predict numpy array datatype after array calculations

Question

How can I predict the datatype of the resulting array after combining two arrays of certain datatypes, without actually performing the calculation? E.g.

import numpy as np
arr1 = np.array([0.5, 0.1, 1.3, 1.5]).astype(np.float32)
arr2 = np.array([1, 2, 3, 4, 5]).astype(np.uint8)
arr3 = np.array([-10, 2, 3, 4, 120]).astype(np.int8)

output_dtype1 = np.multiply(arr1, arr2).dtype  # np.float32

output_dtype2 = np.multiply(arr2, arr3).dtype  # np.int16

Would it be possible to write a function that returns output_dtype1 and output_dtype2, using arr1.dtype, arr2.dtype, arr3.dtype and the array operator (multiply, add, subtract etc) as arguments ?

If the arrays aren't too large the actual calculation is probably fastest. ufunc do have signature information, but using that with python level calls is probably slower. — hpaulj
– hpaulj, Commented Mar 17, 2021 at 18:18
Study numpy.org/doc/stable/reference/ufuncs.html#casting-rules — hpaulj
– hpaulj, Commented Mar 17, 2021 at 20:00
My arrays are not necessarily large. My use-case is writing geospatial rasters to file using rasterio, for which the output profile has to be compiled beforehand. — Hans Roelofsen
– Hans Roelofsen, Commented Mar 18, 2021 at 8:35
For future reference, I've also found the rasterio dtype module, especially the get_minimum_dtype function. — Hans Roelofsen
– Hans Roelofsen, Commented Mar 18, 2021 at 10:03

hpaulj · Accepted Answer · 2021-03-17 20:05:35Z

1

Based on reading https://numpy.org/doc/stable/reference/ufuncs.html#casting-rules

I tried:

In [196]: np.result_type(np.float32, np.uint8)
Out[196]: dtype('float32')
In [197]: np.result_type(np.int8, np.uint8)
Out[197]: dtype('int16')

answered Mar 17, 2021 at 20:05

hpaulj

233k14 gold badges260 silver badges392 bronze badges

Sign up to request clarification or add additional context in comments.

Collectives™ on Stack Overflow

predict numpy array datatype after array calculations

1 Answer 1

Comments

Your Answer

Hot Network Questions

Collectives™ on Stack Overflow

1 Answer 1

Comments

Your Answer

Sign up or log in

Post as a guest

Related