Libraries/Microsoft.ML.StandardLearners.xml
|
<?xml version="1.0"?>
<doc> <assembly> <name>Microsoft.ML.StandardLearners</name> </assembly> <members> <member name="T:Microsoft.ML.Runtime.Numeric.DifferentiableFunction"> <summary> A delegate for functions with gradients. </summary> <param name="input">The point at which to evaluate the function</param> <param name="gradient">The gradient vector, which must be filled in (its initial contents are undefined)</param> <param name="progress">The progress channel provider that can be used to report calculation progress. Can be null.</param> <returns>The value of the function</returns> </member> <member name="T:Microsoft.ML.Runtime.Numeric.IndexedDifferentiableFunction"> <summary> A delegate for indexed sets of functions with gradients. REVIEW: I didn't add an <see cref="T:Microsoft.ML.Runtime.IProgressChannelProvider"/> here, since it looks like this code is not actually accessed from anywhere. Maybe it should go away? </summary> <param name="index">The index of the function</param> <param name="input">The point at which to evaluate the function</param> <param name="gradient">The gradient vector, which must be filled in (its initial contents are undefined)</param> <returns>The value of the function</returns> </member> <member name="T:Microsoft.ML.Runtime.Numeric.DifferentiableFunctionAggregator"> <summary> Class to aggregate an indexed differentiable function into a single function, in parallel </summary> </member> <member name="M:Microsoft.ML.Runtime.Numeric.DifferentiableFunctionAggregator.#ctor(Microsoft.ML.Runtime.Numeric.IndexedDifferentiableFunction,System.Int32,System.Int32,System.Int32)"> <summary> Creates a DifferentiableFunctionAggregator </summary> <param name="func">Indexed function to use</param> <param name="dim">Dimensionality of the function</param> <param name="maxIndex">Max index of the function</param> <param name="threads">Number of threads to use</param> </member> <member name="M:Microsoft.ML.Runtime.Numeric.DifferentiableFunctionAggregator.Eval(Microsoft.ML.Runtime.Data.VBuffer{System.Single}@,Microsoft.ML.Runtime.Data.VBuffer{System.Single}@)"> <summary> Evaluate and sum the function over all indices, in parallel </summary> <param name="input">The point at which to evaluate the function</param> <param name="gradient">The gradient vector, which must be filled in (its initial contents are undefined)</param> <returns>Function value</returns> </member> <member name="T:Microsoft.ML.Runtime.Numeric.GradientTester"> <summary> A class for testing the gradient of DifferentiableFunctions, useful for debugging </summary> <remarks> Works by comparing the reported gradient to the numerically computed gradient. If the gradient is correct, the return value should be small (order of 1e-6). May have false negatives if extreme values cause the numeric gradient to be off, e.g. if the norm of x is very large, or if the gradient is changing rapidly at x. </remarks> </member> <member name="M:Microsoft.ML.Runtime.Numeric.GradientTester.Test(Microsoft.ML.Runtime.Numeric.DifferentiableFunction,Microsoft.ML.Runtime.Data.VBuffer{System.Single}@)"> <summary> Tests the gradient reported by f. </summary> <param name="f">function to test</param> <param name="x">point at which to test</param> <returns>maximum normalized difference between analytic and numeric directional derivative over multiple tests</returns> </member> <member name="M:Microsoft.ML.Runtime.Numeric.GradientTester.Test(Microsoft.ML.Runtime.Numeric.DifferentiableFunction,Microsoft.ML.Runtime.Data.VBuffer{System.Single}@,System.Boolean)"> <summary> Tests the gradient reported by f. </summary> <param name="f">function to test</param> <param name="x">point at which to test</param> <param name="quiet">If false, outputs detailed info.</param> <returns>maximum normalized difference between analytic and numeric directional derivative over multiple tests</returns> </member> <member name="F:Microsoft.ML.Runtime.Numeric.GradientTester.Header"> <summary> The head of the test output </summary> </member> <member name="M:Microsoft.ML.Runtime.Numeric.GradientTester.TestAllCoords(Microsoft.ML.Runtime.Numeric.DifferentiableFunction,Microsoft.ML.Runtime.Data.VBuffer{System.Single}@)"> <summary> Tests the gradient using finite differences on each axis (appropriate for small functions) </summary> <param name="f"></param> <param name="x"></param> </member> <member name="M:Microsoft.ML.Runtime.Numeric.GradientTester.TestCoords(Microsoft.ML.Runtime.Numeric.DifferentiableFunction,Microsoft.ML.Runtime.Data.VBuffer{System.Single}@,System.Collections.Generic.IList{System.Int32})"> <summary> Tests the gradient using finite differences on each axis in the list </summary> <param name="f">Function to test</param> <param name="x">Point at which to test</param> <param name="coords">List of coordinates to test</param> </member> <member name="M:Microsoft.ML.Runtime.Numeric.GradientTester.Test(Microsoft.ML.Runtime.Numeric.DifferentiableFunction,Microsoft.ML.Runtime.Data.VBuffer{System.Single}@,Microsoft.ML.Runtime.Data.VBuffer{System.Single}@,System.Boolean,Microsoft.ML.Runtime.Data.VBuffer{System.Single}@,Microsoft.ML.Runtime.Data.VBuffer{System.Single}@)"> <summary> Tests the gradient reported by <paramref name="f"/>. </summary> <param name="f">Function to test</param> <param name="x">Point at which to test</param> <param name="dir">Direction to test derivative</param> <param name="quiet">Whether to disable output</param> <param name="newGrad">This is a reusable working buffer for intermediate calculations</param> <param name="newX">This is a reusable working buffer for intermediate calculations</param> <returns>Normalized difference between analytic and numeric directional derivative</returns> </member> <member name="T:Microsoft.ML.Runtime.Numeric.L1Optimizer"> <summary> Orthant-Wise Limited-memory Quasi-Newton algorithm for optimization of smooth convex objectives plus L1-regularization If you use this code for published research, please cite Galen Andrew and Jianfeng Gao, "Scalable Training of L1-Regularized Log-Linear Models", ICML 2007 </summary> </member> <member name="M:Microsoft.ML.Runtime.Numeric.L1Optimizer.#ctor(Microsoft.ML.Runtime.IHostEnvironment,System.Int32,System.Single,System.Int32,System.Boolean,Microsoft.ML.Runtime.Numeric.ITerminationCriterion,System.Boolean)"> <summary> Create an L1Optimizer with the supplied value of M and termination criterion </summary> <param name="env">The environment</param> <param name="biasCount">Number of biases</param> <param name="l1weight">Weight of L1 regularizer</param> <param name="m">The number of previous iterations to store</param> <param name="keepDense">Whether the optimizer will keep its internal state dense</param> <param name="term">Termination criterion</param> <param name="enforceNonNegativity">The flag enforcing the non-negativity constraint</param> </member> <member name="T:Microsoft.ML.Runtime.Numeric.L1Optimizer.L1OptimizerState"> <summary> Contains information about the state of the optimizer </summary> </member> <member name="M:Microsoft.ML.Runtime.Numeric.L1Optimizer.L1OptimizerState.EvalCore(Microsoft.ML.Runtime.Data.VBuffer{System.Single}@,Microsoft.ML.Runtime.Data.VBuffer{System.Single}@,Microsoft.ML.Runtime.IProgressChannelProvider)"> <summary> This is the original differentiable function with the injected L1 term. </summary> </member> <member name="M:Microsoft.ML.Runtime.Numeric.L1Optimizer.L1OptimizerState.LineSearch(Microsoft.ML.Runtime.IChannel,System.Boolean)"> <summary> Backtracking line search with Armijo-like condition, from Andrew & Gao </summary> </member> <member name="T:Microsoft.ML.Runtime.Numeric.ILineSearch"> <summary> Line search that does not use derivatives </summary> </member> <member name="M:Microsoft.ML.Runtime.Numeric.ILineSearch.Minimize(System.Func{System.Single,System.Single})"> <summary> Finds a local minimum of the function </summary> <param name="func">Function to minimize</param> <returns>Minimizing value</returns> </member> <member name="T:Microsoft.ML.Runtime.Numeric.DiffFunc1D"> <summary> Delegate for differentiable 1-D functions </summary> <param name="x">Point to evaluate</param> <param name="deriv">Derivative at that point</param> <returns></returns> </member> <member name="T:Microsoft.ML.Runtime.Numeric.IDiffLineSearch"> <summary> Line search that uses derivatives </summary> </member> <member name="M:Microsoft.ML.Runtime.Numeric.IDiffLineSearch.Minimize(Microsoft.ML.Runtime.Numeric.DiffFunc1D,System.Single,System.Single)"> <summary> Finds a local minimum of the function </summary> <param name="func">Function to minimize</param> <param name="initValue">Value of function at 0</param> <param name="initDeriv">Derivative of function at 0</param> <returns>Minimizing value</returns> </member> <member name="T:Microsoft.ML.Runtime.Numeric.CubicInterpLineSearch"> <summary> Cubic interpolation line search </summary> </member> <member name="P:Microsoft.ML.Runtime.Numeric.CubicInterpLineSearch.MaxNumSteps"> <summary> Gets or sets maximum number of steps. </summary> </member> <member name="P:Microsoft.ML.Runtime.Numeric.CubicInterpLineSearch.MinWindow"> <summary> Gets or sets the minimum relative size of bounds around solution. </summary> </member> <member name="P:Microsoft.ML.Runtime.Numeric.CubicInterpLineSearch.MaxStep"> <summary> Gets or sets maximum step size </summary> </member> <member name="M:Microsoft.ML.Runtime.Numeric.CubicInterpLineSearch.