And now we come to the final part—in order to compare models, we would like to cross-validate the model. We've already set aside a portion of the data. Now, we will have to test the model on the data that was set aside, and compute a score.

The score we'll be using is a Root Mean Square Error. It's used because it's simple and straightforward to understand:

  // VERY simple cross validation
  var MSE float64
  for i, row := range testingSet {
    pred, err := r.Predict(row)
    mHandleErr(err)
    correct := testingYs[i]
    eStar := correct - pred
    e2 := eStar * eStar
    MSE += e2
  }
  MSE /= float64(len(testingSet))
  fmt.Printf("RMSE: %v
", math.Sqrt(MSE))

With this, now we're really ready to run the regression analysis.