the components of LDA

[immortal13]

Thanks for your innovative and interesting work！
Could you please tell me why you choose (n_classes-1) components in LDA？
And can sklearn.SLIC() accept a multichannel image? (for PaviaU data set, there supposed to be 8 channels)

[qichaoliu]

Thanks for the questions, and
Answer to Question 1: The LDA is a supervised method that can project the raw data into a subspace according to the given training samples. Assuming that there are C classes in the training data, then the LDA could only reduce its dimension to [1, C-1], which is determined by the algorithm. To preserve as much as information, choosing the (C-1) components seems to be more reasonable.
Answer to Question 2: Yes, it can!

By the way, the LDA is not important here. You can replace it with other methods such as PCA, MNF, or even remove it if you want.

[immortal13]

Thanks for your prompt reply！It helps a lot. (^▽^)