|
Hi there! In the course of pre-processing you usually do some kind of normalisation and reduce the dimensionality via PCA. I was wondering if it's better to calculate e.g. a separate PCA of the test-set, or if I should re-use the principal components that I calculated on the training-set. Same goes for other values such as mean, median, stdev or whatever else is used as normalization. Is it common practice to use the values calculated on the training-set also for the test-set, or not? |