#ctor(System.Int32)"> <summary> Makes a CubicInterpLineSearch </summary> <param name="maxNumSteps">Maximum number of steps before terminating</param> </member> <member name="M:Microsoft.ML.Runtime.Numeric.CubicInterpLineSearch.#ctor(System.Single)"> <summary> Makes a CubicInterpLineSearch </summary> <param name="minWindow">Minimum relative size of bounds around solution</param> </member> <member name="M:Microsoft.ML.Runtime.Numeric.CubicInterpLineSearch.CubicInterp(Microsoft.ML.Runtime.Numeric.CubicInterpLineSearch.StepValueDeriv,Microsoft.ML.Runtime.Numeric.CubicInterpLineSearch.StepValueDeriv)"> <summary> Cubic interpolation routine from Nocedal and Wright </summary> <param name="a">first point, with value and derivative</param> <param name="b">second point, with value and derivative</param> <returns>local minimum of interpolating cubic polynomial</returns> </member> <member name="M:Microsoft.ML.Runtime.Numeric.CubicInterpLineSearch.Minimize(Microsoft.ML.Runtime.Numeric.DiffFunc1D,System.Single,System.Single)"> <summary> Finds a local minimum of the function </summary> <param name="func">Function to minimize</param> <param name="initValue">Value of function at 0</param> <param name="initDeriv">Derivative of function at 0</param> <returns>Minimizing value</returns> </member> <member name="T:Microsoft.ML.Runtime.Numeric.GoldenSectionSearch"> <summary> Finds local minimum with golden section search. </summary> </member> <member name="P:Microsoft.ML.Runtime.Numeric.GoldenSectionSearch.MaxNumSteps"> <summary> Gets or sets maximum number of steps before terminating. </summary> </member> <member name="P:Microsoft.ML.Runtime.Numeric.GoldenSectionSearch.MinWindow"> <summary> Gets or sets minimum relative size of bounds around solution. </summary> </member> <member name="P:Microsoft.ML.Runtime.Numeric.GoldenSectionSearch.MaxStep"> <summary> Gets or sets maximum step size. </summary> </member> <member name="M:Microsoft.ML.Runtime.Numeric.GoldenSectionSearch.#ctor(System.Int32)"> <summary> Makes a new GoldenSectionSearch </summary> <param name="maxNumSteps">Maximum number of steps before terminating (not including bracketing)</param> </member> <member name="M:Microsoft.ML.Runtime.Numeric.GoldenSectionSearch.#ctor(System.Single)"> <summary> Makes a new GoldenSectionSearch </summary> <param name="minWindow">Minimum relative size of bounds around solution</param> </member> <member name="M:Microsoft.ML.Runtime.Numeric.GoldenSectionSearch.Minimize(Microsoft.ML.Runtime.Numeric.DiffFunc1D,System.Single,System.Single)"> <summary> Finds a local minimum of the function </summary> <param name="f">Function to minimize</param> <param name="initVal">Value of function at 0</param> <param name="initDeriv">Derivative of function at 0</param> <returns>Minimizing value</returns> </member> <member name="M:Microsoft.ML.Runtime.Numeric.GoldenSectionSearch.Minimize(Microsoft.ML.Runtime.Numeric.DiffFunc1D)"> <summary> Finds a local minimum of the function </summary> <param name="func">Function to minimize</param> <returns>Minimizing value</returns> </member> <member name="M:Microsoft.ML.Runtime.Numeric.GoldenSectionSearch.Minimize(System.Func{System.Single,System.Single})"> <summary> Finds a local minimum of the function </summary> <param name="func">Function to minimize</param> <returns>Minimizing value</returns> </member> <member name="T:Microsoft.ML.Runtime.Numeric.BacktrackingLineSearch"> <summary> Backtracking line search with Armijo condition </summary> </member> <member name="M:Microsoft.ML.Runtime.Numeric.BacktrackingLineSearch.#ctor(System.Single)"> <summary> Makes a backtracking line search </summary> <param name="c1">Parameter for Armijo condition</param> </member> <member name="M:Microsoft.ML.Runtime.Numeric.BacktrackingLineSearch.Minimize(Microsoft.ML.Runtime.Numeric.DiffFunc1D,System.Single,System.Single)"> <summary> Finds a local minimum of the function </summary> <param name="f">Function to minimize</param> <param name="initVal">Value of function at 0</param> <param name="initDeriv">Derivative of function at 0</param> <returns>Minimizing value</returns> </member> <member name="T:Microsoft.ML.Runtime.Numeric.ITerminationCriterion"> <summary> An object which is used to decide whether to stop optimization. </summary> </member> <member name="P:Microsoft.ML.Runtime.Numeric.ITerminationCriterion.FriendlyName"> <summary> Name appropriate for display to the user. </summary> </member> <member name="M:Microsoft.ML.Runtime.Numeric.ITerminationCriterion.Terminate(Microsoft.ML.Runtime.Numeric.Optimizer.OptimizerState,System.String@)"> <summary> Determines whether to stop optimization </summary> <param name="state">the state of the optimizer</param> <param name="message">a message to be printed (or null for no message)</param> <returns>true iff criterion is met, i.e. optimization should halt</returns> </member> <member name="M:Microsoft.ML.Runtime.Numeric.ITerminationCriterion.Reset"> <summary> Prepares the ITerminationCriterion for a new round of optimization </summary> </member> <member name="T:Microsoft.ML.Runtime.Numeric.GradientCheckingMonitor"> <summary> A wrapper for a termination criterion that checks the gradient at a specified interval </summary> </member> <member name="M:Microsoft.ML.Runtime.Numeric.GradientCheckingMonitor.#ctor(Microsoft.ML.Runtime.Numeric.ITerminationCriterion,System.Int32)"> <summary> Initializes a new instance of the <see cref="T:Microsoft.ML.Runtime.Numeric.GradientCheckingMonitor"/> class. </summary> <param name="termCrit">The termination criterion</param> <param name="gradientCheckInterval">The gradient check interval.</param> </member> <member name="M:Microsoft.ML.Runtime.Numeric.GradientCheckingMonitor.Terminate(Microsoft.ML.Runtime.Numeric.Optimizer.OptimizerState,System.String@)"> <summary> Determines whether to stop optimization </summary> <param name="state">the state of the optimizer</param> <param name="message">a message to be printed (or null for no message)</param> <returns> true iff criterion is met, i.e. optimization should halt </returns> </member> <member name="M:Microsoft.ML.Runtime.Numeric.GradientCheckingMonitor.Reset"> <summary> Prepares the ITerminationCriterion for a new round of optimization </summary> </member> <member name="T:Microsoft.ML.Runtime.Numeric.StaticTerminationCriterion"> <summary> An abstract partial implementation of ITerminationCriterion for those which do not require resetting </summary> </member> <member name="M:Microsoft.ML.Runtime.Numeric.StaticTerminationCriterion.Terminate(Microsoft.ML.Runtime.Numeric.Optimizer.OptimizerState,System.String@)"> <summary> Determines whether to stop optimization </summary> <param name="state">the state of the optimizer</param> <param name="message">a message to be printed (or null for no message)</param> <returns> true iff criterion is met, i.e. optimization should halt </returns> </member> <member name="M:Microsoft.ML.Runtime.Numeric.StaticTerminationCriterion.Reset"> <summary> Prepares the ITerminationCriterion for a new round of optimization </summary> </member> <member name="T:Microsoft.ML.Runtime.Numeric.MeanImprovementCriterion"> <summary> Terminates when the geometrically-weighted average improvement falls below the tolerance </summary> </member> <member name="M:Microsoft.ML.Runtime.Numeric.MeanImprovementCriterion.#ctor(System.Single,System.Single,System.Int32)"> <summary> Initializes a new instance of the <see cref="T:Microsoft.ML.Runtime.Numeric.MeanImprovementCriterion"/> class. </summary> <param name="tol">The tolerance parameter</param> <param name="lambda">The geometric weighting factor. Higher means more heavily weighted toward older values.</param> <param name="maxIterations">Maximum amount of iteration</param> </member> <member name="P:Microsoft.ML.Runtime.Numeric.MeanImprovementCriterion.Tolerance"> <summary> When criterion drops below this value, optimization is terminated </summary> </member> <member name="M:Microsoft.ML.Runtime.Numeric.MeanImprovementCriterion.Terminate(Microsoft.ML.Runtime.Numeric.Optimizer.OptimizerState,System.String@)"> <summary> Determines whether to stop optimization </summary> <param name="state">the state of the optimizer</param> <param name="message">a message to be printed (or null for no message)</param> <returns> true iff criterion is met, i.e. optimization should halt </returns> </member> <member name="M:Microsoft.ML.Runtime.Numeric.MeanImprovementCriterion.Reset"> <summary> Prepares the ITerminationCriterion for a new round of optimization </summary> </member> <member name="T:Microsoft.ML.Runtime.Numeric.MeanRelativeImprovementCriterion"> <summary> Stops optimization when the average objective improvement over the last n iterations, normalized by the function value, is small enough. </summary> <remarks> Inappropriate for functions whose optimal value is non-positive, because of normalization </remarks> </member> <member name="P:Microsoft.ML.Runtime.Numeric.MeanRelativeImprovementCriterion.Tolerance"> <summary> When criterion drops below this value, optimization is terminated </summary> </member> <member name="P:Microsoft.ML.Runtime.Numeric.MeanRelativeImprovementCriterion.Iters"> <summary> Number of previous iterations to store </summary> </member> <member name="M:Microsoft.ML.Runtime.Numeric.MeanRelativeImprovementCriterion.