moygit / exact_cover_np Goto Github PK

Hi,

Thanks for writing this. Here are some corner cases:

import exact_cover_np
import numpy as np

exact_cover_np.get_exact_cover(np.array([[1,0,0],[0,1,1]],dtype=np.int32))
#array([0, 0], dtype=int32)

exact_cover_np.get_exact_cover(np.array([[1,1,1]],dtype=np.int32))
#Segmentation fault (core dumped)

Currently the entire matrix is created in Python and then passed whole to the C code, which then translates it into a sparse matrix. So the example on the front page says S = ... and then calls get_exact_cover(S). If this matrix is large, though, the caller might prefer to construct it piecemeal, say row by row, and pass each row to the C code to be sparsified individually.

What this means in code: currently get_exact_cover (or, more precisely, dlx_get_exact_cover) calls create_sparse_matrix. A more flexible and memory-efficient approach, though less time-efficient, would be to have an API like init_create_sparse_matrix, add_row_to_sparse_matrix, and done_creating_sparse_matrix.

Hi Moy,

Thanks for this code!

I've made a few updates to get it working with Python 3. (They break Python 2 compatibility but I figured that was ok since it's end-of-lifed. Making it work with both shouldn't be too much harder.) You can see what I've done at jwg4#1

I just wondered whether you would care to merge these in. If you're not bothered, I might submit to pypi under a slightly different name? Let me know what you think.