Pulse Documentation

name: op-agg-vec-mean
description: Per-member mean of a vector — one average per member, in member order (a group's average profile).
kind: operator
category: AGG
operator: AGG_VEC_MEAN
type: reference
applies_to: process, compose, predict
examples_tags: [multivariate, streaming-friendly]

Use when

Average of each item of a battery or mix, as one list per group: the group’s average profile across the items.

Questions it answers:

• What is each segment’s average rating on each of the eight brand attributes?
• What does the typical store’s channel mix look like in each region?

Use something else:

• ATTR_SCALE_SCORE when you want one score per row from the items.
• AGG_AVERAGE when you want the average of a single field.

Params

• labels — inline fields only: one display label per member (default the member names).
• expand — true: one scalar column per member, <label>[<element label>], instead of the array (grouped: per row); the label is absent. Predict lists them (aggregations[i].columns). Element labels must be distinct.

Slot: exactly one of vector (a vectors entry, its labels kept) / fields; no field. Default label AGG_VEC_MEAN_<vector>. Weight-aware: Σw·x / Σw per member.

Inputs

Members: u4…u64, f32/f64; source columns only.

Output

Rich []float64, one mean per member; per-element missing data (a null member skips only that element). An element no row answered is NaN (null).

Components

Floor {n, n_null} — a row is present when ANY member answered (weighted adds sum_weights, n_eff, n_weight_invalid). Plus labels, element_n (raw per member), element_sum, element_sum_weights (weighted). Mergeable; streams; bit-identical for any worker count or shard split.

Gotchas

• Array-valued: refused (PROCESSING_CONFIG, reason array_valued) as a crosstab cell, a Sort key, a window or post-test input, an overlay host (request, Compose, whole-chain) and a chain stage feeding another.
• Expanded columns are presentation only — refused in the same places. A return path cannot name a bracketed column: keep them with data[*].<label>*.

See

• pulse_examples_search tags=[multivariate, streaming-friendly]
• Skills: op-agg-vec-sum, op-attr-scale-score, multivariate-design