Assuming I get a good model after the training and testing phases, then I am using it to predict on some new input data. Just wondering if I should use the mean and std which I get from the training set to standardize the input data? Or should I just do the independent standardization based on the new input data themselves?

Many thanks in advance!