In the notebook of Gaussian classifiers there is a loop in which PCA is performed. The loop was meant to speed up the process of writing code but by doing so we applied PCA with m=8, so the results on the pdf refer to this case, not to the case without PCA.
When applying some pre-processing techniques on the evaluation set we need to consider information taken from the training set (like the mean, the variance, etc., for example for the Z-normalization). Instead, we computed again mean and other statistics ON THE EVALUATION SET, which led us to overoptimistic results.