2

I have a recarray that comes from reading a csv file. I am interested in converting a subset of columns to a continuous float array. I'd like to avoid converting them to list or stacking them one by one. I tried the suggestions in https://stackoverflow.com/a/11792956 and https://stackoverflow.com/a/7842620 but I get

ValueError: new type not compatible with array.

Here is my code:

a = np.recfromcsv(r"myfile.csv")
#a has many columns of type int, float or string. I want to extract those called coeff*
coeffs_columns = [n for n in a.dtype.names if n.startswith('coeff')] 
coeffs_recarray = a[coeffs_columns]
newtype=[(n,'<f8') for n in coeffs_columns]
b = coeffs_recarray.astype(newtype)
#b is:
#array((0.0, 0.0, 0.0, 0.0, 0.0, 0.0, 0.0, 0.0, 0.0, 0.0, 0.0, 0.0, 0.0, 0.0, 0.0),  dtype=[('coefficients00', '<f8'), ('coefficients1', '<f8'), ('coefficients2', '<f8'), ('coefficients3', '<f8'), ('coefficients4', '<f8'), ('coefficients5', '<f8'), ('coefficients6', '<f8'), ('coefficients7', '<f8'), ('coefficients8', '<f8'), ('coefficients9', '<f8'), ('coefficients100', '<f8'), ('coefficients11', '<f8'), ('coefficients12', '<f8'), ('coefficients13', '<f8'), ('coefficients14', '<f8')])
coeffs = b.view('<f8')

The "funny" thing is that if I extract only one column, or if I work with a recarray created as 

x = np.array([(1.0, 2,7.0), (3.0, 4, 9.9)], 
                 dtype=[('x', '<f8'), ('y', '<f8'), ('z', '<f8')])

the conversion works.

Community
  • 1
  • 1
lib
  • 2,918
  • 3
  • 27
  • 53
  • Have you tried [np.ascontiguousarray](http://docs.scipy.org/doc/numpy/reference/generated/numpy.ascontiguousarray.html) on your array of interest? – Bort Apr 14 '15 at 15:45
  • Does `a[coeffs_columns].view('f')` work? – hpaulj Apr 14 '15 at 16:05
  • @hpaulj no, it gives the same result – lib Apr 15 '15 at 06:54
  • @Bort , thanks, that has solved the problem! If you convert it to an answer I will accept it – lib Apr 15 '15 at 06:55
  • another "funny" thing is that coeffs_recarray.flags['C_CONTIGUOUS'] is True , but it works only if I use explicitly np.ascontguousarray – lib Apr 15 '15 at 07:12

1 Answers1

3

Numpy provides numpy.ascontiguousarray.

This function returns a contiguous array in memory (C order) of its input array. This is especially helpful when dealing with non contiguous views on arrays.

If Fortran order is desired, use numpy.asfortranarray.

Bort
  • 2,423
  • 14
  • 22