#ctor(System.Single,System.Int32,System.Int32)"> <summary> Create a MeanRelativeImprovementCriterion </summary> <param name="tol">tolerance level</param> <param name="n">number of past iterations to average over</param> <param name="maxIterations">Maximum amount of iteration</param> </member> <member name="M:Microsoft.ML.Runtime.Numeric.MeanRelativeImprovementCriterion.Terminate(Microsoft.ML.Runtime.Numeric.Optimizer.OptimizerState,System.String@)"> <summary> Returns true if the average objective improvement over the last n iterations, normalized by the function value, is less than the tolerance </summary> <param name="state">current state of the optimizer</param> <param name="message">the current value of the criterion</param> <returns>true if criterion is less than tolerance</returns> </member> <member name="M:Microsoft.ML.Runtime.Numeric.MeanRelativeImprovementCriterion.ToString"> <summary> String summary of criterion </summary> <returns>summary of criterion</returns> </member> <member name="M:Microsoft.ML.Runtime.Numeric.MeanRelativeImprovementCriterion.Reset"> <summary> Prepares the ITerminationCriterion for a new round of optimization </summary> </member> <member name="T:Microsoft.ML.Runtime.Numeric.UpperBoundOnDistanceWithL2"> <summary> Uses the gradient to determine an upper bound on (relative) distance from the optimum. </summary> <remarks> Works if the objective uses L2 prior (or in general if the hessian H is such that H > (1 / sigmaSq) * I at all points) Inappropriate for functions whose optimal value is non-positive, because of normalization </remarks> </member> <member name="P:Microsoft.ML.Runtime.Numeric.UpperBoundOnDistanceWithL2.Tolerance"> <summary> When criterion drops below this value, optimization is terminated </summary> </member> <member name="M:Microsoft.ML.Runtime.Numeric.UpperBoundOnDistanceWithL2.#ctor(System.Single,System.Single)"> <summary> Create termination criterion with supplied value of sigmaSq and tolerance </summary> <param name="sigmaSq">value of sigmaSq in L2 regularizer</param> <param name="tol">tolerance level</param> </member> <member name="M:Microsoft.ML.Runtime.Numeric.UpperBoundOnDistanceWithL2.Terminate(Microsoft.ML.Runtime.Numeric.Optimizer.OptimizerState,System.String@)"> <summary> Returns true if the proved bound on the distance from the optimum, normalized by the function value, is less than the tolerance </summary> <param name="state">current state of the optimizer</param> <param name="message">value of criterion</param> <returns>true if criterion is less than tolerance</returns> </member> <member name="M:Microsoft.ML.Runtime.Numeric.UpperBoundOnDistanceWithL2.ToString"> <summary> String summary of criterion </summary> <returns>summary of criterion</returns> </member> <member name="T:Microsoft.ML.Runtime.Numeric.RelativeNormGradient"> <summary> Criterion based on the norm of the gradient being small enough </summary> <remarks> Inappropriate for functions whose optimal value is non-positive, because of normalization </remarks> </member> <member name="P:Microsoft.ML.Runtime.Numeric.RelativeNormGradient.Tolerance"> <summary> When criterion drops below this value, optimization is terminated </summary> </member> <member name="M:Microsoft.ML.Runtime.Numeric.RelativeNormGradient.#ctor(System.Single)"> <summary> Create a RelativeNormGradient with the supplied tolerance </summary> <param name="tol">tolerance level</param> </member> <member name="M:Microsoft.ML.Runtime.Numeric.RelativeNormGradient.Terminate(Microsoft.ML.Runtime.Numeric.Optimizer.OptimizerState,System.String@)"> <summary> Returns true if the norm of the gradient, divided by the value, is less than the tolerance. </summary> <param name="state">current state of the optimzer</param> <param name="message">the current value of the criterion</param> <returns>true iff criterion is less than the tolerance</returns> </member> <member name="M:Microsoft.ML.Runtime.Numeric.RelativeNormGradient.ToString"> <summary> String summary of criterion </summary> <returns>summary of criterion</returns> </member> <member name="T:Microsoft.ML.Runtime.Numeric.Optimizer"> <summary> Limited-memory BFGS quasi-Newton optimization routine </summary> </member> <member name="F:Microsoft.ML.Runtime.Numeric.Optimizer.EnforceNonNegativity"> Based on Nocedal and Wright, "Numerical Optimization, Second Edition" </member> <member name="F:Microsoft.ML.Runtime.Numeric.Optimizer.Env"> <summary> The host environment to use for reporting progress and exceptions. </summary> </member> <member name="P:Microsoft.ML.Runtime.Numeric.Optimizer.M"> <summary> Number of previous iterations to remember for estimate of Hessian. </summary> <remarks> Higher M means better approximation to Newton's method, but uses more memory, and requires more time to compute direction. The optimal setting of M is problem specific, depending on such factors as how expensive is function evaluation compared to choosing the direction, how easily approximable is the function's Hessian, etc. M = 15..20 is usually reasonable but if necessary even M=2 is better than gradient descent </remarks> </member> <member name="P:Microsoft.ML.Runtime.Numeric.Optimizer.TotalMemoryLimit"> <summary> Gets or sets a bound on the total number of bytes allowed. If the whole application is using more than this, no more vectors will be allocated. </summary> </member> <member name="M:Microsoft.ML.Runtime.Numeric.Optimizer.#ctor(Microsoft.ML.Runtime.IHostEnvironment,System.Int32,System.Boolean,Microsoft.ML.Runtime.Numeric.ITerminationCriterion,System.Boolean)"> <summary> Create an optimizer with the supplied value of M and termination criterion </summary> <param name="env">The host environment</param> <param name="m">The number of previous iterations to store</param> <param name="keepDense">Whether the optimizer will keep its internal state dense</param> <param name="term">Termination criterion, defaults to MeanRelativeImprovement if null</param> <param name="enforceNonNegativity">The flag enforcing the non-negativity constraint</param> </member> <member name="T:Microsoft.ML.Runtime.Numeric.Optimizer.OptimizerException"> <summary> A class for exceptions thrown by the optimizer. </summary> </member> <member name="P:Microsoft.ML.Runtime.Numeric.Optimizer.OptimizerException.State"> <summary> The state of the optimizer when premature convergence happened. </summary> </member> <member name="T:Microsoft.ML.Runtime.Numeric.Optimizer.OptimizerState"> <summary> Contains information about the state of the optimizer </summary> </member> <member name="F:Microsoft.ML.Runtime.Numeric.Optimizer.OptimizerState.Dim"> <summary> The dimensionality of the function </summary> </member> <member name="P:Microsoft.ML.Runtime.Numeric.Optimizer.OptimizerState.Function"> <summary> The function being optimized </summary> </member> <member name="P:Microsoft.ML.Runtime.Numeric.Optimizer.OptimizerState.X"> <summary> The current point being explored </summary> </member> <member name="P:Microsoft.ML.Runtime.Numeric.Optimizer.OptimizerState.Grad"> <summary> The gradient at the current point </summary> </member> <member name="P:Microsoft.ML.Runtime.Numeric.Optimizer.OptimizerState.LastDir"> <summary> The direction of search that led to the current point </summary> </member> <member name="P:Microsoft.ML.Runtime.Numeric.Optimizer.OptimizerState.Value"> <summary> The current function value </summary> </member> <member name="P:Microsoft.ML.Runtime.Numeric.Optimizer.OptimizerState.LastValue"> <summary> The function value at the last point </summary> </member> <member name="P:Microsoft.ML.Runtime.Numeric.Optimizer.OptimizerState.Iter"> <summary> The number of iterations so far </summary> </member> <member name="P:Microsoft.ML.Runtime.Numeric.Optimizer.OptimizerState.GradientCalculations"> <summary> The number of completed gradient calculations in the current iteration. </summary> <remarks>This is updated in derived classes, since they may call Eval at different times.