save_npz sometimes fails

Hi,

save_npz sometimes fails because of this line of code, where params is a list of numpy arrays. When the first dimension of arrays in params is the same but other dimensions may be not, numpy would raise a broadcast error.

Here is a piece of code for reproducing the numpy error.

import numpy as np
t1 = np.arange(8).reshape((2, 2, 2))
t2 = np.arange(12).reshape((2, 2, 3))
t3 = np.arange(16).reshape((2, 2, 4))
np.savez('test.npz', params=[t1, t2, t3])

I use python 3.5 and numpy 1.12. I googled around and it seems to be an implementation problem in numpy, as suggested by this. A possible solution is not to use a list of numpy arrays for the params.

[zsdonghao]

thank you for reporting this.

Could we set all params to list before saving them?

np.savez('test.npz', [list(t1),list(t2),list(t3)])

@zsdonghao the error would still be raised.