I found this while testing Application Classification. A CSV file is parsed by pandas, then sliced into a numpy array, only using some of the columns. When passing this array into MLIR as a dense tensor, things got weird fast. Forcing the numpy array through np.ascontiguousarray fixed everything, so somehow we are not fully handling all the strides, etc of the array.
I found this while testing Application Classification. A CSV file is parsed by pandas, then sliced into a numpy array, only using some of the columns. When passing this array into MLIR as a dense tensor, things got weird fast. Forcing the numpy array through
np.ascontiguousarrayfixed everything, so somehow we are not fully handling all the strides, etc of the array.