Dataclasses, immutability(?), and ChatGPT
Thomas Passin
list1 at tompassin.net
Wed Apr 12 08:26:47 EDT 2023
On 4/12/2023 12:58 AM, dn via Python-list wrote:
> Are dataclasses (or instances thereof) mutable or immutable?
> - and in what sense?
>
>
> Have been experimenting with ChatGPT. In particular: its possibilities
> as a trainer, good ideas for methods of introducing new topics, its
> capability for drawing-up demonstrations or examples, its interpretation
> of assessment questions (ie evaluating my own work, eg do they test
> desired specifics, is it possible/likely to misinterpret the question, ...)
>
>
> Conversed with ChatGPT about the differences between (standard)
> custom-classes and dataclasses, and when one might prefer one over the
> other.
>
> We(?) started discussing how dataclasses might be introduced to trainees
> currently only familiar with Python's built-ins: "...how they are
> superior to using built-in data-constructs..."
>
> Amongst the four benefits ChatGPT listed was:
> «
> Immutable instances: By default, instances of dataclasses are immutable,
> which means that once created, their attributes cannot be modified. This
> can help to prevent unintended modifications to the data.
> »
>
> Huh? If we'd been discussing namedtuples over (say) dictionaries, I'd
> perhaps have accepted the reply.
>
>
> The next step was to write some code to double-check my understanding
> (reproduced below). No change there.
> (given that I'm perfect...cough, cough, splutter)
>
>
> Thus, back to ChatGPT to seek clarification:
> «
> You said "Immutable instances: By default, instances of dataclasses are
> immutable, which means that once created, their attributes cannot be
> modified. This can help to prevent unintended modifications to the
> data." The attribute-values within a data-class can be changed. In what
> way are Python dataclasses immutable?
> »
>
> ChatGPT did indeed clarify that:
> «
> You're correct that the attribute values within a dataclass can be
> changed, even though the instance itself is considered immutable.
> »
>
> That's better. The machine is as smart as I am - or I, as smart as it -
> or something(?)
>
>
> However, the supporting-explanation then appears to argue against its
> own clarification (and the assertion in the question):
> «
> So to clarify, when we say that instances of dataclasses are immutable,
> we mean that the attributes of an instance cannot be changed after it is
> created.
>
> In other words, if you create an instance of a dataclass, its attributes
> cannot be modified by any external code. If you want to change the value
> of an attribute, you need to create a new instance of the dataclass with
> the updated attribute value.
>
> ... [code excised - see below]
>
> This is because the instance is considered immutable, and its attributes
> cannot be changed directly. Instead, we need to create a new instance of
> the dataclass with the updated attribute value, like this...
>
> Now, new_person is a new instance of the Person dataclass with the
> updated attribute values, while the original person instance remains
> unchanged.
>
> So to summarize, while the attribute values within a dataclass can be
> changed, instances of dataclasses are considered immutable because their
> attributes cannot be modified directly once they are created. Any
> updates to the attributes require the creation of a new instance of the
> dataclass.
> »
>
> [Note]
> In and amongst the above explanatory-paragraphs it gives example
> code-snippets. Those are condensed into:
>
> >>> from dataclasses import dataclass
> >>>
> >>> @dataclass
> ... class Person:
> ... name: str
> ... age: int
> ...
> >>> person = Person(name='Alice', age=20)
> >>>
> >>> person.name = 'Bob' # raises AttributeError: can't set attribute
> >>> person.age = 21 # raises AttributeError: can't set attribute
>
> - note the erroneous comments about exceptions being raised
> (they weren't/aren't)!
>
>
> Here's a home-grown version illustrating all the points made during the
> investigation, by way of proof/disproof:
>
> """ PythonTraining:dataclass.py
> Prove/disprove claim made by Chat-GPT.
> """
>
> __author__ = "dn, IT&T Consultant"
> __python__ = "3.11"
> __created__ = "PyCharm, 12 Apr 2023"
> __copyright__ = "Copyright © 2023~"
> __license__ = "MIT"
>
> # PSL
> from dataclasses import dataclass
>
>
> @dataclass
> class Coordinates():
> """Sample dataclass. """
> x:int
> y:int
>
>
> if __name__ == "__main__":
> print( "\nCommencing execution\n" )
>
> coordinates = Coordinates( 1, 2, )
> print( coordinates, id( coordinates ), )
> coordinates.x = 3
> print( coordinates, id( coordinates ), )
> coordinates.z = 4
> print( coordinates, id( coordinates ), )
> print( coordinates.x, coordinates.y, coordinates.z, )
>
>
> ### output:
> Commencing execution
>
> Coordinates(x=1, y=2) 140436963150928
> Coordinates(x=3, y=2) 140436963150928
> Coordinates(x=3, y=2) 140436963150928
> 3 2 4
>
> Terminating
> ###
>
> Not only are a dataclass instance's attribute-values mutable, but
> further attributes can be dynamically-added to the object-instance!
>
> Yes, if the code included:
>
> coordinates = Coordinates( 5, 6, )
>
> the new "coordinates" identifier would point to a different id()
> 'address', ie a fresh immutable-instance.
>
>
> The 'book of words' (https://docs.python.org/3/library/dataclasses.html)
> does mention immutability (wrt to dataclasses) in that it is possible to
> add a __hash__() method (any object defined with is (technically)
> immutable). However, apart from the default_factory argument, there
> doesn't appear to be other discussion of [im]mutability.
>
>
> Anything I've 'missed'?
> - or a salutary tale of not depending upon ChatGPT etc?
People need to remember that ChatGPT-like systems put words together the
way that many humans usually do. So what they emit usually sounds
smooth and human-like. If it's code they emit, it will tend to seem
plausible because lines of code are basically sentences, and learning
how to construct plausible sentences is what these systems are built to
do. That's **plausible**, not "logical" or "correct".
The vast size of these systems means that they can include a larger
context in figuring out what words to place next compared with earlier,
smaller systems.
But consider: what if you wrote code as a stream-of-consciousness
process? That code might seem plausible, but why would you have any
confidence in it? Or to put it another way, what if most of ChatGPT's
exposure to code came from StackOverflow archives?
On top of that, ChapGPT-like systems do not know your requirements nor
the reasons behind your requests. They only know that when other people
put words and phrases together like you did, they tended to make
responses that sound like what the chatbot emits next. It's basically
cargo-culting its responses.
Apparently researchers have been learning that the more parameters that
a system like this has, the more likely it is to learn how to emit
responses that the questioner likes. Essentially, it could become the
ultimate yes-man!
So there is some probability that the system will tell you interesting
or useful things, some probability that it will try to tell you what it
thinks you want hear, some probability that it will tell you incorrect
things that other people have repeated, and some probability that it
will perseverate - simply make things up.
If I were going to write a novel about an alternate history, I think
that a chatGPT-like system would be a fantastic writing assistant.
Code? Not so much.
More information about the Python-list
mailing list