CLN: enforce deprecation of interpolate with object dtype

[natmokval]

xref #53638

enforced deprecation of interpolate with object dtype

[natmokval]

I enforced deprecation of interpolate with object dtype. @phofl could you please take a look at this PR? I think CI failures are unrelated to my changes.

[mroeschke]

Could you combine these conditionals? np.any and np.all overlap

[natmokval]

thank you for the comment. I combined two if statement into one.

[mroeschke]

Looks like part of this is already handled below (if obj.ndim == 2 and np.all(obj.dtypes == object)).

In instead nearby that check could you add a if obj.ndim == 1 and obj.dtype == object:?

[natmokval]

Thank you. To avoid duplicate checks I changed blocks of if statements as you suggested.

[natmokval]

I think we can now simplify things and remove this check. The reason: we raise

TypeError(
    f"{type(self).__name__} cannot interpolate with object dtype."
)

in case np.any(obj.dtypes == object) below and there is no need to raise Type Error if np.all(obj.dtypes == object). Should I remove raising

if np.all(obj.dtypes == object):
    raise TypeError(
        "Cannot interpolate with all object-dtype columns "
        "in the DataFrame. Try setting at least one "
        "column to a numeric dtype."
    )

?

[mroeschke]

Nice catch! Yeah try that out

[mroeschke]

But the messaging will need to probably change, this could be a Series

[natmokval]

sorry, I don't understand why is it a Series? I thought it's a DataFrame, because we are in the block
if obj.ndim == 2:

I suggest to use the message f"{type(self).__name__} cannot interpolate with object dtype.", the same one we use in the block below

elif np.any(obj.dtypes == object):
    raise TypeError(
        f"{type(self).__name__} cannot interpolate with object dtype."
    )

I think we can just remove the block with the old message:

if obj.ndim == 2:
    if np.all(obj.dtypes == object):
...

and change the old message in tests to the new one "DataFrame cannot interpolate with object dtype"

[mroeschke]

Ah sorry I had misunderstood your initial suggestion.

You can also go a step further and replace All of these checks with

if np.any(obj.dtypes == object):
    raise TypeError(f"{type(self).__name__} cannot interpolate with object dtype.")

Since Series.dtypes and DataFrame.dtypes can be checked at the same time

[natmokval]

sorry, my wording wasn't clear.

My idea was to remove the second part of the error message:

"Cannot interpolate with all object-dtype columns in the DataFrame. Try setting at least one column to a numeric dtype."

"Try setting at least one column to a numeric dtype" won't work from now on because of the enforcing in this PR. Then I thought why not to replace the first part: "Cannot interpolate with all object-dtype columns in the DataFrame." with a new msg: "... cannot interpolate with object dtype."
If we do this we can remove one check block.

I made changes. Could you please take a look at it?

[mroeschke]

Thanks @natmokval

[jbrockmendel]

I'd suggest doing this check inside Block.interpolate

[jbrockmendel]

there's a TODO(3.0) comment related to this inside Block.interpolate

[natmokval]

thank you, I found the comment. In #58083 I moved raising TypeError from NDFrame to Block.

I think maybe it's better to change the error message from "NumpyBlock cannot interpolate with object dtype." to "Can not interpolate with object dtype."?