Thank you for your feedback, and good question.

The Patsy syntax-ed regression expression is used only to carve out the y and X matrices. It is just telling Patsy that BB_COUNT is the dependent variable and thethe variables on the RHS, namely, DAY, DAY_OF_WEEK, MONTH, HIGH_T, LOW_T, PRECIP are the covariates. You may notice that after carving out the y and X matrices, we do not pass the regression expression into the GLM model. Instead, we pass the matrices, and we tell the GLM model to use the Poisson link function.

I should have explained this out in the article. Thanks for pointing it out.