Libraries/Microsoft.ML.PipelineInference.xml
|
<?xml version="1.0"?>
<doc> <assembly> <name>Microsoft.ML.PipelineInference</name> </assembly> <members> <member name="T:Microsoft.ML.Runtime.PipelineInference.AutoInference"> <summary> Class for generating potential recipes/pipelines, testing them, and zeroing in on the best ones. For now, only works with maximizing metrics (AUC, Accuracy, etc.). </summary> </member> <member name="T:Microsoft.ML.Runtime.PipelineInference.AutoInference.LevelDependencyMap"> <summary> Alias to refer to this construct by an easy name. </summary> </member> <member name="T:Microsoft.ML.Runtime.PipelineInference.AutoInference.DependencyMap"> <summary> Alias to refer to this construct by an easy name. </summary> </member> <member name="T:Microsoft.ML.Runtime.PipelineInference.AutoInference.SupportedMetric"> <summary> AutoInference will support metrics as they are added here. </summary> </member> <member name="T:Microsoft.ML.Runtime.PipelineInference.AutoInference.EntryPointGraphDef"> <summary> Class for encapsulating an entrypoint experiment graph and keeping track of the input and output nodes. </summary> </member> <member name="M:Microsoft.ML.Runtime.PipelineInference.AutoInference.EntryPointGraphDef.GetSubgraphFirstNodeDataVarName(Microsoft.ML.Runtime.IExceptionContext)"> <summary> Get the name of the variable asssigned to the Data or Training Data input, based on what is the first node of the subgraph. A better way to do this would be with a ICanBeSubGraphFirstNode common interface between ITransformInput and ITrainerInputs and a custom deserializer. </summary> </member> <member name="T:Microsoft.ML.Runtime.PipelineInference.AutoInference.RunSummary"> <summary> Class containing some information about an exectuted pipeline. These are analogous to IRunResult for smart sweepers. </summary> </member> <member name="T:Microsoft.ML.Runtime.PipelineInference.AutoInference.AutoMlMlState"> <summary> Class that holds state for an autoML search-in-progress. Should be able to resume search, given this object. </summary> </member> <member name="M:Microsoft.ML.Runtime.PipelineInference.AutoInference.AutoMlMlState.ComputeSearchSpace(System.Int32,Microsoft.ML.Runtime.PipelineInference.RecipeInference.SuggestedRecipe.SuggestedLearner[],System.Func{Microsoft.ML.Runtime.Data.IDataView,Microsoft.ML.Runtime.PipelineInference.TransformInference.Arguments,Microsoft.ML.Runtime.PipelineInference.TransformInference.SuggestedTransform[]})"> <summary> Search space is transforms X learners X hyperparameters. </summary> </member> <member name="M:Microsoft.ML.Runtime.PipelineInference.AutoInference.InferPipelines(Microsoft.ML.Runtime.IHostEnvironment,Microsoft.ML.Runtime.PipelineInference.PipelineOptimizerBase,Microsoft.ML.Runtime.Data.IDataView,Microsoft.ML.Runtime.Data.IDataView,System.Int32,System.Int32,Microsoft.ML.Runtime.PipelineInference.AutoInference.SupportedMetric,Microsoft.ML.Runtime.PipelineInference.PipelinePattern@,Microsoft.ML.Runtime.PipelineInference.ITerminator,Microsoft.ML.Runtime.EntryPoints.MacroUtils.TrainerKinds)"> <summary> The InferPipelines methods are just public portals to the internal function that handle different types of data being passed in: training IDataView, path to training file, or train and test files. </summary> </member> <member name="T:Microsoft.ML.Runtime.PipelineInference.DefaultsEngine"> <summary> AutoML engine that goes through all lerners using defaults. Need to decide how to handle which transforms to try. </summary> </member> <member name="T:Microsoft.ML.Runtime.PipelineInference.UniformRandomEngine"> <summary> Example class of an autoML engine (a pipeline optimizer) that simply tries random enumeration. If we use a third-party solution for autoML, we can just implement a new wrapper for it as a PipelineOptimizerBase, and use our existing autoML body code to take advantage of it. This design should allow for easy development of new autoML methods. </summary> </member> <member name="M:Microsoft.ML.Runtime.PipelineInference.AutoMlUtils.AreColumnsConsistent(Microsoft.ML.Runtime.PipelineInference.TransformInference.SuggestedTransform[],Microsoft.ML.Runtime.PipelineInference.AutoInference.DependencyMap)"> <summary> Using the dependencyMapping and included transforms, determines whether every transform present only consumes columns produced by a lower- or same-level transform, or existed