Before Python 3.7, namedtuple was the easiest way to use class-like objects without having to implement a class for each object with the associated constructor. Since Python 3.7, dataclasses can also be used for this purpose. dataclass allows a class to easily declare that it contains only data. This is available in the dataclasses module of the standard library. The same snippet can be written as follows with dataclasses:

from dataclasses import dataclass

@dataclass
class Basket:
    count
    what

@dataclass
class FruitCount:
    count
    double

obs = Observable.from_([Basket(count=5, what='apple'),
                        Basket(count=3, what='orange')]) 
    .filter(lambda i: i.what == 'apple')  
    .map(lambda i: i.count) 
    .map(lamda i: FruitCount(count=i, double=i*2)) 
...

Compared to namedtuple, using dataclass is the same, but the definition is done as a class definition. If an application requires Python 3.7 at a minimum, then data classes can be used. However, named tuples still require less code to declare, and their immutability is a nice sanity check in functional code.