IntegerScalar

[ozgurkalan]

Hi,

when count(), we get an IntegerScalar, how to cast it to python int type?

f"{t.count():,}"
>>>TypeError: unsupported format string passed to IntegerScalar.__format_

[cpcloud]

@ozgurkalan Thanks for the question!

You'll need to call to_pandas() on the expression to trigger computation:

In [1]: from ibis.interactive import *

In [2]: t = ex.penguins.fetch(table_name="penguins")

In [3]: f"{t.count().to_pandas():,}"
Out[3]: '344'

[cpcloud]

is there any way for Ibis to make this cleaner?

Not without triggering computation AFAICT.

[cpcloud]

I honestly can't tell what would be more surprising:

• Triggering computation, one consequence being that formatting can take a very long time.
• Not triggering computation and returning a strange error message like the one from the OP.
• Not triggering computation and printing the repr of the expression.

[cpcloud]

I could see the argument for implementing __format__ like this on Expr:

    def __format__(self, spec: str) -> str:
        return format(self.to_pandas(), spec)

If a user is operating in interactive mode, they'll see the output of the expression as something that looks like a value, especially if it's a scalar result. Maybe that's an argument against repring a lazy object that way, but given that's the existing behavior, implementing __format__ in a way that makes it consistent with the interactive repr might make sense.

On the other hand, anywhere to_pandas() is invoked the reader of the code knows without a doubt that computation is going to happen and so expectations are set, and something taking a long time won't be surprising.

[lostmygithubaccount]

to explain my personal issue with to_pandas() here, I just don't like it conceptually -- in my case I'm converting scalars (ints and floats) to numbers for a metrics dashboard, or sometimes strings. to_pandas() makes me think dataframe, but as a user I really just want to pull out the one thing computed and put it in a string

the __format__ option seems decent to me...or keeping the call explicit and having a to_scalar() output for this case? or do nothing and document is has to be to_pandas()? I don't have a strong opinion on this

[NickCrews]

@cpcloud Strongly feel like the user must call some explicit method such as .to_pandas().

Neither returning the basic error, nor using repr, point the user towards what they probably want. I would do this:

def __format__(self):
    return SomeError(f"""cannot format {self.__class__.__name__}. Execute first: replace `f'result={v}'` with `f'result={v.to_pandas()}'`""")