sdm.processing

Processor API

Processor	Base processor for tensor-aware table transformations.
InvertibleMixin	Extend a Processor by an inverse transformation.
EnsembleProcessor	Base processor for ensemble-aware table transformations.
EnsembleInvertibleMixin	Extend a EnsembleProcessor by an inverse transformation.
TargetAwareEnsembleProcessor	Base processor with access to unprocessed targets.
SeriesEnsembleProcessor	Base processor for consecutive, equally spaced time-series data.
Recipe	Processing contract around an external model boundary.

Common Processors

Identity	Return inputs unchanged.
Callable	Adapt a stateless callable to the Processor interface.
Sequential	Apply processors and callables to a table in sequence.
DropStypes	Remove all columns for specific semantic types.
StypeDispatch	Apply separate processor pipelines to columns grouped by semantic type.
TaskDispatch	Apply separate processors based on the semantic type of the target.
TableDispatch	Apply separate feature processors to task and related tables.
EnsembleProcessorAdapter	Adapt an ordinary processor to ensemble-aware processing.
Choice	Route each table or ensemble member through one selected option.
ToNumerical	Move categorical columns into the numerical block.
ShuffleColumns	Permute numerical columns and their names.
SelectColumns	Select a subset of columns for each semantic type.

Numerical Processors

Cast	Cast numerical columns to a floating-point dtype.
Clip	Clip numerical values to a fixed interval.
ClipQuantiles	Clamp each numerical column to its fitted quantile bounds.
ClipSigma	Two-stage z-score outlier clipping.
ImputeMean	Replace missing numerical values with fitted per-column means.
PowerTransform	Apply a Yeo-Johnson power transform to each numerical column.
QuantileTransform	Map numerical columns through their empirical quantiles.
RankGaussian	Map interpolated empirical mid-ranks to standard normal quantiles.
Standardize	Center and scale each numerical column.
RobustScale	Center and scale each numerical column with median and quantile range.
MinMaxScale	Scale each numerical column by its fitted min-max range.
FlipSign	Randomly negate numerical columns.
Asinh	Apply the inverse hyperbolic sine to numerical values.
DropConstantColumns	Remove non-informative numerical columns learned during fit.
PCA	Project numerical columns onto their principal components.
RandomProjection	Randomly project numerical columns.
Detrend	Remove a per-column linear trend over step coordinates.
Interpolate	Interpolate missing numerical values along the row dimension.

Categorical Processors

AlignCategories	Align categorical columns to categories observed during fitting.
ShuffleCategories	Permute the integer codes of categorical columns.
ImputeMode	Replace missing categorical values with fitted per-column modes.
AddCategoryCounts	Add the log row count of each categorical value.
OneHotEncode	One-hot encode categorical columns.

Text Processors

TFIDF	Encode text columns as character n-gram TF-IDF vectors.
SentenceTransformer	Transform text columns with a sentence_transformers.SentenceTransformer model.

Datetime Processors

AddCalendarFields

Add numerical calendar fields derived from datetime columns.

Post-Processors

AverageEstimators	Reduce the leading ensemble dimension of model outputs.
Softmax	Apply softmax to logits after temperature scaling.
SortQuantiles	Sort quantile predictions in ascending order.
ClipNonnegative	Clip outputs at zero for targets with nonnegative observations.