euromod_linking.methods.cells

Shared cell machinery for method methodologies.

Turns shock-table group strings into population cells on the input microdata: dimension label derivation, region-level collapse, op semantics, and period resolution. Used by every methodology; lives outside any single method.

Attributes

CELL_COL

Functions

apply_op(→ float)
collapse_shock_groups() → tuple[pandas.DataFrame, ...)	Rewrite shock groups to the data's granularity. Region codes finer than
data_region_len(→ int)	Representative code length of the data's region labels (0 = country).
prepare_cells(→ tuple[pandas.DataFrame, dict])	Add per-key label columns (_dim_<key>) + CELL_COL, labelling each row
resolve_period(→ str)	The period whose shocks to apply: params['period'] when given, else the
specs_in_shocks(→ dict[str, list[str]])	Group key -> sorted distinct value specs used across (non-constant) shocks.

Module Contents

euromod_linking.methods.cells.apply_op(current: float, op: str, value: float) → float

euromod_linking.methods.cells.collapse_shock_groups(shocks: pandas.DataFrame, used, region_len: int, allow_coarser: bool = False, intensive_metrics=frozenset()) → tuple[pandas.DataFrame, list[str]]

Rewrite shock groups to the data’s granularity. Region codes finer than the data are truncated to region_len (0 -> drop the region key); values are aggregated per (metric, collapsed group): grow/mult by mean, set/add by sum.

Region codes coarser than the data are an error by default (count targets cannot be disaggregated); with allow_coarser=True they are kept as-is for the caller to prefix-match (broadcast semantics — correct for scaling).

Why mean vs sum: growth rates and multipliers are intensive quantities — merging two NUTS-2 rates into their NUTS-1 parent takes an average, with equal weights, since without subregional population counts in the microdata there is no defensible weighting. Levels and absolute deltas (set/add) are extensive — subregional amounts add up.

intensive_metrics names metrics that are intensive whatever the op — rates, where a value is a share or a percentage-point change. Without it a set/add on a rate would be summed, so two NUTS-2 shocks of -3.5pp would become -7pp in their NUTS-1 parent: silently double the intended size.

euromod_linking.methods.cells.data_region_len(df: pandas.DataFrame, used) → int

Representative code length of the data’s region labels (0 = country).

euromod_linking.methods.cells.prepare_cells(df: pandas.DataFrame, used: dict[str, list[str]]) → tuple[pandas.DataFrame, dict]

Add per-key label columns (_dim_<key>) + CELL_COL, labelling each row with the value spec it satisfies. Returns (df, info).

euromod_linking.methods.cells.resolve_period(shocks: pandas.DataFrame, params: dict) → str

The period whose shocks to apply: params[‘period’] when given, else the table’s single distinct period; ambiguity is an error listing the options.

euromod_linking.methods.cells.specs_in_shocks(shocks: pandas.DataFrame) → dict[str, list[str]]

Group key -> sorted distinct value specs used across (non-constant) shocks.

Cells are defined by the shock table itself: the specs a scenario actually uses (e.g. deh: [“0-2”, “3-4”]) become that variable’s categories for this run. Nothing is recoded ahead of time, so any input variable works and the labels stay readable in the diagnostics.

euromod_linking.methods.cells.CELL_COL = '_cell'