Transformations

q.transform owns stateful transformations learned at the model-training boundary. Deterministic model-input calculations may run for all dates, while fitted transformations must learn only from fit partitions.

Split-aware inputs use q.dataset: transforms discover allowed fit rows from partition roles rather than assuming a partition name.

Namespace	Role
q.transform.impute	Missing-value indicators and imputation
q.transform.scale	Numeric normalization
q.transform.outlier	Outlier detection and treatment
q.transform.encode	Categorical representation
q.transform.reduction	Learned projection and decomposition
q.transform.selection	Selection of model-input columns

q.transform.Pipeline wraps sklearn.pipeline.Pipeline for split-aware QRT datasets. It fits sklearn transformers only on role-fit partitions, applies the learned state to every row, preserves aligned dataset components, and records fit provenance. Individual transform namespaces remain reserved for QRT-provided estimator adapters. Deterministic cleaning that learns no state belongs in q.data, not here.