Using Lasso for non-linear regression (Python)

Question

I have a set of independent data points X, and a set of dependent points Y, and I would like to find a model of the form:

(a₀+a₁*x₁+a₂*x₂+...+a_mx_m)(a_m+1*x_m+1+a_m+2*x_m+2)

I know I can use scipy's curve_fit, but to avoid overfitting, I want to use Lasso for the linear part (i.e. the part in the first set of parenthesis).

Is there a simple way of doing that in Python?

Answer 1

You can fit a lasso regressor to the whole lot, multiplying out your brackets giving you 2m+2 coefficients. Then by performing a change of variables you can make this a linear regression problem again.

See this link for more details: http://scikit-learn.org/stable/modules/linear_model.html#polynomial-regression