in the original dataset. Note, a column could be produced by a transform on the same level, such as in multipart (atomic group) transforms. </summary> </member> <member name="M:Microsoft.ML.Runtime.PipelineInference.AutoMlUtils.ValidationWrapper(Microsoft.ML.Runtime.PipelineInference.TransformInference.SuggestedTransform[],Microsoft.ML.Runtime.PipelineInference.AutoInference.DependencyMap)"> <summary> Simple wrapper which allows the call signature to match the signature needed for the PipelineOptimizerBase interface. </summary> </member> <member name="M:Microsoft.ML.Runtime.PipelineInference.AutoMlUtils.GetExcludedColumnIndices(Microsoft.ML.Runtime.PipelineInference.TransformInference.SuggestedTransform[],Microsoft.ML.Runtime.Data.IDataView,Microsoft.ML.Runtime.PipelineInference.AutoInference.DependencyMap)"> <summary> Using the dependencyMapping and included transforms, computes which subset of columns in dataSample will be present in the final transformed dataset when only the transforms present are applied. </summary> </member> <member name="M:Microsoft.ML.Runtime.PipelineInference.AutoMlUtils.BuildAtomicIdDependencyGraph(Microsoft.ML.Runtime.PipelineInference.TransformInference.SuggestedTransform[])"> <summary> Builds dependency dictionary of which atomic groups depend on which. Assumes that a column will be produced by the highest level possible producing transform. </summary> <param name="allTransforms">The set of possible transforms for a dataset for all levels.</param> </member> <member name="M:Microsoft.ML.Runtime.PipelineInference.AutoMlUtils.GetFinalFeatureConcat(Microsoft.ML.Runtime.IHostEnvironment,Microsoft.ML.Runtime.Data.IDataView,System.Int32[],System.Int32,System.Int32)"> <summary> Gets a final transform to concatenate all numeric columns into a "Features" vector column. Note: May return empty set if Features column already present and is only relevant numeric column. (In other words, if there would be nothing for that concatenate transform to do.) </summary> </member> <member name="M:Microsoft.ML.Runtime.PipelineInference.AutoMlUtils.GetFinalFeatureConcat(Microsoft.ML.Runtime.IHostEnvironment,Microsoft.ML.Runtime.Data.IDataView,Microsoft.ML.Runtime.PipelineInference.AutoInference.DependencyMap,Microsoft.ML.Runtime.PipelineInference.TransformInference.SuggestedTransform[],Microsoft.ML.Runtime.PipelineInference.TransformInference.SuggestedTransform[])"> <summary> Exposed version of the method. </summary> </member> <member name="M:Microsoft.ML.Runtime.PipelineInference.AutoMlUtils.ComputeColumnResponsibilities(Microsoft.ML.Runtime.Data.IDataView,Microsoft.ML.Runtime.PipelineInference.TransformInference.SuggestedTransform[])"> <summary> Creates a dictionary mapping column names to the transforms which could have produced them. </summary> </member> <member name="M:Microsoft.ML.Runtime.PipelineInference.AutoMlUtils.UpdateProperties(System.Object,Microsoft.ML.Runtime.EntryPoints.TlcModule.SweepableParamAttribute[])"> <summary> Updates properties of entryPointObj instance based on the values in sweepParams </summary> </member> <member name="M:Microsoft.ML.Runtime.PipelineInference.AutoMlUtils.PopulateSweepableParams(Microsoft.ML.Runtime.PipelineInference.RecipeInference.SuggestedRecipe.SuggestedLearner)"> <summary> Updates properties of entryPointObj instance based on the values in sweepParams </summary> </member> <member name="M:Microsoft.ML.Runtime.PipelineInference.AutoMlUtils.ConvertToSweepArgumentStrings(Microsoft.ML.Runtime.EntryPoints.TlcModule.SweepableParamAttribute[])"> <summary> Method to convert set of sweepable hyperparameters into strings of a format understood by the current smart hyperparameter sweepers. </summary> </member> <member name="T:Microsoft.ML.Runtime.PipelineInference.ColumnGroupingInference"> <summary> This class incapsulates logic for grouping together the inferred columns of the text file based on their type and purpose, and generating column names. </summary> </member> <member name="T:Microsoft.ML.Runtime.PipelineInference.ColumnGroupingInference.GroupingColumn"> <summary> This is effectively a merger of <see cref="T:Microsoft.ML.Runtime.PipelineInference.PurposeInference.Column"/> and a <see cref="T:Microsoft.ML.Runtime.PipelineInference.ColumnTypeInference.Column"/> with support for vector-value columns. </summary> </member> <member name="M:Microsoft.ML.Runtime.PipelineInference.ColumnGroupingInference.InferGroupingAndNames(Microsoft.ML.Runtime.IHostEnvironment,System.Boolean,Microsoft.ML.Runtime.PipelineInference.ColumnTypeInference.Column[],Microsoft.ML.Runtime.PipelineInference.PurposeInference.Column[])"> <summary> Group together the single-valued columns with the same type and purpose and generate column names. </summary> <param name="env">The host environment to use.