</remarks> </member> <member name="F:Microsoft.ML.Runtime.Numeric.Optimizer.OptimizerState._keepDense"> <summary> Whether the optimizer state will keep its internal vectors dense or not. This being true may lead to reduced load on the garbage collector, at the cost of possibly higher overall memory utilization. </summary> </member> <member name="M:Microsoft.ML.Runtime.Numeric.Optimizer.OptimizerState.CreateWorkingVector"> <summary> Convenience function to construct a working vector of length <c>Dim</c>. </summary> <returns></returns> </member> <member name="M:Microsoft.ML.Runtime.Numeric.Optimizer.OptimizerState.LineSearch(Microsoft.ML.Runtime.IChannel,System.Boolean)"> <summary> An implementation of the line search for the Wolfe conditions, from Nocedal & Wright </summary> </member> <member name="M:Microsoft.ML.Runtime.Numeric.Optimizer.OptimizerState.CubicInterp(Microsoft.ML.Runtime.Numeric.Optimizer.OptimizerState.PointValueDeriv,Microsoft.ML.Runtime.Numeric.Optimizer.OptimizerState.PointValueDeriv)"> <summary> Cubic interpolation routine from Nocedal and Wright </summary> <param name="p0">first point, with value and derivative</param> <param name="p1">second point, with value and derivative</param> <returns>local minimum of interpolating cubic polynomial</returns> </member> <member name="M:Microsoft.ML.Runtime.Numeric.Optimizer.Minimize(Microsoft.ML.Runtime.Numeric.DifferentiableFunction,Microsoft.ML.Runtime.Data.VBuffer{System.Single}@,System.Single,Microsoft.ML.Runtime.Data.VBuffer{System.Single}@,System.Single@)"> <summary> Minimize a function using the MeanRelativeImprovement termination criterion with the supplied tolerance level </summary> <param name="function">The function to minimize</param> <param name="initial">The initial point</param> <param name="tolerance">Convergence tolerance (smaller means more iterations, closer to exact optimum)</param> <param name="result">The point at the optimum</param> <param name="optimum">The optimum function value</param> <exception cref="T:Microsoft.ML.Runtime.Numeric.Optimizer.PrematureConvergenceException">Thrown if successive points are within numeric precision of each other, but termination condition is still unsatisfied.</exception> </member> <member name="M:Microsoft.ML.Runtime.Numeric.Optimizer.Minimize(Microsoft.ML.Runtime.Numeric.DifferentiableFunction,Microsoft.ML.Runtime.Data.VBuffer{System.Single}@,Microsoft.ML.Runtime.Data.VBuffer{System.Single}@,System.Single@)"> <summary> Minimize a function. </summary> <param name="function">The function to minimize</param> <param name="initial">The initial point</param> <param name="result">The point at the optimum</param> <param name="optimum">The optimum function value</param> <exception cref="T:Microsoft.ML.Runtime.Numeric.Optimizer.PrematureConvergenceException">Thrown if successive points are within numeric precision of each other, but termination condition is still unsatisfied.</exception> </member> <member name="M:Microsoft.ML.Runtime.Numeric.Optimizer.Minimize(Microsoft.ML.Runtime.Numeric.DifferentiableFunction,Microsoft.ML.Runtime.Data.VBuffer{System.Single}@,Microsoft.ML.Runtime.Numeric.ITerminationCriterion,Microsoft.ML.Runtime.Data.VBuffer{System.Single}@,System.Single@)"> <summary> Minimize a function using the supplied termination criterion </summary> <param name="function">The function to minimize</param> <param name="initial">The initial point</param> <param name="term">termination criterion to use</param> <param name="result">The point at the optimum</param> <param name="optimum">The optimum function value</param> <exception cref="T:Microsoft.ML.Runtime.Numeric.Optimizer.PrematureConvergenceException">Thrown if successive points are within numeric precision of each other, but termination condition is still unsatisfied.</exception> </member> <member name="T:Microsoft.ML.Runtime.Numeric.Optimizer.PrematureConvergenceException"> <summary> This exception is thrown if successive differences between points reach the limits of numerical stability, but the termination condition still hasn't been satisfied </summary> </member> <member name="M:Microsoft.ML.Runtime.Numeric.Optimizer.PrematureConvergenceException.#ctor(Microsoft.ML.Runtime.Numeric.Optimizer.OptimizerState,System.String)"> <summary> Makes a PrematureConvergenceException with the supplied message </summary> <param name="state">The OptimizerState when the exception was thrown</param> <param name="message">message for exception</param> </member> <member name="P:Microsoft.ML.Runtime.Numeric.Optimizer.Quiet"> <summary> If true, suppresses all output. </summary> </member> <member name="T:Microsoft.ML.Runtime.Numeric.DTerminate"> <summary> Delegate for functions that determine whether to terminate search. Called after each update. </summary> <param name="x">Current iterate</param> <returns>True if search should terminate</returns> </member> <member name="T:Microsoft.ML.Runtime.Numeric.SgdOptimizer"> <summary> Stochastic gradient descent with variations (minibatch, momentum, averaging). </summary> </member> <member name="P:Microsoft.ML.Runtime.Numeric.SgdOptimizer.BatchSize"> <summary> Size of minibatches </summary> </member> <member name="P:Microsoft.ML.Runtime.Numeric.SgdOptimizer.Momentum"> <summary> Momentum parameter </summary> </member> <member name="P:Microsoft.ML.Runtime.Numeric.SgdOptimizer.T0"> <summary> Base of step size schedule s_t = 1 / (t0 + f(t)) </summary> </member> <member name="F:Microsoft.ML.Runtime.Numeric.SgdOptimizer._terminate"> <summary> Termination criterion </summary> </member> <member name="P:Microsoft.ML.Runtime.Numeric.SgdOptimizer.Averaging"> <summary> If true, iterates are averaged </summary> </member> <member name="P:Microsoft.ML.Runtime.Numeric.SgdOptimizer.RateSchedule"> <summary> Gets/Sets rate schedule type </summary> </member> <member name="P:Microsoft.ML.Runtime.Numeric.SgdOptimizer.MaxSteps"> <summary> Gets/Sets maximum number of steps. Set to 0 for no max </summary> </member> <member name="T:Microsoft.ML.Runtime.Numeric.SgdOptimizer.RateScheduleType"> <summary> Annealing schedule for learning rate </summary> </member> <member name="F:Microsoft.ML.Runtime.Numeric.SgdOptimizer.RateScheduleType.Constant"> <summary> r_t = 1 / t0 </summary> </member> <member name="F:Microsoft.ML.Runtime.Numeric.SgdOptimizer.RateScheduleType.Sqrt"> <summary> r_t = 1 / (t0 + sqrt(t)) </summary> </member> <member name="F:Microsoft.ML.Runtime.Numeric.SgdOptimizer.RateScheduleType.Linear"> <summary> r_t = 1 / (t0 + t) </summary> </member> <member name="M:Microsoft.ML.Runtime.Numeric.SgdOptimizer.#ctor(Microsoft.ML.Runtime.Numeric.DTerminate,Microsoft.ML.Runtime.Numeric.SgdOptimizer.RateScheduleType,System.Boolean,System.Single,System.Int32,System.Single,System.Int32)"> <summary> Creates SGDOptimizer and sets optimization parameters </summary> <param name="terminate">Termination criterion</param> <param name="rateSchedule">Annealing schedule type for learning rate</param> <param name="averaging">If true, all iterates are averaged</param> <param name="t0">Base for learning rate schedule</param> <param name="batchSize">Average this number of stochastic gradients for each update</param> <param name="momentum">Momentum parameter</param> <param name="maxSteps">Maximum number of updates (0 for no max)</param> </member> <member name="T:Microsoft.ML.Runtime.Numeric.SgdOptimizer.DStochasticGradient"> <summary> Delegate for functions to query stochastic gradient at a point </summary> <param name="x">Point at which to evaluate</param> <param name="grad">Vector to be filled in with gradient</param> </member> <member name="M:Microsoft.ML.Runtime.Numeric.SgdOptimizer.Minimize(Microsoft.ML.Runtime.Numeric.SgdOptimizer.DStochasticGradient,Microsoft.ML.Runtime.Data.VBuffer{System.Single}@,Microsoft.ML.Runtime.Data.VBuffer{System.Single}@)"> <summary> Minimize the function represented by <paramref name="f"/>. </summary> <param name="f">Stochastic gradients of function to minimize</param> <param name="initial">Initial point</param> <param name="result">Approximate minimum of <paramref name="f"/></param> </member> <member name="T:Microsoft.ML.Runtime.Numeric.GDOptimizer"> <summary> Deterministic gradient descent with line search </summary> </member> <member name="P:Microsoft.ML.Runtime.Numeric.GDOptimizer.LineSearch"> <summary> Line search to use. </summary> </member> <member name="P:Microsoft.ML.Runtime.Numeric.GDOptimizer.MaxSteps"> <summary> Gets/Sets maximum number of steps. Set to 0 for no max. </summary> </member> <member name="P:Microsoft.ML.Runtime.Numeric.GDOptimizer.Terminate"> <summary> Gets/sets termination criterion. </summary> </member> <member name="P:Microsoft.ML.Runtime.Numeric.GDOptimizer.UseCG"> <summary> Gets/sets whether to use nonlinear conjugate gradient. </summary> </member> <member name="M:Microsoft.ML.Runtime.Numeric.GDOptimizer.