The numpy manual mentions use case for numpy.save
Annie Analyst has been using large nested record arrays to represent her statistical data.
Is it possible to have nested records array without dtype=object? If so, how?
-
Does stackoverflow.com/questions/19201868/… answer your question?Eric– Eric2017-07-14 15:24:03 +00:00Commented Jul 14, 2017 at 15:24
Add a comment
|
2 Answers
Yes, like so:
engine_dt = np.dtype([('volume', float), ('cylinders', int)])
car_dt = np.dtype([('color', int, 3), ('engine', engine_dt)]) # nest the dtypes
cars = np.rec.array([
([255, 0, 0], (1.5, 8)),
([255, 0, 255], (5, 24)),
], dtype=car_dt)
print(cars.engine.cylinders)
# array([ 8, 24])
The np.dtype function isn't strictly necessary here, but it's usually a good idea, and gives a small speed boost over letting array call it every time.
Note that rec.array is only necessary here to use the .engine notation. If you used a plain np.array, then you'd use cars['engine']['cylinders']
4 Comments
hpaulj
The data for
cars has to be properly nested as you (right mix of [] and ()). That can be a source of errors. But when loading from a csv genfromtxt can use a 'flat' list of columns, as long as there's enough data.hpaulj
Just checked:
genfromtxt first creates an array with a flattened version of the dtype, and then returns a view with the nested dtype.
lumbric
Can this be done with lists of different lengths? E.g. how would it look like if I want to add a field
photo to the car_dt which is a 2-dimensional array of unknown size and type int? (let's use a black/white image for simplicity)
Eric
No, numpy has no mechanism for subarrays of "unknown size". You'll have to use an
object field to store that arrayYou can construct nested arrays the same way you construct nested lists:
nested_list = [['a',1],['b',2],['c',3]]
import numpy as np
nested_array = np.array(nested_list)
Comments