</param> <param name="hasHeader">Whether the original file had a header. If yes, the <see cref="F:Microsoft.ML.Runtime.PipelineInference.ColumnTypeInference.Column.SuggestedName"/> fields are used to generate the column names, otherwise they are ignored.</param> <param name="types">The (detected) column types.</param> <param name="purposes">The (detected) column purposes. Must be parallel to <paramref name="types"/>.</param> <returns>The struct containing an array of grouped columns specifications.</returns> </member> <member name="M:Microsoft.ML.Runtime.PipelineInference.ColumnGroupingInference.GetRange(System.Int32[])"> <summary> Generates a range selector from the array of indices. </summary> </member> <member name="M:Microsoft.ML.Runtime.PipelineInference.ColumnGroupingInference.GetRange(System.String)"> <summary> Generates individual indices from a string range selector. </summary> </member> <member name="T:Microsoft.ML.Runtime.PipelineInference.ColumnTypeInference"> <summary> This class incapsulates logic for automatic inference of column types for the text file. It also attempts to guess whether there is a header row. </summary> </member> <member name="T:Microsoft.ML.Runtime.PipelineInference.ColumnTypeInference.Experts"> <summary> Current design is as follows: there's a sequence of 'experts' that each look at all the columns. Every expert may or may not assign the 'answer' (suggested type) to a column. If the expert needs some information about the column (for example, the column values), this information is lazily calculated by the column object, not the expert itself, to allow the reuse of the same information by another expert. </summary> </member> <member name="M:Microsoft.ML.Runtime.PipelineInference.ColumnTypeInference.InferTextFileColumnTypes(Microsoft.ML.Runtime.IHostEnvironment,Microsoft.ML.Runtime.Data.IMultiStreamSource,Microsoft.ML.Runtime.PipelineInference.ColumnTypeInference.Arguments)"> <summary> Auto-detect column types of the file. </summary> </member> <member name="T:Microsoft.ML.Runtime.PipelineInference.DatasetFeatureInference"> <summary> Featurization ideas inspired from: http://aad.informatik.uni-freiburg.de/papers/15-NIPS-auto-sklearn-supplementary.pdf </summary> </member> <member name="T:Microsoft.ML.Runtime.PipelineInference.DatasetFeatureInference.ColumnSchema"> <summary> Meta features about column schema: 1) Number of features such as Text, Categorical, Numerical, counted in terms of slots. 2) Column meta data such as Name, DataKind, Purpose and slot indices. 3) Feature ratio: number of categorical columns to text feature columns, etc. 4) Count of columns and slots by column purpose. 5) Log of total feature count. </summary> </member> <member name="T:Microsoft.ML.Runtime.PipelineInference.DatasetFeatureInference.LabelFeatures"> <summary> Meta features for Label columns: 1) Column info such as name, data kind, column purpose, slot indices. 2) Stats about label counts such as Min, Max counts, standard deviation, mean, median, kurtosis, variance, min/max class probabilities. 3) Missing value count. </summary> </member> <member name="T:Microsoft.ML.Runtime.PipelineInference.DatasetFeatureInference.MissingValues"> <summary> Meta features about missing values across the dataset. </summary> </member> <member name="T:Microsoft.ML.Runtime.PipelineInference.DatasetFeatureInference.ColumnFeatures"> <summary> Meta features about columns: For numeric column types it caculates mean, variance, kurtosis, min, max, median, standard deviation. For non-numeric column types it caculates the same but over string length and number of spaces as seperate stats. </summary> </member> <member name="T:Microsoft.ML.Runtime.PipelineInference.ExperimentsGenerator.Sweep"> <summary> Holds the sweep and the trainer for sweep commands. </summary> </member> <member name="T:Microsoft.ML.Runtime.PipelineInference.ExperimentsGenerator.Pattern"> <summary> Pattern of the Sweep command. </summary> </member> <member