imputation_policy#
Back to L2 | Browse all axes | Browse all options
Axis
imputation_policyon sub-layerl2_d(layerl2).
Sub-layer#
l2_d
Axis metadata#
Default:
'em_factor'Sweepable: True
Status: operational
Operational status summary#
Operational: 6 option(s)
Future: 0 option(s)
Options#
em_factor – operational#
EM-factor imputation (McCracken-Ng default).
See em_factor function page for full documentation + parameters + standalone usage. Standalone: mf.functions.em_factor_impute_clean.
em_multivariate – operational#
Multivariate-Gaussian EM imputation.
See em_multivariate function page for full documentation + parameters + standalone usage. Standalone: mf.functions.em_multivariate_impute_clean.
mean – operational#
Replace missing cells with the per-series rolling mean.
See mean function page for full documentation + parameters + standalone usage. Standalone: mf.functions.mean_impute_clean.
forward_fill – operational#
Carry the last observed value forward.
See forward_fill function page for full documentation + parameters + standalone usage. Standalone: mf.functions.forward_fill_clean.
linear_interpolation – operational#
Linear interpolation between adjacent observations.
See linear_interpolation function page for full documentation + parameters + standalone usage. Standalone: mf.functions.linear_interpolate_clean.
none_propagate – operational#
Pass NaN through; downstream layers handle it.
Useful when the recipe wants L3 / L4 to see the missing pattern (e.g., for missingness-as-feature studies).
When to use
Studies that treat missingness as informative; or panels with no missings.
References
macroforecast design Part 2, L2: ‘preprocessing is the only layer with a strict A→B→C→D→E execution order; every cell follows the same pipeline.’
Related options: em_factor, mean, forward_fill
Last reviewed 2026-05-04 by macroforecast author.