python structured recarray type conversion behaviour

Question

The behaviour of type conversion is pretty confusing when constructing a structured/recarray:

For example, the following code takes in numerical fields but defines the type as string:

data = [(1.0, 2), (3.0, 4)]
np.array(data, dtype=[('x', str), ('y', int)])
 

Which produces:

array([('', 2), ('', 4)], dtype=[('x', 'S'), ('y', '<i8')])
 

So the values were converted to empty strings which is not what you would expect from:

str(1.0)
 

Which produces the string '1.0'. Can anyone tell me the problem and how to solve this?

Answer 1

You need to specify a string width, e.g. 'a3':

>>> np.array([(1.0, 2),(3.0,4)],dtype=[('x','a3'),('y',int)])
array([('1.0', 2), ('3.0', 4)],
      dtype=[('x', 'S3'), ('y', '<i4')])
 

Just using str effectively means a string field of 0 bytes - which of course is too small to hold the string rendition of the float.