ping @jnothman

This seems overly lenient and strange behaviour, as in #9342 where @qinhanmin2014 shows that check_array(['a', 'b', 'c'], dtype='numeric') works without error and produces an array of strings!

I think you mean #9835 (#9835 (comment)) ?

Yes, from my perspective, if it is not intended, it seems a bug.

Five now!
And I'm not entirely sure what my intended behavior was, but I agree with your assessment. This should error on strings.

Well, I guess it's one more bit, though ;)

@jnothman I'd be happy to give this a go with some limited guidance if no one else is working on it already. Looks like the behavior you noted comes from this line, where we're checking the array against the numpy object type when we'd like to check it against string and unicode types as well -- the [['a', 'b', 'c']] list in your example appears to be cast to the numpy unicode array type in your example by the time it reaches that line. Sound right?

I'd also be curious to hear what you had in mind in terms of longer term solution, i.e., what would replace check_array if deprecated?

Perhaps we need a deprecation cycle

We wouldn't deprecate check_array entirely, but we would warn for two releases that "In the future, this data with dtype('Uxx') would be rejected because it is not of a numeric dtype."

Hi, I'm the newcomer here, can I take this?

There is already a PR open on this one #10229, so I would encourage you to find another issue to work on.

If you are a newcomer I would encourage you to read our contribution guidelines, in particular look at issues with the "good first issue" tag.

Iesteve, I find this issue by this tag.
As far as I see, it was reopened, so I think, that I can help.
Nevertheless thank you, I will find another issue, where I can start my contributing way :)

I'm confused about the fix here. I'm using sklearn 0.23.2, and the behavior that @jnothman called out as a problem is still the same as he described. To reproduce:

arr = np.array([[1, 's'],
                [1, 1]])
check_array(arr, dtype='numeric')

FutureWarning: Beginning in version 0.22, arrays of bytes/strings will be converted to decimal numbers
 if dtype='numeric'. It is recommended that you convert the array to a float dtype before using it in 
scikit-learn, for example by using your_array = your_array.astype(np.float64).
  return f(**kwargs)

array([['1', 's'],
       ['1', '1']], dtype='<U21')

Now everything's a string. It looks like the warning message was added in 2018 but the behavior was never changed. Am I missing something?

@zeromh I agree. We have a new opportunity to fix it in #18496.