write1 seems to give wrong shape info

Question

write1 seems to give wrong shape info

nilsbecker opened this issue 7 years ago · 10 comments

i just wrote a fortran-ordered bigarray using write1. the file starts with

�NUMPY��F�{'descr': '<f8', 'fortran_order': True, 'shape': (184), }

note that the shape is not a python tuple (this would have to be (184,)).
numpy refuses to load it, probably for this reason. indeed after i edit the file manually i can import it.

(maybe add a test case for write/read in all dimensions?)

Answer 1 · 2017-08-24T12:54:46.000Z

actually the file still seems to be broken: after editing the shape, the file is loaded but contains garbled values ?!

Answer 2 · 2017-08-24T12:58:41.000Z

the same happens if i first convert the array to a c-layout Bigarray.Array1.t

Answer 3 · 2017-08-24T13:02:15.000Z

os x, python 2.7.11, numpy 1.11.2

Answer 4 · 2017-08-24T13:06:16.000Z

if i wrap the vector in an extra dimension and write it with write2, the import in numpy works as expected. conclusion: write1 is broken, write2 ok.

Answer 5 · 2017-08-24T13:44:50.000Z

Could you attach the problematic npy file ? Also are you using the current github tip or the opam package ?

Answer 6 · 2017-08-24T15:05:10.000Z

i was using the opam package. the file generated with write1 is here:
tvals.npy.zip

Answer 7 · 2017-08-24T15:06:02.000Z

the input bigarrays were actually vectors from lacaml - but that should not matter i think..

Answer 8 · 2017-08-24T15:11:36.000Z

I just tweaked the github tip to fix this and added some tests for the single dimension case, would you mind giving it a try ?

Answer 9 · 2017-08-24T15:21:13.000Z

it's looking good now! i opam pinned and redid my example, and the shape is now reported as (184,) in the file. numpy successfully loads it. thanks!

Answer 10 · 2017-08-24T15:28:24.000Z

Perfect, thanks for reporting the issue!