name="T:Microsoft.ML.Runtime.PipelineInference.ExperimentsGenerator.TrainerSweeper"> <summary> Sweep parameters for the trainer. </summary> </member> <member name="T:Microsoft.ML.Runtime.PipelineInference.IPipelineOptimizer"> <summary> Interface that defines what an autoML engine looks like, so that it can plug into the autoML body defined in this file. Will allow us to change autoML techniques or use third- party services as needed, just by swapping out the pipeline optimizer used. </summary> </member> <member name="T:Microsoft.ML.Runtime.PipelineInference.ITerminator"> <summary> Interface defining various stopping criteria for pipeline sweeps. This could include number of total iterations, compute time, budget expended, etc. </summary> </member> <member name="T:Microsoft.ML.Runtime.PipelineInference.PipelinePattern"> <summary> A runnable pipeline. Contains a learner and set of transforms, along with a RunSummary if it has already been exectued. </summary> </member> <member name="M:Microsoft.ML.Runtime.PipelineInference.PipelinePattern.ToEntryPointGraph(Microsoft.ML.Runtime.Experiment)"> <summary> Constructs an entrypoint graph from the current pipeline. </summary> </member> <member name="M:Microsoft.ML.Runtime.PipelineInference.PipelinePattern.ToString"> <summary> This method will return some indentifying string for the pipeline, based on transforms, learner, and (eventually) hyperparameters. </summary> </member> <member name="M:Microsoft.ML.Runtime.PipelineInference.PipelinePattern.RunTrainTestExperiment(Microsoft.ML.Runtime.Data.IDataView,Microsoft.ML.Runtime.Data.IDataView,Microsoft.ML.Runtime.PipelineInference.AutoInference.SupportedMetric,Microsoft.ML.Runtime.EntryPoints.MacroUtils.TrainerKinds)"> <summary> Runs a train-test experiment on the current pipeline, through entrypoints. </summary> </member> <member name="T:Microsoft.ML.Runtime.PipelineInference.PurposeInference"> <summary> Automatic inference of column purposes for the data view. This is used in the context of text import wizard, but can be used outside as well. </summary> </member> <member name="T:Microsoft.ML.Runtime.PipelineInference.PurposeInference.IPurposeInferenceExpert"> <summary> The design is the same as for <see cref="T:Microsoft.ML.Runtime.PipelineInference.ColumnTypeInference"/>: there's a sequence of 'experts' that each look at all the columns. Every expert may or may not assign the 'answer' (suggested purpose) to a column. If the expert needs some information about the column (for example, the column values), this information is lazily calculated by the column object, not the expert itself, to allow the reuse of the same information by another expert. </summary> </member> <member name="M:Microsoft.ML.Runtime.PipelineInference.PurposeInference.InferPurposes(Microsoft.ML.Runtime.IHostEnvironment,Microsoft.ML.Runtime.Data.IDataView,System.Collections.Generic.IEnumerable{System.Int32},Microsoft.ML.Runtime.PipelineInference.PurposeInference.Arguments)"> <summary> Auto-detect purpose for the data view columns. </summary> <param name="env">The host environment to use.</param> <param name="data">The data to use for inference.</param> <param name="columnIndices">Indices of columns that we're interested in.</param> <param name="args">Additional arguments to inference.</param> <returns>The result includes the array of auto-detected column purposes.</returns> </member> <member name="M:Microsoft.ML.Runtime.PipelineInference.RecipeInference.AllowedLearners(Microsoft.ML.Runtime.IHostEnvironment,Microsoft.ML.Runtime.EntryPoints.MacroUtils.TrainerKinds)"> <summary> Given a predictor type returns a set of all permissible learners (with their sweeper params, if defined). </summary> <returns>Array of viable learners.</returns> </member> <member name="T:Microsoft.ML.Runtime.PipelineInference.TextFileContents"> <summary> Utilities for various heuristics against text files. Currently, separator inference and column count detection. </summary> </member> <member name="M:Microsoft.ML.Runtime.PipelineInference.TextFileContents.TrySplitColumns(Microsoft.ML.Runtime.IHostEnvironment,Microsoft.ML.Runtime.Data.IMultiStreamSource,System.String[],System.Nullable{System.Boolean},System.Nullable{System.Boolean},System.Boolean)"> <summary> Attempt to detect text loader arguments. The algorithm selects the first 'acceptable' set: the one that recognizes the same number of columns in at least <see