#ctor(Microsoft.ML.Runtime.Numeric.DTerminate,Microsoft.ML.Runtime.Numeric.IDiffLineSearch,System.Boolean,System.Int32)"> <summary> Makes a new GDOptimizer with the given optimization parameters </summary> <param name="terminate">Termination criterion</param> <param name="lineSearch">Line search to use</param> <param name="maxSteps">Maximum number of updates</param> <param name="useCG">Use Cubic interpolation line search or Backtracking line search with Armijo condition</param> </member> <member name="M:Microsoft.ML.Runtime.Numeric.GDOptimizer.Minimize(Microsoft.ML.Runtime.Numeric.DifferentiableFunction,Microsoft.ML.Runtime.Data.VBuffer{System.Single}@,Microsoft.ML.Runtime.Data.VBuffer{System.Single}@)"> <summary> Finds approximate minimum of the function </summary> <param name="function">Function to minimize</param> <param name="initial">Initial point</param> <param name="result">Approximate minimum</param> </member> <member name="T:Microsoft.ML.Runtime.Numeric.TerminateTester"> <summary> Terminates the optimization if NA value appears in result or no progress is made. </summary> </member> <member name="M:Microsoft.ML.Runtime.Numeric.TerminateTester.ShouldTerminate(Microsoft.ML.Runtime.Data.VBuffer{System.Single}@,Microsoft.ML.Runtime.Data.VBuffer{System.Single}@)"> <summary> Test whether the optimization should terminate. Returns true if x contains NA or +/-Inf or x equals xprev. </summary> <param name="x">The current value.</param> <param name="xprev">The value from the previous iteration.</param> <returns>True if the optimization routine should terminate at this iteration.</returns> </member> <member name="P:Microsoft.ML.Runtime.Learners.LinearTrainerBase`1.ShuffleData"> <summary> Whether data is to be shuffled every epoch. </summary> </member> <member name="P:Microsoft.ML.Runtime.Learners.LinearTrainerBase`1.WeightArraySize"> <summary> Gets the size of weights and bias array. For binary classification and regression, this is 1. For multi-class classification, this equals the number of classes. </summary> </member> <member name="M:Microsoft.ML.Runtime.Learners.LinearTrainerBase`1.PrepareDataFromTrainingExamples(Microsoft.ML.Runtime.IChannel,Microsoft.ML.Runtime.Data.RoleMappedData)"> <summary> This method ensures that the data meets the requirements of this trainer and its subclasses, injects necessary transforms, and throws if it couldn't meet them. </summary> </member> <member name="M:Microsoft.ML.Runtime.Learners.SdcaTrainerBase`1.InitializeConvergenceMetrics(System.String[]@,System.Double[]@)"> <summary> Returns the names of the metrics reported by <see cref="M:Microsoft.ML.Runtime.Learners.SdcaTrainerBase`1.CheckConvergence(Microsoft.ML.Runtime.IProgressChannel,System.Int32,Microsoft.ML.Runtime.Training.FloatLabelCursor.Factory,Microsoft.ML.Runtime.Learners.SdcaTrainerBase{`0}.DualsTableBase,Microsoft.ML.Runtime.Learners.SdcaTrainerBase{`0}.IdToIdxLookup,Microsoft.ML.Runtime.Data.VBuffer{System.Single}[],Microsoft.ML.Runtime.Data.VBuffer{System.Single}[],System.Single[],System.Single[],System.Single[],System.Single[],System.Int64,System.Double[],System.Double@,System.Int32@)"/>, as well as the initial values. </summary> </member> <member name="M:Microsoft.ML.Runtime.Learners.SdcaTrainerBase`1.TrainWithoutLock(Microsoft.ML.Runtime.IProgressChannelProvider,Microsoft.ML.Runtime.Training.FloatLabelCursor.Factory,Microsoft.ML.Runtime.IRandom,Microsoft.ML.Runtime.Learners.SdcaTrainerBase{`0}.IdToIdxLookup,System.Int32,Microsoft.ML.Runtime.Learners.SdcaTrainerBase{`0}.DualsTableBase,System.Single[],System.Single[],System.Single,Microsoft.ML.Runtime.Data.VBuffer{System.Single}[],System.Single[],Microsoft.ML.Runtime.Data.VBuffer{System.Single}[],System.Single[],System.Single[])"> <summary> Train the SDCA optimizer with one iteration over the entire training examples. </summary> <param name="progress">The progress reporting channel.</param> <param name="cursorFactory">The cursor factory to create cursors over the training examples.</param> <param name="rand"> The random number generator to generate random numbers for randomized shuffling of the training examples. It may be null. When it is null, the training examples are not shuffled and are cursored in its original order. </param> <param name="idToIdx"> The id to index mapping. May be null. If it is null, the index is given by the corresponding lower bits of the id. </param> <param name="numThreads">The number of threads used in parallel training. It is used in computing the dual update.</param> <param name="duals"> The dual variables. For binary classification and regression, there is one dual variable per row. For multiclass classification, there is one dual variable per class per row. </param> <param name="biasReg">The array containing regularized bias terms. For binary classification or regression, it contains only a single value. For multiclass classification its size equals the number of classes.</param> <param name="invariants"> The dual updates invariants. It may be null. If not null, it holds an array of pre-computed numerical quantities that depend on the training example label and features, not the value of dual variables. </param> <param name="lambdaNInv">The precomputed numerical quantity 1 / (l2Const * (count of training examples)).</param> <param name="weights"> The weights array. For binary classification or regression, it consists of only one VBuffer. For multiclass classification, its size equals the number of classes. </param> <param name="biasUnreg"> The array containing unregularized bias terms. For binary classification or regression, it contains only a single value. For multiclass classification its size equals the number of classes. </param> <param name="l1IntermediateWeights"> The array holding the intermediate weights prior to making L1 shrinkage adjustment. It is null iff l1Threshold is zero. Otherwise, for binary classification or regression, it consists of only one VBuffer; for multiclass classification, its size equals the number of classes. </param> <param name="l1IntermediateBias"> The array holding the intermediate bias prior to making L1 shrinkage adjustment. It is null iff l1Threshold is zero. Otherwise, for binary classification or regression, it consists of only one value; for multiclass classification, its size equals the number of classes. </param> <param name="featureNormSquared"> The array holding the pre-computed squared L2-norm of features for each training example. It may be null. It is always null for binary classification and regression because this quantity is not needed. </param> </member> <member name="M:Microsoft.ML.Runtime.Learners.SdcaTrainerBase`1.CheckConvergence(Microsoft.ML.Runtime.IProgressChannel,System.Int32,Microsoft.ML.Runtime.Training.FloatLabelCursor.Factory,Microsoft.ML.Runtime.Learners.SdcaTrainerBase{`0}.DualsTableBase,Microsoft.ML.Runtime.Learners.SdcaTrainerBase{`0}.IdToIdxLookup,Microsoft.ML.Runtime.Data.VBuffer{System.Single}[],Microsoft.ML.Runtime.Data.VBuffer{System.Single}[],System.Single[],System.Single[],System.Single[],System.Single[],System.Int64,System.Double[],System.Double@,System.Int32@)"> <summary> Returns whether the algorithm converged, and also populates the <paramref name="metrics"/> (which is expected to be parallel to the names returned by <see cref="M:Microsoft.ML.Runtime.Learners.SdcaTrainerBase`1.InitializeConvergenceMetrics(System.String[]@,System.Double[]@)"/>). When called, the <paramref name="metrics"/> is expected to hold the previously reported values. </summary> <param name="pch">The progress reporting channel.</param> <param name="iter">The iteration number, zero based.</param> <param name="cursorFactory">The cursor factory to create cursors over the training data.</param> <param name="duals"> The dual variables. For binary classification and regression, there is one dual variable per row. For multiclass classification, there is one dual variable per class per row. </param> <param name="idToIdx"> The id to index mapping. May be null. If it is null, the index is given by the corresponding lower bits of the id. </param> <param name="weights"> The weights array. For binary classification or regression, it consists of only one VBuffer. For multiclass classification, its size equals the number of classes. </param> <param name="bestWeights"> The weights array that corresponds to the best model obtained from the training iterations thus far. </param> <param name="biasUnreg"> The array containing unregularized bias terms. For binary classification or regression, it contains only a single value. For multiclass classification its size equals the number of classes. </param> <param name="bestBiasUnreg"> The array containing unregularized bias terms corresponding to the best model obtained from the training iterations thus far. For binary classification or regression, it contains only a single value. For multiclass classification its size equals the number of classes. </param> <param name="biasReg"> The array containing regularized bias terms. For binary classification or regression, it contains only a single value. For multiclass classification its size equals the number of classes. </param> <param name="bestBiasReg"> The array containing regularized bias terms corresponding to the best model obtained from the training iterations thus far. For binary classification or regression, it contains only a single value. For multiclass classification its size equals the number of classes. </param> <param name="count"> The count of (valid) training examples. Bad training examples are excluded from this count. </param> <param name="metrics"> The array of metrics for progress reporting. </param> <param name="bestPrimalLoss"> The primal loss function value corresponding to the best model obtained thus far. </param> <param name="bestIter">The iteration number when the best model is obtained.