Shape · Statistics

Linear Regression

Fits a straight-line model of one column from one or more others. Outputs the coefficients, or your rows with predictions and residuals.

Also known as: Least squares, trend line

How it works

Pick the Target column and one or more Feature columns. Fit intercept is on by default. Rows with a NULL in the target or any feature are left out of the fit, and the node stops with an error when fewer rows remain than parameters. Output Coefficients gives a table with term and coefficient: an intercept row when fitted, one row per feature and an r_squared row. Predictions returns the rows used in the fit, so rows with a NULL in the target or any feature are dropped, with two appended columns, named predicted_target and residual_target unless you set Prediction column and Residual column. This node runs on the desktop with the rows in memory.

Inputs and outputs

Settings

{
  targetColumn: string
  featureColumns: string[]
  fitIntercept: boolean
  output: "coefficients"|"predictions"
  predictionColumn?: string
  residualColumn?: string
}

Example

Explain weekly sales from price and footfall: target sales, features price and footfall, output Coefficients.

More in Statistics

• Describe — Summarizes each column in one row: count, blanks, mean, standard deviation, min, percentiles and max. A quick first look at a new dataset.
• Value Counts — Counts how often each value, or combination of values, appears: one row per distinct value with its count or share.
• Column Profile — Counts the distinct values in each chosen column, as long rows of column, value and count. Useful for spotting typos or dominant categories.
• Correlation Matrix — Shows how strongly each pair of number columns moves together, as a grid of values from −1 to 1.
• Covariance Matrix — Works out the covariance of every pair of number columns, as a grid.
• Standardize — Rescales number columns to a common scale: z-score or 0 to 1. Use it before comparing measures in different units.
• Outlier Flag — Adds a true/false column marking rows with unusually high or low values. Rows are kept, so you can filter or review them afterwards.