I have a list like this:
l=[(1,2),(3,4)]
I want to convert it to a numpy array,and keep array item type as tuple:
array([(1,2),(3,4)])
but numpy.array(l) will give:
array([[1,2],[3,4)]])
and item type has been changed from tuple to numpy.ndarray,then I specified item types
numpy.array(l,numpy.dtype('float,float'))
this gives:
array([(1,2),(3,4)])
but item type isn't tuple but numpy.void,so question is:
how to convert it to a numpy.array of tuple,not of numpy.void?
You can have an array of object dtype, letting each element of the array being a tuple, like so -
out = np.empty(len(l), dtype=object)
out[:] = l
Sample run -
In [163]: l = [(1,2),(3,4)]
In [164]: out = np.empty(len(l), dtype=object)
In [165]: out[:] = l
In [172]: out
Out[172]: array([(1, 2), (3, 4)], dtype=object)
In [173]: out[0]
Out[173]: (1, 2)
In [174]: type(out[0])
Out[174]: tuple
np.array([(1,2),(3,4)], dtype=object) doesn't?
np.array([ [1,2], [2,3] ]) so you want np.array to be able to recursively unpack lists. Unfortunately this is not desired in this case. With the other method you avoid the np.array initialization and this way numpy does not recurse into the hierarchyFor some reason you can't simply do this if you're looking for a single line of code (even though Divakar's answer ultimately leaves you with dtype=object):
np.array([(1,2),(3,4)], dtype=object)
Instead you have to do this:
np.array([(1,2),(3,4)], dtype="f,f")
"f,f" signals to the array that it's receiving tuples of two floats (or you could use "i,i" for integers). If you wanted, you could cast back to an object by adding .astype(object) to the end of the line above).
"f,f" works but object does not: I had a numpy array of tuples and wanted to check if each tuple existed in another array of tuples with np.isin(). Using object array did not work but using "f,f"array did work.>>> np.fromiter(l, object)
array([(1, 2), (3, 4)], dtype=object)
np.fromiter supports object dtype since NumPy 1.23.
https://numpy.org/doc/stable/reference/generated/numpy.fromiter.html
Just discovered that pandas has a way to take care of this.
You can use their MultiIndex class to create an array of tuples
since all pandas Indexes are wrapped 1-D numpy arrays.
It's as easy as calling the Index constructor on a list of tuples.
>>> import pandas as pd
>>> tups = [(1, 2), (1, 3)]
>>> tup_array = pd.Index(tups).values
>>> print(type(tup_array), tup_array)
<class 'numpy.ndarray'> [(1, 2) (1, 3)]