Search code examples
pythonscikit-learnregressionpolynomialsnon-linear-regression

Run a polynomial regression without combinations of the features

I am running a polynomial regression for order p. To make it simple, we use order p = 2 in this question.

Suppose we have X with two feature x1, x2 and y. And I am trying to run a polynomial regression of

y = ε + α + β1•x1 + β2•x2 + β3•x1^2 + β4•x2^2

I find that the sklearn have a sklearn.preprocessing.PolynomialFeatures. However, if I use order p = 2 and it automatically gives the combination of features. It will result in a regression of:

y = ε + α + β1•x1 + β2•x2 + β3•x1^2 + β4•x2^2 + β5•x1x2

However, I do not want the combination of the features, i.e. things like x1x2. Is there any package that can do the polynomial regress as I wished?

Thanks!

Solution

  • numpy.polynomial.polynomial.polyfit seems to serve your needs.

    For even more specific needs use this statistics tool