Suppose I have a 2D numpy array of dtype=np.float. Let's say the shape is (n, 4).
I want to convert that array to a new array of a new dtype: new_type = np.dtype([('a', np.float, 1), ('b', np.float, 1), ('c', np.float, 2)]), i.e., type with 3 fields ('a', 'b', 'c') of type float where the first field has 1 element, the second has 1 element and the third has 2 elements.
I can't find a way to convert the 2D array into a 1D array of my new dtype (so each element as 3 fields with overall 4 elements that correspondence to the matrix rows) without using loops.
I've tried using the astype() function but it converts each element to the new type instead of each row as I wish.
For example I have the array:
array([[1., 2., 3., 4.],[5., 6., 7., 8.]])
and I want to get
array([(1., 2., [3., 4.]), (5., 6., [7., 8.])], dtype=[('a', '<f8'), ('b', '<f8'), ('c', '<f8', (2,))]