internal CountTargetEncodingEstimator(IHostEnvironment env, string labelColumnName, CountTableEstimator.SharedColumnOptions[] columnOptions,
CountTableBuilderBase countTableBuilder, int numberOfBits = HashingEstimator.Defaults.NumberOfBits,
bool combine = HashingEstimator.Defaults.Combine, uint hashingSeed = HashingEstimator.Defaults.Seed)
: this(env, new CountTableEstimator(env, labelColumnName, countTableBuilder,
columnOptions.Select(col => new CountTableEstimator.SharedColumnOptions(col.Name, col.Name, col.PriorCoefficient, col.LaplaceScale, col.Seed)).ToArray()),
columnOptions, numberOfBits, combine, hashingSeed)
{
}
/// <summary>
/// Transforms a categorical column into a set of features that includes the count of each label class,
/// the log-odds for each label class and the back-off indicator.
/// </summary>
/// <param name="catalog">The transforms catalog.</param>
/// <param name="outputColumnName">Name of the column resulting from the transformation of <paramref name="inputColumnName"/>.</param>
/// <param name="inputColumnName">Name of the column to transform. If set to <see langword="null"/>, the value of the <paramref name="outputColumnName"/> will be used as source.</param>
/// <param name="labelColumn">The name of the label column.</param>
/// <param name="builder">The builder that creates the count tables from the training data.</param>
/// <param name="priorCoefficient">The coefficient with which to apply the prior smoothing to the features.</param>
/// <param name="laplaceScale">The Laplacian noise diversity/scale-parameter. Recommended values are between 0 and 1. Note that the noise
/// will only be applied if the estimator is part of an <see cref="EstimatorChain{TLastTransformer}"/>, when fitting the next estimator in the chain.</param>
/// <param name="numberOfBits">The number of bits to hash the input into. Must be between 1 and 31, inclusive.</param>
/// <param name="combine">In case the input is a vector column, indicates whether the values should be combined into a single hash to create a single
/// count table, or be left as a vector of hashes with multiple count tables.</param>
/// <param name="hashingSeed">The seed used for hashing the input columns.</param>
public static CountTargetEncodingEstimator CountTargetEncode(this TransformsCatalog catalog, string outputColumnName, string inputColumnName = null,
string labelColumn = DefaultColumnNames.Label,
CountTableBuilderBase builder = null,
float priorCoefficient = CountTableTransformer.Defaults.PriorCoefficient,
float laplaceScale = CountTableTransformer.Defaults.LaplaceScale,
int numberOfBits = HashingEstimator.Defaults.NumberOfBits,
bool combine = HashingEstimator.Defaults.Combine,
uint hashingSeed = HashingEstimator.Defaults.Seed)
{
var env = CatalogUtils.GetEnvironment(catalog);
env.CheckNonEmpty(outputColumnName, nameof(outputColumnName));
inputColumnName = string.IsNullOrEmpty(inputColumnName) ? outputColumnName : inputColumnName;
builder = builder ?? new CMCountTableBuilder();
return(new CountTargetEncodingEstimator(env, labelColumn,
new[] { new CountTableEstimator.ColumnOptions(outputColumnName, inputColumnName, builder, priorCoefficient, laplaceScale) },
numberOfBits, combine, hashingSeed));
}
/// <summary>
/// Transforms a categorical column into a set of features that includes the count of each label class,
/// the log-odds for each label class and the back-off indicator.
