I have a list of a certain Python object (let's call it MyClass) that can be interpreted as a multi-dimensional Numpy array. However, I'd like to convert that list to a numpy array of MyClass, and not try to convert MyClass to an inner Numpy array. Just for the sake of the question, you can use a simple list instead of MyClass:
a = [1, 2, 3]
b = [4, 5, 6]
data = [a, b]
You can achieve what I want with:
import numpy
arr = np.empty(len(data), dtype=object)
for i,v in enumerate(data):
arr[i] = v
assert arr.shape == (len(data),) # Works
But I'm surprised as to why this doesn't work:
arr = np.array(data, dtype=object)
print(arr.shape) # prints (2 ,3), and I want (2,)
Is there a way I can limit Numpy to not delve into the inner lists and instantiate them as arrays? Why isn't dtype=object argument implies that? I really hope to avoid that i,v enumerating loop.
Another option is to use fromiter:
In [13]: data
Out[13]: [[1, 2, 3], [4, 5, 6]]
In [14]: arr = np.fromiter(data, dtype=object, count=len(data))
In [15]: arr
Out[15]: array([list([1, 2, 3]), list([4, 5, 6])], dtype=object)
In [16]: arr.shape
Out[16]: (2,)