Default handling of empty strings about nvstrings

[davidwendt]

This is great feedback. So there is a little trick here using device memory to hold the results from contains() where the None value is turned identified as -1 as so can be ignored/counted:

>>> import nvstrings
>>> import numpy as np
>>> from numba import cuda
>>>
>>> a = nvstrings.to_device(["test #woot","strings are awesome","#hashme",None,""])
>>> a.count('\#')
[1, 0, 1, None, 0]

>>> arr = np.arange(a.size(),dtype=np.int32)
>>> d_arr = cuda.to_device(arr)
>>> a.count('\#',d_arr.device_ctypes_pointer.value)

>>> d_arr.copy_to_host()
array([ 1,  0,  1, -1,  0], dtype=int32)

>>> c = d_arr.copy_to_host().tolist()
>>> c.count(0)
2
>>> c.count(-1)
1

The -1 is a place holder identifying None strings for contains() in this case.

[davidwendt]

In the next release of nvstrings, we are adding a couple more parameters to the from_csv method including a set of flags. This will include a flag to treat nulls as empty strings. Also, the next release should include some documentation on from_csv

[davidwendt]

Parameter added in new 0.2.0 release.