Numpy float sizes

Numpy uses 64-bit floats by default, although supposedly this is implementation dependent. If you are reading this several years from now, then you will have to re-check the validity of this number. In any case, in order to use, say, 32-bit floats instead, use np.float32. You also have a float16 and a float128, as of now.