DEPR: Deprecate numpy argument in read_json

[WillAyd]

I've never really been clear on the purpose of the numpy argument in to_json. Some digging brought me here #3876 (comment) where it is explained that this maintains some kind of sequence to elements. To illustrate the only difference I could find

>>> pd.read_json('[{"a": 1, "b": 2}, {"b":2, "a" :1}]',numpy=False, orient='records')
   a  b
0  1  2
1  1  2

>>> pd.read_json('[{"a": 1, "b": 2}, {"b":2, "a" :1}]',numpy=True, orient='records')
   a  b
0  1  2
1  2  1

I might be missing the point but I don't understand why this would be useful. Objects in JSON are by definition not ordered, so this is non-compliant and I think just plain confusing.

So I think good to deprecate unless anyone has objections.

[WillAyd]

Should also note there are not tests that I can see that actually test numpy=True making any kind of difference

[lucaionescu]

Hi, I'd like to work on this one. What would be the best way to go? Removing the argument from the function and adding a deprecation warning to the next release notes?

[WillAyd]

You should be able to apply the decorate_kwarg decorator to read_json:

pandas/pandas/util/_decorators.py

Line 104 in bd8dbf9

def deprecate_kwarg(

And then update tests in pandas.tests.io.json . Right now this argument is used for parametrization but doesn't actually change anything, so could probably just remove parametrization

[WillAyd]

Don't remove the argument yet though just add warning

[alimcmaster1]

@WillAyd

According to user_guide/io.rst there are also some performance considerations.

Are we still good to deprecate do you think?

If ``numpy=True`` is passed to ``read_json`` an attempt will be made to sniff
an appropriate dtype during deserialization and to subsequently decode directly
to NumPy arrays, bypassing the need for intermediate Python objects.

This can provide speedups if you are deserialising a large amount of numeric
data:

   randfloats = np.random.uniform(-100, 1000, 10000)
   randfloats.shape = (1000, 10)
   dffloats = pd.DataFrame(randfloats, columns=list('ABCDEFGHIJ'))
   jsonfloats = dffloats.to_json()
   
   pd.read_json(jsonfloats) 
   pd.read_json(jsonfloats, numpy=True)

[WillAyd]

Yea should still deprecate