cref="F:Microsoft.ML.Runtime.PipelineInference.TextFileContents.UniformColumnCountThreshold"/> of the sample's lines, and this number of columns is more than 1. We sweep on separator, allow sparse and allow quote parameter. </summary> </member> <member name="T:Microsoft.ML.Runtime.PipelineInference.TextFileSample"> <summary> This class holds an in-memory sample of the text file, and serves as an <see cref="T:Microsoft.ML.Runtime.Data.IMultiStreamSource"/> proxy to it. </summary> </member> <member name="M:Microsoft.ML.Runtime.PipelineInference.TextFileSample.CreateFromFullFile(Microsoft.ML.Runtime.IHostEnvironment,System.String)"> <summary> Create a <see cref="T:Microsoft.ML.Runtime.PipelineInference.TextFileSample"/> by reading multiple chunks from the file (or other source) and then stitching them together. The algorithm is as follows: 0. If the source is not seekable, revert to <see cref="M:Microsoft.ML.Runtime.PipelineInference.TextFileSample.CreateFromHead(System.String)"/>. 1. If the file length is less than 2 * <see cref="F:Microsoft.ML.Runtime.PipelineInference.TextFileSample.BufferSizeMb"/>, revert to <see cref="M:Microsoft.ML.Runtime.PipelineInference.TextFileSample.CreateFromHead(System.String)"/>. 2. Read first <see cref="F:Microsoft.ML.Runtime.PipelineInference.TextFileSample.FirstChunkSizeMb"/> MB chunk. Determine average line length in the chunk. 3. Determine how large one chunk should be, and how many chunks there should be, to end up with <see cref="F:Microsoft.ML.Runtime.PipelineInference.TextFileSample.BufferSizeMb"/> * <see cref="F:Microsoft.ML.Runtime.PipelineInference.TextFileSample.OversamplingRate"/> MB worth of lines. 4. Determine seek locations and read the chunks. 5. Stitch and return a <see cref="T:Microsoft.ML.Runtime.PipelineInference.TextFileSample"/>. </summary> </member> <member name="M:Microsoft.ML.Runtime.PipelineInference.TextFileSample.CreateFromHead(System.String)"> <summary> Create a <see cref="T:Microsoft.ML.Runtime.PipelineInference.TextFileSample"/> by reading one chunk from the beginning. </summary> </member> <member name="M:Microsoft.ML.Runtime.PipelineInference.TextFileSample.StitchChunks(System.Boolean,System.Byte[][])"> <summary> Given an array of chunks of the text file, of which the first chunk is the head, this method trims incomplete lines from the beginning and end of each chunk (except that it doesn't trim the beginning of the first chunk and end of last chunk if we read whole file), then joins the rest together to form a final byte buffer and returns a <see cref="T:Microsoft.ML.Runtime.PipelineInference.TextFileSample"/> wrapped around it. </summary> <param name="wholeFile">did we read whole file</param> <param name="chunks">chunks of data</param> <returns></returns> </member> <member name="M:Microsoft.ML.Runtime.PipelineInference.TextFileSample.IsEncodingOkForSampling(System.Byte[])"> <summary> Detect whether we can auto-detect EOL characters without parsing. If we do, we can cheaply sample from different file locations and trim the partial strings. The encodings that pass the test are UTF8 and all single-byte encodings. </summary> </member> <member name="T:Microsoft.ML.Runtime.PipelineInference.TransformInference"> <summary> Auto-generate set of transforms for the data view, given the purposes of specified columns. The design is the same as for <see cref="T:Microsoft.ML.Runtime.PipelineInference.ColumnTypeInference"/>: there's a sequence of 'experts' that each look at all the columns. Every expert may or may not suggest additional transforms. If the expert needs some information about the column (for example, the column values), this information is lazily calculated by the column object, not the expert itself, to allow the reuse of the same information by another expert. </summary> </member> <member name="F:Microsoft.ML.Runtime.PipelineInference.TransformInference.Arguments.EstimatedSampleFraction"> <summary> Relative size of the inspected data view vs. the 'real' data size. </summary> </member> <member name="M:Microsoft.ML.Runtime.PipelineInference.TransformInference.InferTransforms(Microsoft.ML.Runtime.IHostEnvironment,Microsoft.ML.Runtime.Data.IDataView,Microsoft.ML.Runtime.PipelineInference.PurposeInference.Column[],Microsoft.ML.Runtime.PipelineInference.TransformInference.Arguments)"> <summary> Automatically infer transforms for the data view </summary> </member> </members> </doc> |