/// </summary>
/// <param name="catalog">The transforms catalog.</param>
/// <param name="columns">The input and output columns.</param>
/// <param name="labelColumn">The name of the label column.</param>
/// <param name="builder">The builder that creates the count tables from the training data.</param>
/// <param name="priorCoefficient">The coefficient with which to apply the prior smoothing to the features.</param>
/// <param name="laplaceScale">The Laplacian noise diversity/scale-parameter. Recommended values are between 0 and 1. Note that the noise
/// will only be applied if the estimator is part of an <see cref="EstimatorChain{TLastTransformer}"/>, when fitting the next estimator in the chain.</param>
/// <param name="sharedTable">Indicates whether to keep counts for all columns and slots in one shared count table. If true, the keys in the count table
/// will include a hash of the column and slot indices.</param>
/// <param name="numberOfBits">The number of bits to hash the input into. Must be between 1 and 31, inclusive.</param>
/// <param name="combine">In case the input is a vector column, indicates whether the values should be combined into a single hash to create a single
/// count table, or be left as a vector of hashes with multiple count tables.</param>
/// <param name="hashingSeed">The seed used for hashing the input columns.</param>
/// <returns></returns>
public static CountTargetEncodingEstimator CountTargetEncode(this TransformsCatalog catalog,
InputOutputColumnPair[] columns, string labelColumn = DefaultColumnNames.Label,
CountTableBuilderBase builder = null,
float priorCoefficient = CountTableTransformer.Defaults.PriorCoefficient,
float laplaceScale = CountTableTransformer.Defaults.LaplaceScale,
bool sharedTable = CountTableTransformer.Defaults.SharedTable,
int numberOfBits = HashingEstimator.Defaults.NumberOfBits,
bool combine = HashingEstimator.Defaults.Combine,
uint hashingSeed = HashingEstimator.Defaults.Seed)
{
var env = CatalogUtils.GetEnvironment(catalog);
env.CheckValue(columns, nameof(columns));
builder = builder ?? new CMCountTableBuilder();
CountTargetEncodingEstimator estimator;
if (sharedTable)
{
var columnOptions = new CountTableEstimator.SharedColumnOptions[columns.Length];
for (int i = 0; i < columns.Length; i++)
{
columnOptions[i] = new CountTableEstimator.SharedColumnOptions(
columns[i].OutputColumnName, columns[i].InputColumnName, priorCoefficient, laplaceScale);
}
estimator = new CountTargetEncodingEstimator(env, labelColumn, columnOptions, builder, numberOfBits, combine, hashingSeed);
}
else
{
var columnOptions = new CountTableEstimator.ColumnOptions[columns.Length];
for (int i = 0; i < columns.Length; i++)
{
columnOptions[i] = new CountTableEstimator.ColumnOptions(
columns[i].OutputColumnName, columns[i].InputColumnName, builder, priorCoefficient, laplaceScale);
}
estimator = new CountTargetEncodingEstimator(env, labelColumn, columnOptions, numberOfBits: numberOfBits, combine: combine, hashingSeed: hashingSeed);
}
return(estimator);
}
public BagMultiCountTableBuilder(IHostEnvironment env, DataViewSchema.Column[] inputColumns, CountTableBuilderBase builder, long labelCardinality)
{
Contracts.CheckValue(env, nameof(env));
env.CheckNonEmpty(inputColumns, nameof(inputColumns));
_host = env.Register(LoaderSignature);
// REVIEW: how to disallow non-zero garbage bin for bag dict count table? Or maybe just ignore?
_builder = builder.GetInternalBuilder(labelCardinality);
_colCount = inputColumns.Length;
_slotCount = new int[_colCount];
for (int i = 0; i < _colCount; i++)
{
_slotCount[i] = inputColumns[i].Type.GetValueCount();
}
}
internal CountTableEstimator(IHostEnvironment env, string labelColumnName, CountTableBuilderBase countTableBuilder, params SharedColumnOptions[] columns)
: this(env, labelColumnName, columns)
{
_sharedBuilder = countTableBuilder;
}
public ColumnOptions(string name, string inputColumnName, CountTableBuilderBase countTableBuilder, float priorCoefficient = 1,
float laplaceScale = 0, int seed = 314489979)
: base(name, inputColumnName, priorCoefficient, laplaceScale, seed)
{
CountTableBuilder = countTableBuilder;
}
