linear_regression

linear_regression fits one ordinary least squares line: one numeric feature, one numeric target.

It is intentionally small. Use it for a reportable single-feature relationship, not as a substitute for a modeling workflow.

Model contract

Field	Required	Notes
inputs	yes	Exactly one table.
feature	yes	Single numeric predictor.
target	yes	Numeric outcome.
testFraction	no	Optional holdout fraction between 0 and 1.
seed	no	Optional integer seed for deterministic splitting.

How to read the result

default includes n, slope, intercept, r2, p-value, a 95% slope confidence interval, and effect size.

testFraction can reserve a deterministic holdout split. Add seed when you want that split to be repeatable.

When not to use it

• Multiple predictors, interactions, robust standard errors, diagnostics, and nonlinear models belong in Python/R.
• LINEAR_REGRESSION_SAMPLE_SMALL appears when n is below 20.
• LINEAR_REGRESSION_HIGH_RESIDUAL_OUTLIERS warns when at least 5% of observations have residuals at or beyond 3 residual standard deviations.

Example

- id: lr
  kind: linear_regression
  inputs: [training]
  feature: x
  target: y
  testFraction: 0.2               # optional
  seed: 42                        # optional

Related

• correlation - lighter-weight association check