numpy.savetxt resulting a formatting mismatch error in python 3.5

Question

I'm trying to save a numpy matrix (Nx3, float64) into a txt file using numpy.savetxt:

np.savetxt(f, mat, fmt='%.5f', delimiter=' ')

This line worked in python 2.7, but in python 3.5, I'm getting the following error:

TypeError: Mismatch between array dtype ('float64') and format specifier ('%.5f %.5f %.5f')

When I'm stepping into the savetxt code, the print the error (traceback.format_exc()) in the catch block (numpy.lib.npyio, line 1158), the error is completely different:

TypeError: write() argument must be str, not bytes

The line of code resulting the exception is the following:

fh.write(asbytes(format % tuple(row) + newline))

I tried to remove the asbytes, and it seems to fix this error. Is it a bug in numpy?

Answer 1

savetxt opens the file in wb mode, and thus writes everything as bytes.

If instead I open the file with 'w', I get your second error:

In [403]: x=np.ones((3,3),dtype=np.float64)
In [404]: with open('test.txt','w') as f:
    np.savetxt(f,x,fmt='%.5f')
   .....:     
TypeError: must be str, not bytes

But there's no problem with

In [405]: with open('test.txt','wb') as f:
    np.savetxt(f,x,fmt='%.5f')
   .....:     
In [406]: cat test.txt
1.00000 1.00000 1.00000
1.00000 1.00000 1.00000
1.00000 1.00000 1.00000

This is on Py3.4; I don't have numpy installed with my 3.5 Python. But I wouldn't expect a difference.

Does

'%.5f'%1.2342

work on your system? You could also try

'%.5f %.5f %.5f'%tuple(mat[0,:])