Assessing performance with cross-validation

2. for each part in turn:

   • use the remaining $K − 1$ parts as training data to define the interpolation

   • use that part as testing data to predict

   • compute the RMSE by comparing the testing and predicted data in each of the $K$ parts

     $RMSE = \sqrt{\frac{\sum_{i = 1}^{n_{test}}(y_{i, test} - \hat{y}_{i, test})^2}{n_{test}}}$

3. average the RMSE values obtained in each of the $K$ parts