Scikit-learn based interface with fit/predict functions added for easier usage. Added labels_ and centroids_ attributes of two classes KShapeClustering CPU and KShapeClustering GPU.
Added n_jobs parameter which allows to adjust number of processes in multiprocessing calculations
Fix issue #14 which used depreciated symeig function. It was changed to torch.linalg.eigh which is equivalent.
Fix issue with nan_to_num_cuda error for complex numbers that were returned by eigh function
Fix issue with multiprocessing pool added in _extract_shape function. The multiprocessing at this point was removed which speeds up computations a lot
Following changes were made to the original code:
labels_
andcentroids_
attributes of two classesKShapeClustering CPU
andKShapeClustering GPU
.n_jobs
parameter which allows to adjust number of processes in multiprocessing calculationssymeig
function. It was changed totorch.linalg.eigh
which is equivalent.nan_to_num_cuda
error for complex numbers that were returned byeigh
function_extract_shape
function. The multiprocessing at this point was removed which speeds up computations a lot