KernelExpansionCatalog.ApproximatedKernelMap Method

Definition

Namespace:

Microsoft.ML

Assembly:

Microsoft.ML.Transforms.dll

Package:

Microsoft.ML v3.0.1

Create an ApproximatedKernelMappingEstimator that maps input vectors to a low dimensional feature space where inner products approximate a shift-invariant kernel function.

public static Microsoft.ML.Transforms.ApproximatedKernelMappingEstimator ApproximatedKernelMap (this Microsoft.ML.TransformsCatalog catalog, string outputColumnName, string inputColumnName = default, int rank = 1000, bool useCosAndSinBases = false, Microsoft.ML.Transforms.KernelBase generator = default, int? seed = default);

static member ApproximatedKernelMap : Microsoft.ML.TransformsCatalog * string * string * int * bool * Microsoft.ML.Transforms.KernelBase * Nullable<int> -> Microsoft.ML.Transforms.ApproximatedKernelMappingEstimator

<Extension()>
Public Function ApproximatedKernelMap (catalog As TransformsCatalog, outputColumnName As String, Optional inputColumnName As String = Nothing, Optional rank As Integer = 1000, Optional useCosAndSinBases As Boolean = false, Optional generator As KernelBase = Nothing, Optional seed As Nullable(Of Integer) = Nothing) As ApproximatedKernelMappingEstimator

Parameters

catalog

TransformsCatalog

The transform's catalog.

outputColumnName

String

Name of the column resulting from the transformation of inputColumnName. The data type on this column will be a known-sized vector of Single.

inputColumnName

String

Name of column to transform. If set to null, the value of the outputColumnName will be used as source. This estimator operates on known-sized vector of Single data type.

rank

Int32

The dimension of the feature space to map the input to.

useCosAndSinBases

Boolean

If true, use both of cos and sin basis functions to create two features for every random Fourier frequency. Otherwise, only cos bases would be used. Note that if set to true, the dimension of the output feature space will be 2*rank.

generator

KernelBase

The argument that indicates which kernel to use. The two available implementations are GaussianKernel and LaplacianKernel.

seed

Nullable<Int32>

The seed of the random number generator for generating the new features (if unspecified, the global random is used).

Returns

ApproximatedKernelMappingEstimator

Examples

using System;
using System.Collections.Generic;
using System.Linq;
using Microsoft.ML;
using Microsoft.ML.Data;
using Microsoft.ML.Transforms;

namespace Samples.Dynamic
{
    public static class ApproximatedKernelMap
    {
        // Transform feature vector to another non-linear space. See
        // https://people.eecs.berkeley.edu/~brecht/papers/07.rah.rec.nips.pdf.
        public static void Example()
        {
            // Create a new ML context, for ML.NET operations. It can be used for
            // exception tracking and logging, as well as the source of randomness.
            var mlContext = new MLContext();
            var samples = new List<DataPoint>()
            {
                new DataPoint(){ Features = new float[7] { 1, 1, 0, 0, 1, 0, 1} },
                new DataPoint(){ Features = new float[7] { 0, 0, 1, 0, 0, 1, 1} },
                new DataPoint(){ Features = new float[7] {-1, 1, 0,-1,-1, 0,-1} },
                new DataPoint(){ Features = new float[7] { 0,-1, 0, 1, 0,-1,-1} }
            };
            // Convert training data to IDataView, the general data type used in
            // ML.NET.
            var data = mlContext.Data.LoadFromEnumerable(samples);
            // ApproximatedKernel map takes data and maps it's to a random
            // low -dimensional space.
            var approximation = mlContext.Transforms.ApproximatedKernelMap(
                "Features", rank: 4, generator: new GaussianKernel(gamma: 0.7f),
                seed: 1);

            // Now we can transform the data and look at the output to confirm the
            // behavior of the estimator. This operation doesn't actually evaluate
            // data until we read the data below.
            var tansformer = approximation.Fit(data);
            var transformedData = tansformer.Transform(data);

            var column = transformedData.GetColumn<float[]>("Features").ToArray();
            foreach (var row in column)
                Console.WriteLine(string.Join(", ", row.Select(x => x.ToString(
                    "f4"))));

            // Expected output:
            // -0.0119, 0.5867, 0.4942,  0.7041
            //  0.4720, 0.5639, 0.4346,  0.2671
            // -0.2243, 0.7071, 0.7053, -0.1681
            //  0.0846, 0.5836, 0.6575,  0.0581
        }

        private class DataPoint
        {
            [VectorType(7)]
            public float[] Features { get; set; }
        }

    }
}