</param> <returns>Whether the optimization has converged.</returns> </member> <member name="T:Microsoft.ML.Runtime.Learners.SdcaTrainerBase`1.DualsTableBase"> <summary> Encapsulates the common functionality of storing and retrieving the dual variables. </summary> </member> <member name="T:Microsoft.ML.Runtime.Learners.SdcaTrainerBase`1.StandardArrayDualsTable"> <summary> Implementation of <see cref="T:Microsoft.ML.Runtime.Learners.SdcaTrainerBase`1.DualsTableBase"/> using a standard array. </summary> </member> <member name="T:Microsoft.ML.Runtime.Learners.SdcaTrainerBase`1.BigArrayDualsTable"> <summary> Implementation of <see cref="T:Microsoft.ML.Runtime.Learners.SdcaTrainerBase`1.DualsTableBase"/> using a big array. </summary> </member> <member name="M:Microsoft.ML.Runtime.Learners.SdcaTrainerBase`1.GetIndexFromIdGetter(Microsoft.ML.Runtime.Learners.SdcaTrainerBase{`0}.IdToIdxLookup)"> <summary> Returns a function delegate to retrieve index from id. This is to avoid redundant conditional branches in the tight loop of training. </summary> </member> <member name="M:Microsoft.ML.Runtime.Learners.SdcaTrainerBase`1.GetIndexFromIdAndRowGetter(Microsoft.ML.Runtime.Learners.SdcaTrainerBase{`0}.IdToIdxLookup)"> <summary> Returns a function delegate to retrieve index from id and row. Only works if the cursor is not shuffled. This is to avoid redundant conditional branches in the tight loop of training. </summary> </member> <member name="T:Microsoft.ML.Runtime.Learners.SdcaTrainerBase`1.IdToIdxLookup"> <summary> A hash table data structure to store Id of type <see cref="T:Microsoft.ML.Runtime.Data.UInt128"/>, and accommodates size larger than 2 billion. This class is an extension based on BCL. Two operations are supported: adding and retrieving an id with asymptotically constant complexity. The bucket size are prime numbers, starting from 3 and grows to the next prime larger than double the current size until it reaches the maximum possible size. When a table growth is triggered, the table growing operation initializes a new larger bucket and rehash the existing entries to the new bucket. Such operation has an expected complexity proportional to the size. </summary> </member> <member name="P:Microsoft.ML.Runtime.Learners.SdcaTrainerBase`1.IdToIdxLookup.Count"> <summary> Gets the count of id entries. </summary> </member> <member name="M:Microsoft.ML.Runtime.Learners.SdcaTrainerBase`1.IdToIdxLookup.#ctor(System.Int64)"> <summary> Initializes an instance of the <see cref="T:Microsoft.ML.Runtime.Learners.SdcaTrainerBase`1.IdToIdxLookup"/> class with the specified size. </summary> </member> <member name="M:Microsoft.ML.Runtime.Learners.SdcaTrainerBase`1.IdToIdxLookup.Add(Microsoft.ML.Runtime.Data.UInt128)"> <summary> Make sure the given id is in this lookup table and return the index of the id. </summary> </member> <member name="M:Microsoft.ML.Runtime.Learners.SdcaTrainerBase`1.IdToIdxLookup.TryGetIndex(Microsoft.ML.Runtime.Data.UInt128,System.Int64@)"> <summary> Find the index of the given id. Returns a bool representing if id is present. Index outputs the index that the id, -1 otherwise. </summary> </member> <member name="M:Microsoft.ML.Runtime.Learners.SdcaTrainerBase`1.IdToIdxLookup.GetIndexCore(Microsoft.ML.Runtime.Data.UInt128,System.Int64)"> <summary> Return the index of value, -1 if it is not present. </summary> </member> <member name="M:Microsoft.ML.Runtime.Learners.SdcaTrainerBase`1.IdToIdxLookup.AddCore(Microsoft.ML.Runtime.Data.UInt128,System.Int64)"> <summary> Adds the value as a TItem. Does not check whether the TItem is already present. Returns the index of the added value. </summary> </member> <member name="T:Microsoft.ML.Runtime.Learners.CompensatedSum"> <summary> Sum with underflow compensation for better numerical stability. </summary> </member> <member name="T:Microsoft.ML.Runtime.Learners.Sdca"> <summary> A component to train an SDCA model. </summary> <summary> A component to train an SDCA model. </summary> <summary> A component to train an SDCA model. </summary> </member> <member name="P:Microsoft.ML.Runtime.Learners.LinearPredictor.Weights2"> <summary> The predictor's feature weight coefficients.</summary> </member> <member name="P:Microsoft.ML.Runtime.Learners.LinearPredictor.Bias"> <summary> The predictor's bias term.</summary> </member> <member name="M:Microsoft.ML.Runtime.Learners.LinearPredictor.#ctor(Microsoft.ML.Runtime.IHostEnvironment,System.String,Microsoft.ML.Runtime.Data.VBuffer{System.Single}@,System.Single)"> <summary> Constructs a new linear predictor. </summary> <param name="env">The host environment.</param> <param name="name">Component name.</param> <param name="weights">The weights for the linear predictor. Note that this will take ownership of the <see cref="T:Microsoft.ML.Runtime.Data.VBuffer`1"/>.</param> <param name="bias">The bias added to every output score.</param> </member> <member name="M:Microsoft.ML.Runtime.Learners.LinearPredictor.CombineParameters(System.Collections.Generic.IList{Microsoft.ML.Runtime.Internal.Internallearn.IParameterMixer{System.Single}},Microsoft.ML.Runtime.Data.VBuffer{System.Single}@,System.Single@)"> <summary> Combine a bunch of models into one by averaging parameters </summary> </member> <member name="M:Microsoft.ML.Runtime.Learners.LinearBinaryPredictor.#ctor(Microsoft.ML.Runtime.IHostEnvironment,Microsoft.ML.Runtime.Data.VBuffer{System.Single}@,System.Single,Microsoft.ML.Runtime.Learners.LinearModelStatistics)"> <summary> Constructs a new linear binary predictor. </summary> <param name="env">The host environment.</param> <param name="weights">The weights for the linear predictor. Note that this will take ownership of the <see cref="T:Microsoft.ML.Runtime.Data.VBuffer`1"/>.</param> <param name="bias">The bias added to every output score.</param> <param name="stats"></param> </member> <member name="M:Microsoft.ML.Runtime.Learners.LinearBinaryPredictor.CombineParameters(System.Collections.Generic.IList{Microsoft.ML.Runtime.Internal.Internallearn.IParameterMixer{System.Single}})"> <summary> Combine a bunch of models into one by averaging parameters </summary> </member> <member name="M:Microsoft.ML.Runtime.Learners.LinearBinaryPredictor.GetSummaryInKeyValuePairs(Microsoft.ML.Runtime.Data.RoleMappedSchema)"> <inheritdoc/> </member> <member name="M:Microsoft.ML.Runtime.Learners.RegressionPredictor.SaveAsIni(System.IO.TextWriter,Microsoft.ML.Runtime.Data.RoleMappedSchema,Microsoft.ML.Runtime.Internal.Calibration.ICalibrator)"> <summary> Output the INI model to a given writer </summary> </member> <member name="M:Microsoft.ML.Runtime.Learners.LinearRegressionPredictor.#ctor(Microsoft.ML.Runtime.IHostEnvironment,Microsoft.ML.Runtime.Data.VBuffer{System.Single}@,System.Single)"> <summary> Constructs a new linear regression predictor. </summary> <param name="env">The host environment.</param> <param name="weights">The weights for the linear predictor. Note that this will take ownership of the <see cref="T:Microsoft.ML.Runtime.Data.VBuffer`1"/>.</param> <param name="bias">The bias added to every output score.</param> </member> <member name="M:Microsoft.ML.Runtime.Learners.LinearRegressionPredictor.CombineParameters(System.Collections.Generic.IList{Microsoft.ML.Runtime.Internal.Internallearn.IParameterMixer{System.Single}})"> <summary> Combine a bunch of models into one by averaging parameters </summary> </member> <member name="M:Microsoft.ML.Runtime.Learners.LinearRegressionPredictor.GetSummaryInKeyValuePairs(Microsoft.ML.Runtime.Data.RoleMappedSchema)"> <inheritdoc/> </member> <member name="M:Microsoft.ML.Runtime.Learners.PoissonRegressionPredictor.CombineParameters(System.Collections.Generic.IList{Microsoft.ML.Runtime.Internal.Internallearn.IParameterMixer{System.Single}})"> <summary> Combine a bunch of models into one by averaging parameters </summary> </member> <member name="T:Microsoft.ML.Runtime.Learners.LinearPredictorUtils"> <summary> Helper methods for linear predictors </summary> </member> <member name="M:Microsoft.ML.Runtime.Learners.LinearPredictorUtils.SaveAsCode(System.IO.TextWriter,Microsoft.ML.Runtime.Data.VBuffer{System.Single}@,System.Single,Microsoft.ML.Runtime.Data.RoleMappedSchema,System.String)"> <summary> print the linear model as code </summary> </member> <member name="M:Microsoft.ML.Runtime.Learners.LinearPredictorUtils.FeatureNameAsCode(System.String,System.Int32)"> <summary> Ensure that feature name is a legitimate variable name </summary> </member> <member name="M:Microsoft.ML.Runtime.Learners.LinearPredictorUtils.LinearModelAsIni(Microsoft.ML.Runtime.Data.VBuffer{System.Single}@,System.Single,Microsoft.ML.Runtime.IPredictor,Microsoft.ML.Runtime.Data.RoleMappedSchema,Microsoft.ML.Runtime.Internal.Calibration.PlattCalibrator)"> <summary> Build