I have to convert a code written using cuml (RAPIDS) into sklearn.
I found out that in cuml.truncatedSVD the parameter n_components which is the output dimensions (number of singular values) can equal to the number of inputs/features in cuml, but not in sklearn.decomposition.truncatedSVD which requires a value strictly inferior to the input dimensions.
The cuml code I'm converting takes two features as inputs and computes two singular values, which is impossible with sklearn.
Is there a way workaround or a way to make it work with sklearn?
The solution is to use the SVD methods in scipy (faster in my case) or numpy. You can find more information in this discussion.