aggregate_idata_time#

pymc_marketing.data.idata.utils.aggregate_idata_time(idata, period, method='sum')[source]#

Aggregate DataTree over time periods.

Parameters:

idataxr.DataTree: DataTree to aggregate
period{“original”, “weekly”, “monthly”, “quarterly”, “yearly”, “all_time”}: Time period to aggregate to. Use “original” for no aggregation (returns unchanged), “all_time” to aggregate over the entire time dimension (removes the date dimension).
method{“sum”, “mean”}, default “sum”: Aggregation method

Returns:

xr.DataTree: New DataTree with aggregated groups (or unchanged if period=”original”). Its root attrs record the aggregation as time_aggregation=period so that consumers of per-date quantities can tell the data apart from the original dates.

Notes

Each variable is reduced as it is, so a variable without a date dim passes through unchanged. A time-invariant MMM intercept is such a variable although the model adds it in every period: summing MMM contributions with this function counts it once for the whole window. Use aggregate_time(), which broadcasts it over date first.

Examples

>>> original = aggregate_idata_time(idata, "original")  # No aggregation
>>> monthly = aggregate_idata_time(idata, "monthly", method="sum")
>>> total = aggregate_idata_time(idata, "all_time", method="sum")