I also found that np.isinf gives the same error

Code:

import numpy as np

list = np.array([1, np.nan, np.inf], dtype=object)

if np.any(np.isinf(list)):
    print("Foo")

Error:

Traceback (most recent call last):
  File "...\test.py", line 5, in <module>
    if np.any(np.isinf(list)):
              ~~~~~~~~^^^^^^
TypeError: ufunc 'isinf' not supported for the input types, and the inputs could not be safely coerced to any supported types according to the casting rule ''safe''

Hi @Tontonio3, can I work on this?

@Mohit-Kundu Yeah, I'm not currently working on it. But we usually don't assign issues

Before starting to implement, it would be good to think through what should happen. To me, it's not quite obvious what the expectations are. E.g., what should np.isnan(np.array([np.nan, "nan", None, {}], dtype=object)) return? I guess in principle one could pass through math.isnan, which would error. Is that desired? Or should it be the equivalent of np.isnan(array.astype(float))? Maybe refusing the temptation to guess is in fact the right thing to do?

Yeah, there are some old issues on this, which would be good to at least check briefly first, since this is a duplicate.

I don't really like special casing floats and math.isnan converts to float(), although that should typically work for numeric NaN-like values...
My preference would be -- if anything -- to argue clearly that bool(value != value) is a reasonable definition of "NaN-like".

To me, it's not quite obvious what the expectations are. E.g., what should np.isnan(np.array([np.nan, "nan", None, {}], dtype=object)) return?

You're right, considering that np.nan is a "mask" and the representation in float of things that are not numbers. So the question I ask is "nan" "NaN-like"? Is False "NaN-like"?

I do believe that it should have the same behaviour for a dtype=object array.