MMMSummaryFactory.waterfall#

MMMSummaryFactory.waterfall(hdi_probs=None, dims=None, original_scale=True, output_format=None)[source]#

Create waterfall decomposition summary DataFrame.

Summarizes per-component mean contributions (averaged over date) with mean, median, HDI bounds, and percentage of total contribution.

Parameters:
hdi_probssequence of float, optional

HDI probability levels (default: uses factory default)

dimsdict, optional

Dimension filters, e.g. {"geo": ["CA"]}.

original_scalebool, default True

Whether to use original-scale contributions.

output_format{“pandas”, “polars”}, optional

Output DataFrame format (default: uses factory default)

Returns:
pd.DataFrame or pl.DataFrame

Summary DataFrame with columns:

  • component: Contribution component name

  • mean, median: Point estimates of mean per-period contributions

  • abs_error_{prob}_lower/upper: HDI bounds on mean contributions

  • pct_of_total: Mean share of total (from per-draw ratios; see share_*)

  • share_mean, share_median: Share point estimates with uncertainty

  • share_abs_error_{prob}_lower/upper: HDI bounds on share of total

  • <custom_dims>: One column per custom dimension when present

Notes

mean/abs_error_* describe mean per-period contributions (matching waterfall() bar heights). share_* and pct_of_total describe each component’s share of the total per posterior draw (ratio HDIs, not derived from component-total HDIs).