a Bing TreeEnsemble .ini representation of the given predictor </summary> </member> <member name="M:Microsoft.ML.Runtime.Learners.LinearPredictorUtils.LinearModelAsText(System.String,System.String,System.String,Microsoft.ML.Runtime.Data.VBuffer{System.Single}@,System.Single,Microsoft.ML.Runtime.Data.RoleMappedSchema,Microsoft.ML.Runtime.Internal.Calibration.PlattCalibrator)"> <summary> Output the weights of a linear model to a given writer </summary> </member> <member name="M:Microsoft.ML.Runtime.Learners.LinearPredictorUtils.SaveLinearModelWeightsInKeyValuePairs(Microsoft.ML.Runtime.Data.VBuffer{System.Single}@,System.Single,Microsoft.ML.Runtime.Data.RoleMappedSchema,System.Collections.Generic.List{System.Collections.Generic.KeyValuePair{System.String,System.Object}})"> <summary> Output the weights of a linear model to key value pairs. </summary> </member> <member name="F:Microsoft.ML.Runtime.Learners.LbfgsTrainerBase`2.ArgumentsBase.Quiet"> <summary> Features must occur in at least this many instances to be included </summary> <remarks>If greater than 1, forces an initialization pass over the data</remarks> </member> <member name="F:Microsoft.ML.Runtime.Learners.LbfgsTrainerBase`2.ArgumentsBase.InitWtsDiameter"> <summary> Init Weights Diameter </summary> </member> <member name="F:Microsoft.ML.Runtime.Learners.LbfgsTrainerBase`2.ArgumentsBase.NumThreads"> <summary> Number of threads. Null means use the number of processors. </summary> </member> <member name="M:Microsoft.ML.Runtime.Learners.LbfgsTrainerBase`2.InitializeWeightsSgd(Microsoft.ML.Runtime.IChannel,Microsoft.ML.Runtime.Training.FloatLabelCursor.Factory)"> <summary> Initialize weights by running SGD up to specified tolerance. </summary> </member> <member name="M:Microsoft.ML.Runtime.Learners.LbfgsTrainerBase`2.Train(Microsoft.ML.Runtime.Data.RoleMappedData)"> <summary> The basic training calls the optimizer </summary> </member> <member name="M:Microsoft.ML.Runtime.Learners.LbfgsTrainerBase`2.DifferentiableFunction(Microsoft.ML.Runtime.Data.VBuffer{System.Single}@,Microsoft.ML.Runtime.Data.VBuffer{System.Single}@,Microsoft.ML.Runtime.IProgressChannelProvider)"> <summary> The gradient being used by the optimizer </summary> </member> <member name="M:Microsoft.ML.Runtime.Learners.LbfgsTrainerBase`2.DifferentiableFunctionMultithreaded(Microsoft.ML.Runtime.Data.VBuffer{System.Single}@,Microsoft.ML.Runtime.Data.VBuffer{System.Single}@,Microsoft.ML.Runtime.IProgressChannel)"> <summary> Batch-parallel optimizer </summary> <remarks> REVIEW: consider getting rid of multithread-targeted members Using TPL, the distinction between Multithreaded and Sequential implementations is unnecessary </remarks> </member> <member name="T:Microsoft.ML.Runtime.Learners.LogisticRegression"> <summary> A component to train a logistic regression model. </summary> </member> <member name="M:Microsoft.ML.Runtime.Learners.MulticlassLogisticRegressionPredictor.#ctor(Microsoft.ML.Runtime.IHostEnvironment,Microsoft.ML.Runtime.Data.VBuffer{System.Single}[],System.Single[],System.Int32,System.Int32,System.String[],Microsoft.ML.Runtime.Learners.LinearModelStatistics)"> <summary> Initializes a new instance of the <see cref="T:Microsoft.ML.Runtime.Learners.MulticlassLogisticRegressionPredictor"/> class. This constructor is called by <see cref="T:Microsoft.ML.Runtime.Learners.SdcaMultiClassTrainer"/> to create the predictor. </summary> <param name="env">The host environment.</param> <param name="weights">The array of weights vectors. It should contain <paramref name="numClasses"/> weights.</param> <param name="bias">The array of biases. It should contain contain <paramref name="numClasses"/> weights.</param> <param name="numClasses">The number of classes for multi-class classification. Must be at least 2.</param> <param name="numFeatures">The logical length of the feature vector.</param> <param name="labelNames">The optional label names. If specified not null, it should have the same length as <paramref name="numClasses"/>.</param> <param name="stats">The model statistics.</param> </member> <member name="M:Microsoft.ML.Runtime.Learners.MulticlassLogisticRegressionPredictor.SaveAsText(System.IO.TextWriter,Microsoft.ML.Runtime.Data.RoleMappedSchema)"> <summary> Output the text model to a given writer </summary> </member> <member name="M:Microsoft.ML.Runtime.Learners.MulticlassLogisticRegressionPredictor.GetSummaryInKeyValuePairs(Microsoft.ML.Runtime.Data.RoleMappedSchema)"> <inheritdoc/> </member> <member name="M:Microsoft.ML.Runtime.Learners.MulticlassLogisticRegressionPredictor.SaveAsCode(System.IO.TextWriter,Microsoft.ML.Runtime.Data.RoleMappedSchema)"> <summary> Output the text model to a given writer </summary> </member> <member name="M:Microsoft.ML.Runtime.Learners.MulticlassLogisticRegressionPredictor.GetWeights(Microsoft.ML.Runtime.Data.VBuffer{System.Single}[]@,System.Int32@)"> <summary> Copies the weight vector for each class into a set of buffers. </summary> <param name="weights">A possibly reusable set of vectors, which will be expanded as necessary to accomodate the data.</param> <param name="numClasses">Set to the rank, which is also the logical length of <paramref name="weights"/>.</param> </member> <member name="T:Microsoft.ML.Runtime.Learners.CoefficientStatistics"> <summary> Represents a coefficient statistics object. </summary> </member> <member name="T:Microsoft.ML.Runtime.Learners.LinearModelStatistics"> <summary> The statistics for linear predictor. </summary> </member> <member name="M:Microsoft.ML.Runtime.Learners.LinearModelStatistics.GetCoefficientStatistics(Microsoft.ML.Runtime.Learners.LinearBinaryPredictor,Microsoft.ML.Runtime.Data.RoleMappedSchema,System.Int32)"> <summary> Gets the coefficient statistics as an object. </summary> </member> <member name="T:Microsoft.ML.Runtime.Learners.Ova.Arguments"> <summary> Arguments passed to OVA. </summary> </member> <member name="M:Microsoft.ML.Runtime.Learners.OvaPredictor.Create(Microsoft.ML.Runtime.IHost,Microsoft.ML.Runtime.IPredictorProducing{System.Single}[])"> <summary> Create a OVA predictor from an array of predictors. </summary> </member> <member name="T:Microsoft.ML.Runtime.Learners.Pkpd.Arguments"> <summary> Arguments passed to PKPD. </summary> </member> <member name="M:Microsoft.ML.Runtime.Learners.OlsLinearRegressionTrainer.ProbClamp(System.Double)"> <summary> In several calculations, we calculate probabilities or other quantities that should range from 0 to 1, but because of numerical imprecision may, in entirely innocent circumstances, land outside that range. This is a helper function to "reclamp" this to sane ranges. </summary> <param name="p">The quantity that should be clamped from 0 to 1</param> <returns>Either p, or 0 or 1 if it was outside the range 0 to 1</returns> </member> <member name="M:Microsoft.ML.Runtime.Learners.OlsLinearRegressionTrainer.Mkl.Pptrf(Microsoft.ML.Runtime.Learners.OlsLinearRegressionTrainer.Mkl.Layout,Microsoft.ML.Runtime.Learners.OlsLinearRegressionTrainer.Mkl.UpLo,System.Int32,System.Double[])"> <summary> Cholesky factorization of a symmetric positive-definite double matrix, using packed storage. The <c>pptrf</c> name comes from LAPACK, and means PositiveDefinitePackedTriangular(Cholesky)Factorize. </summary> <param name="layout">The storage order of this matrix</param> <param name="uplo">Whether the passed in matrix stores the upper or lower triangular part of the matrix</param> <param name="n">The order of the matrix</param> <param name="ap">An array with at least n*(n+1)/2 entries, containing the packed upper/lower part of the matrix. The triangular factorization is stored in this passed in matrix, when it returns. (U^T U or L L^T depending on whether this was upper or lower.)</param> </member> <member name="M:Microsoft.ML.Runtime.Learners.OlsLinearRegressionTrainer.Mkl.Pptrs(Microsoft.ML.Runtime.Learners.OlsLinearRegressionTrainer.Mkl.Layout,Microsoft.ML.Runtime.Learners.OlsLinearRegressionTrainer.Mkl.UpLo,System.Int32,System.Int32,System.Double[],System.Double[],System.Int32)"> <summary> Solves a system of linear equations, using the Cholesky factorization of the <c>A</c> matrix, typically returned from <c>Pptrf</c>. The <c>pptrf</c> name comes from LAPACK, and means PositiveDefinitePackedTriangular(Cholesky)Solve. </summary> <param name="layout">The storage order of this matrix</param> <param name="uplo">Whether the passed in matrix stores the upper or lower triangular part of the matrix</param> <param name="n">The order of the matrix</param> <param name="nrhs">The number of columns in the right hand side matrix</param> <param name="ap">An array with at least n*(n+1)/2 entries, containing a Cholesky factorization of the matrix in the linear equation.