Maintain numpy array data type when reloading data

Question

Say you create a dataframe like this, where one column contains a numpy array:

data = {'col_1': [np.array([1,2,3])], 'col_2': 3}

x = pd.DataFrame.from_dict(data)

Then you write it to csv and (in a different file) import it as a dataframe:

filename = 'test.csv'

x.to_csv(filename)

x2 = pd.read_csv('test.csv')

Now when you print the type of the numpy array:

print(type(x2.iloc[0,2]))

it returns <class 'str'> instead of a numpy array. Is there a way to avoid this and keep the numpy array type, for instance by using Pickle instead of writing to a csv? Or are datatypes always lost when you write a dataframe to a file, then reload it?