name: op-agg-mae
description: Mean absolute error of a prediction against an actual — mean(|y - prediction|).
kind: operator
category: AGG
operator: AGG_MAE
type: reference
applies_to: process, compose, predict
examples_tags: [regression, compose]
Use when
Average size of a prediction’s miss in the actual’s own units, comparing an actual column with a prediction column row by row.
Questions it answers:
- On average, how many units off is the sales model on the held-out weeks?
- Is the forecast miss larger for some stores than for others?
Use something else:
AGG_RMSEwhen large misses should count for more than small ones.
Params
prediction_field — required; the predicted column (typically an ATTR_REG_PREDICT score column). Missing, or a non-numeric column on either side (categorical, set, date, datetime, decimal128), is PROCESSING_CONFIG; an unknown name SERVICE_VALIDATION — predict and runtime alike.
Weight-aware: Σw·|y − ŷ| / Σw. A decimal128 side under a weight → PULSE_WEIGHT_UNSUPPORTED.
Inputs
Field is the actual: u4–u64, f32/f64, packed_bool. No train/holdout param — split with a filter or GROUP_RANGE range grouper.
Output
Scalar float64 in the actual’s units. Rows with a null on either side are skipped; none left → NaN (null).
Reading the output
value: The average size of the miss between prediction and actual, ignoring sign, in the actual’s own units. 0 means every prediction is exact.- Caveat: It says how big the misses are, not which way: a model that is always too high and one that is always too low can share a value.
- Caveat: It has the actual’s units, so compare it only between models scored on the same rows and target.
Components
Floor {n, n_null} (on field only) (weighted adds sum_weights, n_eff, n_weight_invalid) plus loss_sum (Σw·|y − ŷ|), weight_sum, n_scored, n_null_prediction, mae. Mergeable; streams.
Gotchas
- Sign-blind: a model always too high and one always too low can share a value.
See
pulse_examples_search tags=[regression, compose]- Skills:
op-agg-rmse,op-agg-mape,op-attr-reg-predict