relfm_tabular
RelFMTabular: standalone flat-table RelFM — single Concept, no relational joins.
Deliberately independent of PredictiveEstimator/RelFM’s Concept+Relationship
task abstraction. That shared abstraction requires context/validation to reference the
same Concept (validate_task’s schema-match check compares field names, which
default to the concept’s own name) — this class instead takes each split as its own
Concept directly, with no requirement that they share an identity, since there is no
join to align: each Concept’s full row set (identity + features + label, when present)
is the flat table.
Forecasting:
— forecast(domain=...) does need domain to carry pre-populated future
(item_id, time) skeleton rows (no label column), same convention as
predictions(domain=...). The forecast lands exactly on domain’s own
(item_id, time) rows — there is no horizon parameter, each series may ask
for its own dates. Output properties on prediction_concept:
predicted_value (point estimate) plus, when quantile_levels is set,
one quantile_<level> property per level (e.g. quantile_10) — plain
attribute access.
Classification/regression and forecasting have disjoint constructor parameters
(clamp_min/clamp_max/regression_output vs. time_column/
item_id_column/max_context_length) and disjoint
validation (single natural key vs. composite (item_id, time) identity), so
they’re split into RelFMTabularPredictor and RelFMTabularForecaster — two
subclasses of a shared _RelFMTabularBase — rather than one class branching on
task_type throughout. RelFMTabular(...) is a thin factory that dispatches
to whichever one matches task_type, so existing call sites don’t need to
know about the split.
Submodules: base (shared plumbing), predictor/forecaster (the two
concrete estimators), factory (the RelFMTabular(...) dispatcher).
Functions
Functions exposed by this module.
relationalai.semantics.reasoners.predictive.estimators.relfm_tabular.factory.Classes
Classes exposed by this module.
RelFMTabular. Re-exported from relationalai.semantics.reasoners.predictive.estimators.relfm_tabular.forecaster.RelFMTabular. Re-exported from relationalai.semantics.reasoners.predictive.estimators.relfm_tabular.predictor.