We will use `Boto3` and Python SDK and follow the same method of generating the parameters for datasources that we used in *Chapter 7, Command Line and SDK*, to do the **Monte Carlo** validation: we will generate features corresponding to power 2 of *x* to power `P` of `x` and run `N` Monte Carlo cross-validation. The pseudo-code is as follows:

for each power from 2 to P: write sql that extracts power 1 to P from the nonlinear table do N times Create training and evaluation datasource Create model Evaluate model Get evaluation result Delete datasource and model Average results

In this exercise, we will go from *2 to 5* powers of *x* and do 5 trials for each model. The Python code to create a datasource from Redshift using ...