerOptions.LossFunction.CreateComponent(env);
Loss = _loss;
}
/// <summary>
/// Initializes a new instance of <see cref="SdcaRegressionTrainer"/>
/// </summary>
/// <param name="env">The environment to use.</param>
/// <param name="labelColumn">The label, or dependent variable.</param>
/// <param name="featureColumn">The features, or independent variables.</param>
/// <param name="weights">The optional example weights.</param>
/// <param name="loss">The custom loss.</param>
/// <param name="l2Const">The L2 regularization hyperparameter.</param>
/// <param name="l1Threshold">The L1 regularization hyperparameter. Higher values will tend to lead to more sparse model.</param>
/// <param name="maxIterations">The maximum number of passes to perform over the data.</param>
/// <param name="advancedSettings">A delegate to set more settings.
/// The settings here will override the ones provided in the direct method signature,
/// if both are present and have different values.
/// The columns names, however need to be provided directly, not through the <paramref name="advancedSettings"/>.</param>
public SdcaRegressionTrainer(IHostEnvironment env,
string labelColumn = DefaultColumnNames.Label,
string featureColumn = DefaultColumnNames.Features,
string weights = null,
ISupportSdcaRegressionLoss loss = null,
float?l2Const = null,
float?l1Threshold = null,
int?maxIterations = null,
Action <Arguments> advancedSettings = null)
: base(env, featureColumn, TrainerUtils.MakeR4ScalarLabel(labelColumn), TrainerUtils.MakeR4ScalarWeightColumn(weights), advancedSettings,
l2Const, l1Threshold, maxIterations)
{
Host.CheckNonEmpty(featureColumn, nameof(featureColumn));
Host.CheckNonEmpty(labelColumn, nameof(labelColumn));
_loss = loss ?? Args.LossFunction.CreateComponent(env);
Loss = _loss;
}
private RandomizedPcaTrainer(IHostEnvironment env, Options options, string featureColumnName, string exampleWeightColumnName,
int rank = 20, int oversampling = 20, bool center = true, int?seed = null)
: base(Contracts.CheckRef(env, nameof(env)).Register(LoadNameValue), TrainerUtils.MakeR4VecFeature(featureColumnName), default, TrainerUtils.MakeR4ScalarWeightColumn(exampleWeightColumnName))
{
// if the args are not null, we got here from maml, and the internal ctor.
if (options != null)
{
_rank = options.Rank;
_ensureZeroMean = options.EnsureZeroMean;
_oversampling = options.Oversampling;
_seed = options.Seed ?? Host.Rand.Next();
}
else
{
_rank = rank;
_ensureZeroMean = center;
_oversampling = oversampling;
_seed = seed ?? Host.Rand.Next();
}
_featureColumn = featureColumnName;
Host.CheckUserArg(_rank > 0, nameof(_rank), "Rank must be positive");
Host.CheckUserArg(_oversampling >= 0, nameof(_oversampling), "Oversampling must be non-negative");
}