neu_morpho.measure.tables

Turn a measure DB into tidy tables, for cohort selection and comparison.

The notebook-facing half of measure: DataFrame in, DataFrame out, no printing or plotting. pandas and pyarrow (neu-morpho[measure]) are imported inside functions.

Variants: all is the whole body, minus_nucleus removes the nucleus, minus_soma_nucleus the whole cell body. Which matches depends on the other cohort: removing a compartment from one side that is present on the other is a silent bias.

Attributes

DEFAULT_COMPARTMENTS
VARIANTS

Functions

compare(left, right, columns, *[, labels, statistic])	One row per column: each cohort's statistic and the ratio between them.
load_bodies(db_path, *[, voxel_nm, dataset, compartments])	One row per body, with volume, cable, per-compartment volume and the variants.
load_segment_properties(base, *sources)	Wide table of every inline property across one or more properties sources.
log_histogram(df, column, *, lo, hi[, nbins, weight])	Log-spaced histogram as a DataFrame, with under/overflow rows.
merge_properties(bodies, props, *[, how])	Join properties onto a bodies table, keeping boolean columns BOOLEAN.
summarize(df, columns, *[, by, quantiles])	Percentile table for columns, optionally grouped by a column.
write_table(df, path) → str	Write parquet, which unlike csv round-trips a nullable integer column.

Module Contents

neu_morpho.measure.tables.compare(left, right, columns: Sequence[str], *, labels: Sequence[str] = ('left', 'right'), statistic: str = 'median')

One row per column: each cohort’s statistic and the ratio between them.

The ratio is of the statistics: the cohorts are unpaired.

neu_morpho.measure.tables.load_bodies(db_path: str, *, voxel_nm: float = 32.0, dataset: str | None = None, compartments: Mapping[int, str] | None = None)

One row per body, with volume, cable, per-compartment volume and the variants.

Opens the DB read-only, so it is safe against a run in flight. Bodies with no voxel count were never measured and are dropped.

neu_morpho.measure.tables.load_segment_properties(base: str, *sources: str)

Wide table of every inline property across one or more properties sources.

tags become a boolean column per bare tag and, for namespace:value tags, one column per namespace. A namespace with several values for one body keeps the first, and the whole tuple goes in <namespace>_all.

neu_morpho.measure.tables.log_histogram(df, column: str, *, lo: float, hi: float, nbins: int = 48, weight: str | None = None)

Log-spaced histogram as a DataFrame, with under/overflow rows.

The catch-alls keep the weight total exact; a fat overflow row means the range is wrong.

neu_morpho.measure.tables.merge_properties(bodies, props, *, how: str = 'left')

Join properties onto a bodies table, keeping boolean columns BOOLEAN.

Use this rather than bodies.merge(props, on='body_id'). A plain merge widens a bool column to object where a body has no properties row, and ~ on that column is arithmetic (~True is -2), so ~col.fillna(False) silently masks nothing. A body with no properties row gets False for every flag.

neu_morpho.measure.tables.summarize(df, columns: Sequence[str], *, by: str | None = None, quantiles: Sequence[float] = (0.1, 0.5, 0.9))

Percentile table for columns, optionally grouped by a column.

Percentiles rather than mean and sd: these distributions are heavy-tailed.

neu_morpho.measure.tables.write_table(df, path: str) → str

Write parquet, which unlike csv round-trips a nullable integer column.

neu_morpho.measure.tables.DEFAULT_COMPARTMENTS

neu_morpho.measure.tables.VARIANTS