I tried to understand what we should do before PCA: standartization (x-m)/s or normalization (scale into [0, 1] interval). In the sklearn tutorial they use standardization and show that PCA with standardization performs better: