Problem with dtype in bcolz.carray constructor.

[Kyrish]

According to documentation, parameter dtype should "Force this dtype for the carray (rather than the array one)".

Let's consider following:

arr = np.arange(20).reshape((4, 5)) # dtype of array is np.int64
ds = bcolz.carray(arr, dtype=np.float32, expectedlen=arr.size)

And I got the TypeError:

bcolz/carray_ext.pyx in bcolz.carray_ext.carray.__cinit__ (bcolz/carray_ext.c:14817)()

bcolz/carray_ext.pyx in bcolz.carray_ext.carray._create_carray (bcolz/carray_ext.c:15599)()

/usr/local/lib/python2.7/site-packages/bcolz/utils.pyc in to_ndarray(array, dtype, arrlen, safe)
    117     # Ensure that we have an ndarray of the correct dtype
    118     if dtype is not None:
--> 119         if type(array) != np.ndarray or array.dtype != dtype.base:
    120             try:
    121                 array = np.array(array, dtype=dtype.base)

TypeError: data type not understood

[FrancescAlted]

Yep. Perhaps forcing a view of the numpy array should be relatively easy to add. Interested in providing a PR? If not easy, perhaps fixing the documentation would be better.

[Kyrish]

I believe that simple

arr = arr.astype(dtype, copy=False)

can do the trick. At least there are no overheads. If so I'll try to provide PR.

[Venkat214]

@Kyrish

np.dtype('float32')

This might works, try.