spacr.regression_layout

Lossless long/wide conversions for regression predictor tables.

The screen pipeline historically receives one row per (well, gRNA) and the low-level estimators receive a conventional wide design matrix. These helpers make that boundary explicit and also accept count tables that arrive with one guide per column. Conversion is deliberately strict: a value is never silently selected when duplicate rows disagree.

Functions

infer_regression_layout(→ str)	Infer long only from the paired predictor/value columns.
long_to_wide_regression_data(→ pandas.DataFrame)	Pivot one row per observation/predictor to one predictor per column.
normalise_count_table_layout(→ tuple[pandas.DataFrame, ...)	Return a canonical long grna/count count table and its input layout.
wide_to_long_regression_data(→ pandas.DataFrame)	Melt one-predictor-per-column data to one row per predictor.

Module Contents

spacr.regression_layout.infer_regression_layout(frame: pandas.DataFrame, *, predictor_column: str = 'grna', value_column: str = 'count') → str

Infer long only from the paired predictor/value columns.

A table containing exactly one of the two columns is malformed rather than wide. Treating it as wide would accidentally melt the existing value column into a gRNA.

spacr.regression_layout.long_to_wide_regression_data(frame: pandas.DataFrame, *, index_columns: collections.abc.Sequence[str] | str = 'prc', predictor_column: str = 'grna', value_column: str = 'fraction', metadata_columns: collections.abc.Sequence[str] | None = None, fill_value: float = 0.0, predictor_prefix: str = '') → pandas.DataFrame

Pivot one row per observation/predictor to one predictor per column.

Metadata must be constant inside each observation, and duplicate observation/predictor rows must agree. Repeated identical rows (for example after a harmless join) are collapsed once; conflicting values are refused.

spacr.regression_layout.normalise_count_table_layout(frame: pandas.DataFrame, *, layout: str = 'auto', guide_column: str = 'grna', count_column: str = 'count', wide_predictor_columns: collections.abc.Sequence[str] | None = None) → tuple[pandas.DataFrame, str]

Return a canonical long grna/count count table and its input layout.

spacr.regression_layout.wide_to_long_regression_data(frame: pandas.DataFrame, *, predictor_columns: collections.abc.Sequence[str] | None = None, id_columns: collections.abc.Sequence[str] | None = None, predictor_name: str = 'grna', value_name: str = 'count', drop_zero: bool = False) → pandas.DataFrame

Melt one-predictor-per-column data to one row per predictor.

When predictor_columns is omitted, every numeric column not named as observation metadata is used. Non-numeric unclassified columns are rejected because guessing whether they are metadata or a predictor would change the model silently.