Polyfit (polyfit)

The polyfit function uses polynomial regression to predict a smooth, non-linear curve through a bi-variate scatter plot. The polyfit function takes three parameters:

  1. The numeric field containing the independent (x) variable

  2. The numeric field containing the dependent (y) variable

  3. The integer degree of the polynomial

    The degree determines the number of curves to the fitted curve. One degree polynomial performs linear regression. Typically the degree is an integer from 1 to 5.

Sample syntax

select polyfit(petal_length_d, petal_width_d, 3) as prediction,
from iris
       limit 150

Result set

The result set contains a random sample of records that match the WHERE clause. If no WHERE clause is included the random sample will be taken from the entire result set. The size of the result set can be controlled by the LIMIT clause. The default size, if no limit is applied, is 25000.

The polyfit function returns the predicted value for each record. There are three additional fields that can be selected when the polyfit function is used:

  • residual: the residual value for each sample

    The residual value is the samples dependent (y) value minus the predicted value. The residual represents the error of the regression prediction for each sample.

  • the independent variable for each sample

  • the dependent variable for each sample

Sample result set in Apache Zeppelin

Sample result set


There are a number of visualizations that can flow from the regression result set.

The first visualization shown is a scatter plot with petal_length_d on the x-axis and petal_width_d on the y-axis. This can be used to visualize the relationship between the two variables in the regression analysis.

Sample visualization

The second visualization shows the petal_length_d variable on the x-axis and the prediction for petal_width_d on y-axis.

Sample visualization

The last visualization plots the predictions on the x-axis and the residual on the y-axis. This residual plot can be used to visualize the error of the regression model across the full range of predictions.

Sample visualization