I wanted to do SpectralClustering with affinity given by nearest-neighbors but use manhattan metric for the nearest-neighbor task. (i.e. the argument metric in KNeighborsClassifier). But it's not clear from the documentation how to do this. It is possible to pass n_neighbors to SpectralClustering but not the metric. Then there is also the kernel_params argument but the documentation says "Parameters (keyword arguments) and values for kernel passed as callable object. Ignored by other kernels." So I would think passing kernel_params={"metric":"manhattan"} has no effect. If it does, the documentation should be corrected. If it does not, I think it should be supported in some way.