Pulse Documentation

name: op-feat-log
description: Per-row log1p(x) of a numeric field; emits one f64 column. Standard skew-tamer.
kind: operator
category: FEAT
operator: FEAT_LOG
type: reference
applies_to: process, compose, predict
examples_tags: [feature-engineering, distribution-shape, pre-filter, streaming-friendly]

Feature operators emit derived columns; no Response.Components.

Use when

Adds a column holding the natural log of 1 + the value, which pulls in a long tail of large values so a skewed field is easier to model.

Questions it answers:

• Can I tame the long tail of customer spend before fitting a model on it?
• How does income look on a log-like scale, where for large values doubling counts about the same everywhere?

Params

None. Field (required, numeric) — params block is unused.

Inputs

Output

One f64 column written to Label (default LOG_<field>). Formula log1p(x) = ln(1 + x) — the +1 shift keeps x=0 -> 0 instead of -inf.

Reading the output

• value: The natural log of 1 + the value: ln(1 + x). 0 reads 0, e - 1 (about 1.72) reads 1, 9 reads about 2.3 and 99 about 4.6; each step of about 0.69 is a doubling of 1 + x, so equal steps are equal ratios of 1 + x, not equal amounts.
  • Sign: positive means the original value is above 0; negative means the original value is between -1 and 0.
  • Caveat: A value of -1 or less has no log and reads missing, with no error raised: count the missing rows before trusting a model built on the column.
  • Caveat: A missing input gives a missing output.

Gotchas

• Inputs with x <= -1 (where 1 + x <= 0) produce a null, not an error — the rest of the cohort keeps going.
• Null inputs propagate to null outputs.
• Streamable per-row — pairs cleanly with online aggregators downstream.
• Pre-filter slot: a range filter on the log column references LOG_<field> (default label).

See

• pulse_examples_search tags=[feature-engineering]
• Skills: feature-engineering, op-feat-sqrt, op-feat-poly