</param> <param name="b">The right hand side</param> <param name="ldb">The major index step size (typically for row major order, the number of columns, or something larger)</param> </member> <member name="M:Microsoft.ML.Runtime.Learners.OlsLinearRegressionTrainer.Mkl.Pptri(Microsoft.ML.Runtime.Learners.OlsLinearRegressionTrainer.Mkl.Layout,Microsoft.ML.Runtime.Learners.OlsLinearRegressionTrainer.Mkl.UpLo,System.Int32,System.Double[])"> <summary> Compute the inverse of a matrix, using the Cholesky factorization of the <c>A</c> matrix, typically returned from <c>Pptrf</c>. The <c>pptrf</c> name comes from LAPACK, and means PositiveDefinitePackedTriangular(Cholesky)Invert. </summary> <param name="layout">The storage order of this matrix</param> <param name="uplo">Whether the passed in matrix stores the upper or lower triangular part of the matrix</param> <param name="n">The order of the matrix</param> <param name="ap">An array with at least n*(n+1)/2 entries, containing a Cholesky factorization of the matrix in the linear equation. The inverse is returned in this array.</param> </member> <member name="T:Microsoft.ML.Runtime.Learners.OlsLinearRegressionPredictor"> <summary> A linear predictor for which per parameter significance statistics are available. </summary> </member> <member name="M:Microsoft.ML.Runtime.Learners.OlsLinearRegressionPredictor.GetVersionInfo"> <summary> Version information to be saved in binary format </summary> </member> <member name="P:Microsoft.ML.Runtime.Learners.OlsLinearRegressionPredictor.RSquared"> <summary> The coefficient of determination. </summary> </member> <member name="P:Microsoft.ML.Runtime.Learners.OlsLinearRegressionPredictor.RSquaredAdjusted"> <summary> The adjusted coefficient of determination. It is only possible to produce an adjusted R-squared if there are more examples than parameters in the model plus one. If this condition is not met, this value will be <c>NaN</c>. </summary> </member> <member name="P:Microsoft.ML.Runtime.Learners.OlsLinearRegressionPredictor.HasStatistics"> <summary> Whether the model has per parameter statistics. This is false iff <see cref="P:Microsoft.ML.Runtime.Learners.OlsLinearRegressionPredictor.StandardErrors"/>, <see cref="P:Microsoft.ML.Runtime.Learners.OlsLinearRegressionPredictor.TValues"/>, and <see cref="P:Microsoft.ML.Runtime.Learners.OlsLinearRegressionPredictor.PValues"/> are all null. A model may not have per parameter statistics because either there were not more examples than parameters in the model, or because they were explicitly suppressed in training by setting <see cref="F:Microsoft.ML.Runtime.Learners.OlsLinearRegressionTrainer.Arguments.PerParameterSignificance"/> to false. </summary> </member> <member name="P:Microsoft.ML.Runtime.Learners.OlsLinearRegressionPredictor.StandardErrors"> <summary> The standard error per model parameter, where the first corresponds to the bias, and all subsequent correspond to each weight in turn. This is <c>null</c> if and only if <see cref="P:Microsoft.ML.Runtime.Learners.OlsLinearRegressionPredictor.HasStatistics"/> is <c>false</c>. </summary> </member> <member name="P:Microsoft.ML.Runtime.Learners.OlsLinearRegressionPredictor.TValues"> <summary> t-Statistic values corresponding to each of the model standard errors. This is <c>null</c> if and only if <see cref="P:Microsoft.ML.Runtime.Learners.OlsLinearRegressionPredictor.HasStatistics"/> is <c>false</c>. </summary> </member> <member name="P:Microsoft.ML.Runtime.Learners.OlsLinearRegressionPredictor.PValues"> <summary> p-values corresponding to each of the model standard errors. This is <c>null</c> if and only if <see cref="P:Microsoft.ML.Runtime.Learners.OlsLinearRegressionPredictor.HasStatistics"/> is <c>false</c>. </summary> </member> <member name="M:Microsoft.ML.Runtime.Learners.AveragedLinearTrainer`2.AveragedMargin(Microsoft.ML.Runtime.Data.VBuffer{System.Single}@)"> <summary> Return the raw margin from the decision hyperplane </summary> </member> <member name="M:Microsoft.ML.Runtime.Learners.AveragedLinearTrainer`2.IncrementAverageNonLazy"> <summary> Add current weights and bias to average weights/bias. </summary> </member> <member name="T:Microsoft.ML.Runtime.Learners.AveragedPerceptronTrainer"> <summary> This is an averaged perceptron classifier. Configurable subcomponents: - Loss function. By default, hinge loss (aka max-margin avgd perceptron) - Feature normalization. By default, rescaling between min and max values for every feature - Prediction calibration to produce probabilities. Off by default, if on, uses exponential (aka Platt) calibration. </summary> </member> <member name="T:Microsoft.ML.Runtime.Learners.LinearSvm"> <summary> Linear SVM that implements PEGASOS for training. See: http://ttic.uchicago.edu/~shai/papers/ShalevSiSr07.pdf </summary> </member> <member name="M:Microsoft.ML.Runtime.Learners.LinearSvm.Margin(Microsoft.ML.Runtime.Data.VBuffer{System.Single}@)"> <summary> Return the raw margin from the decision hyperplane </summary> </member> <member name="M:Microsoft.ML.Runtime.Learners.LinearSvm.ProcessDataInstance(Microsoft.ML.Runtime.IChannel,Microsoft.ML.Runtime.Data.VBuffer{System.Single}@,System.Single,System.Single)"> <summary> Observe an example and update weights if necessary </summary> </member> <member name="M:Microsoft.ML.Runtime.Learners.LinearSvm.UpdateWeights(Microsoft.ML.Runtime.Data.VBuffer{System.Single}@,System.Single)"> <summary> Updates the weights at the end of the batch. Since weightsUpdate can be an instance feature vector, this function should not change the contents of weightsUpdate. </summary> </member> <member name="M:Microsoft.ML.Runtime.Learners.OnlineGradientDescentTrainer.Arguments.#ctor"> <summary> Set defaults that vary from the base type. </summary> </member> <member name="M:Microsoft.ML.Runtime.Learners.OnlineLinearTrainer`2.ScaleWeights"> <summary> Propagates the <c>_weightsScale </c> to the weights vector. </summary> </member> <member name="M:Microsoft.ML.Runtime.Learners.OnlineLinearTrainer`2.ScaleWeightsIfNeeded"> <summary> Conditionally propagates the <c>_weightsScale</c> to the weights vector when it reaches a scale where additions to weights would start dropping too much precision. ("Too much" is mostly empirically defined.) </summary> </member> <member name="M:Microsoft.ML.Runtime.Learners.OnlineLinearTrainer`2.ProcessDataInstance(Microsoft.ML.Runtime.IChannel,Microsoft.ML.Runtime.Data.VBuffer{System.Single}@,System.Single,System.Single)"> <summary> This should be overridden by derived classes. This implementation simply increments _numIterExamples and dumps debug information to the console. </summary> </member> <member name="M:Microsoft.ML.Runtime.Learners.OnlineLinearTrainer`2.CurrentMargin(Microsoft.ML.Runtime.Data.VBuffer{System.Single}@)"> <summary> Return the raw margin from the decision hyperplane </summary> </member> <member name="T:Microsoft.ML.Runtime.Learners.SdcaMultiClassTrainer"> <summary> SDCA linear multiclass trainer. </summary> </member> <member name="M:Microsoft.ML.Runtime.Learners.SdcaMultiClassTrainer.TrainWithoutLock(Microsoft.ML.Runtime.IProgressChannelProvider,Microsoft.ML.Runtime.Training.FloatLabelCursor.Factory,Microsoft.ML.Runtime.IRandom,Microsoft.ML.Runtime.Learners.SdcaTrainerBase{Microsoft.ML.Runtime.IPredictorProducing{Microsoft.ML.Runtime.Data.VBuffer{System.Single}}}.IdToIdxLookup,System.Int32,Microsoft.ML.Runtime.Learners.SdcaTrainerBase{Microsoft.ML.Runtime.IPredictorProducing{Microsoft.ML.Runtime.Data.VBuffer{System.Single}}}.DualsTableBase,System.Single[],System.Single[],System.Single,Microsoft.ML.Runtime.Data.VBuffer{System.Single}[],System.Single[],Microsoft.ML.Runtime.Data.VBuffer{System.Single}[],System.Single[],System.Single[])"> <inheritdoc/> </member> <member name="M:Microsoft.ML.Runtime.Learners.SdcaMultiClassTrainer.CheckConvergence(Microsoft.ML.Runtime.IProgressChannel,System.Int32,Microsoft.ML.Runtime.Training.FloatLabelCursor.Factory,Microsoft.ML.Runtime.Learners.SdcaTrainerBase{Microsoft.ML.Runtime.IPredictorProducing{Microsoft.ML.Runtime.Data.VBuffer{System.Single}}}.DualsTableBase,Microsoft.ML.Runtime.Learners.SdcaTrainerBase{Microsoft.ML.Runtime.IPredictorProducing{Microsoft.ML.Runtime.Data.VBuffer{System.Single}}}.IdToIdxLookup,Microsoft.ML.Runtime.Data.VBuffer{System.Single}[],Microsoft.ML.Runtime.Data.VBuffer{System.Single}[],System.Single[],System.Single[],System.Single[],System.Single[],System.Int64,System.Double[],System.Double@,System.Int32@)"> <inheritdoc/> </member> <member name="T:Microsoft.ML.Runtime.Learners.RandomTrainer"> <summary> A trainer that trains a predictor that returns random values </summary> </member> <member name="T:Microsoft.ML.Runtime.Learners.RandomPredictor"> <summary> The predictor implements the Predict() interface. The predictor returns a uniform random probability and classification assignment. </summary> </member> <member name="M:Microsoft.ML.Runtime.Learners.RandomPredictor.#ctor(Microsoft.ML.Runtime.IHostEnvironment,Microsoft.ML.Runtime.Model.ModelLoadContext)"> <summary> Load the predictor from the binary format. </summary> </member> <member name="M:Microsoft.ML.Runtime.Learners.RandomPredictor.SaveCore(Microsoft.ML.Runtime.Model.ModelSaveContext)"> <summary> Save the predictor in the binary format. </summary> <param name="ctx"></param> </member> </members> </doc> |