Skip to content

np.asscalar should pass through scalars #4701

New issue

Have a question about this project? Sign up for a free GitHub account to open an issue and contact its maintainers and the community.

Sign up for GitHub

By clicking “Sign up for GitHub”, you agree to our terms of service and privacy statement. We’ll occasionally send you account related emails.

Already on GitHub? Sign in to your account

Jump to bottom
Closed
mforbes opened this issue May 12, 2014 · 4 comments · Fixed by #12123
Closed

np.asscalar should pass through scalars #4701

mforbes opened this issue May 12, 2014 · 4 comments · Fixed by #12123
Labels
54 - Needs decision

Comments

@mforbes
Copy link
Contributor

mforbes commented May 12, 2014

Suggestion
It would be very convenient if np.asscalar() would pass through scalars. The use-case is to ensure that an argument is a scalar, regardless what is passed in (as long as it behaves as a scalar).

>>> np.asarray(1.0)
1.0
>>> np.asarray(1j)
1j

Presently these fail as the implementation simply looks for the .item() method, which scalars do not possess:

>>> np.asarray(1.0)
AttributeError                            Traceback (most recent call last)
<ipython-input-13-c6bc08be135e> in <module>()
----> 1 np.asscalar(1.0)

.../python2.7/site-packages/numpy/lib/type_check.pyc in asscalar(a)
    461 
    462     """
--> 463     return a.item()
    464 
    465 #-----------------------------------------------------------------------------

AttributeError: 'float' object has no attribute 'item'```

Rational
It is common to use np.asarray() to ensure that the argument is an array (in the case that it might be a list etc.). In an analogous use case, it would be useful to be able to use np.asscalar() to ensure that an argument is a scalar (as long as it can be converted to a scalar).

Workaround
Convert the argument to an array first.

np.asscalar(np.asarray(x))

Resolution
Perhaps code like the following would work:

def asscalar(a):
    try:
        return a.item()
    except AttributeError, e:
        return np.asarray(a).item()
@jarl-haggerty jarl-haggerty mentioned this issue Sep 27, 2014
Closed
@liufei11111 liufei11111 mentioned this issue Sep 27, 2014
Closed
@liufei11111 liufei11111 mentioned this issue Nov 6, 2014
Closed
liufei11111 added a commit to liufei11111/numpy that referenced this issue Nov 7, 2014
@liufei11111
BUG: pmt() method of financial.py will yield error because the two pa…
586620c 
…rameters passed in to np.where() is evaluated first before going into the function however, the denominator can be zero at that time. I allocated the intermediate result based on different formulas for rate == 0.0 and not equal to and fixes numpy#5046. Also, a easy fix for asocial doesn’t take integer because it assumes passing in an np.array. I convert everything to np.array first and then retrieve the item. Fixed numpy#4701
@dpdoughe
Copy link

I'd like to reopen this issue. This is not fixed in numpy 1.13. It doesn't have the fix mentioned here

In [4]: np.asscalar(3.4)
---------------------------------------------------------------------------
AttributeError                            Traceback (most recent call last)
<ipython-input-4-8460cfbb0c19> in <module>()
----> 1 np.asscalar(3.4)

/home/foo/.virtualenv/myapp/lib/python2.7/site-packages/numpy/lib/type_check.pyc in asscalar(a)
    478 
    479     """
--> 480     return a.item()
    481 
    482 #-----------------------------------------------------------------------------

AttributeError: 'float' object has no attribute 'item'

def asscalar(a):
    """
    Convert an array of size 1 to its scalar equivalent.

    Parameters
    ----------
    a : ndarray
        Input array of size 1.

    Returns
    -------
    out : scalar
        Scalar representation of `a`. The output data type is the same type
       returned by the input's `item` method.

    Examples
    --------
    >>> np.asscalar(np.array([24]))
    24

    """
    return a.item()

@eric-wieser eric-wieser reopened this Dec 22, 2017
@eric-wieser
Copy link
Member

Looks like the wrong issue number was used in the commit that closed this

seawolf631 added a commit to seawolf631/numpy that referenced this issue Dec 22, 2017
@seawolf631
Attempt to fix issue numpy#4701, np.asscalar should pass through scalars
68c57a0
@seawolf631 seawolf631 mentioned this issue Dec 22, 2017
Closed
seawolf631 added a commit to seawolf631/numpy that referenced this issue Dec 23, 2017
@seawolf631
Fixes to numpy#4701 and numpy#5126
a977598
@eric-wieser eric-wieser mentioned this issue Feb 25, 2018
Closed
SpoorthyBhat added a commit to SpoorthyBhat/numpy that referenced this issue Mar 1, 2018
@SpoorthyBhat
BUG: fixes numpy#4701 np.asscalar
e08d0be 
Allows np.scalar to pass through scalars
SpoorthyBhat added a commit to SpoorthyBhat/numpy that referenced this issue Mar 1, 2018
@SpoorthyBhat
BUG: fixes numpy#4701 related to numpy.asscalar
a944360 
Allows numpy.asscalar to pass scalars
SpoorthyBhat added a commit to SpoorthyBhat/numpy that referenced this issue Mar 1, 2018
@SpoorthyBhat
Merge pull request #1 from SpoorthyBhat/Asscalar_fix
3fb0757 
BUG: fixes numpy#4701 related to numpy.asscalar
SpoorthyBhat added a commit to SpoorthyBhat/numpy that referenced this issue Mar 4, 2018
@SpoorthyBhat
Revert "BUG: fixes numpy#4701 related to numpy.asscalar"
08db841
@M-Mueller M-Mueller mentioned this issue Sep 1, 2018
Closed
@eric-wieser
Copy link
Member

eric-wieser commented Sep 5, 2018 •
edited
Loading

We need to make a decision on this before we get even more PRs attempting to "fix" it. To date:

Un-tagging as easy to stop us getting more of these

@eric-wieser
Copy link
Member

eric-wieser commented Sep 5, 2018 •
edited
Loading

One of the main problems I have with asscalar is that it's just not consistent with isscalar - isscalar(x) currently does not imply asscalar(x) is x:

>>> x = np.float64
>>> isscalar(x)
True
>>> type(x)
<class 'numpy.float64'>
>>> type(asscalar(x))
<class 'float'>

Overall, we have three conflicting definitions of scalar:

  1. Is a python scalar - the type of object returned from asscalar
  2. Is a python or numpy scalar - objects meeting isscalar
  3. is a 0d array - np.ndim(x) == 0

Almost every time I see an internal use of isscalar, the intention was actually np.ndim(x) == 0. The exceptions are places that decide whether *= will modify in-place

@eric-wieser eric-wieser mentioned this issue Sep 5, 2018
Merged
@mattip mattip mentioned this issue Oct 9, 2018
Merged
gvgramazio added a commit to gvgramazio/trpo that referenced this issue Nov 5, 2018
@gvgramazio
Update train.py to avoid attribute errors
2493e79 
As explained in detail in a [numpy issue](numpy/numpy#4701), `np.asscalar` doesn't pass through scalars. This is because it looks for `item()` method that `int` and `float` don't have. A possible workaround suggest there is the one I proposed to you. In this way your script doesn't fail with other environments such as `LunarLanderContinuous-v2`.
@gvgramazio gvgramazio mentioned this issue Nov 5, 2018
Merged
@lostanlen lostanlen mentioned this issue Sep 30, 2019
Closed
Sign up for free to join this conversation on GitHub. Already have an account? Sign in to comment
Assignees
No one assigned
Labels
54 - Needs decision
Projects
None yet
Milestone
No milestone
4 participants
@eric-wieser @juliantaylor @mforbes @dpdoughe