MAINT: Add parameter checks to polynomial integration functions. #9930
Conversation
charris
commented
Oct 26, 2017
It was not being checked that the `lbnd` and `scl` parameters were scalars as required. Closes numpy#9901.
The `lbnd` and `scl` parameters of the various polynomial integration functions are documented to be scalars. Add tests that those parameters are checked and ValueError raised when it is not so.
d58bdf6 to
050cd17
Compare
| if np.ndim(lbnd) != 0: | ||
| raise ValueError("lbnd must be a scalar.") | ||
| if np.ndim(scl) != 0: | ||
| raise ValueError("scl must be a scalar.") |
There was a problem hiding this comment.
Nit: errors above and below have no terminating period, nor do the builtin python error messages.
| @@ -203,6 +203,9 @@ def test_legint(self): | |||
| assert_raises(ValueError, leg.legint, [0], .5) | |||
| assert_raises(ValueError, leg.legint, [0], -1) | |||
| assert_raises(ValueError, leg.legint, [0], 1, [0, 0]) | |||
| assert_raises(ValueError, leg.legint, [0], lbnd=[0]) | |||
| assert_raises(ValueError, leg.legint, [0], scl=[0]) | |||
| assert_raises(ValueError, leg.legint, [0], axis=.5) | |||
There was a problem hiding this comment.
Shouldn't this raise a TypeError?
There was a problem hiding this comment.
Hmm. I was just following the documentation ... It would probably be more correct if all the errors checked here were Type errors except possibly the third and sixth.
050cd17 to
06e62db
Compare
|
Updated. |
| @@ -214,6 +214,9 @@ def test_chebint(self): | |||
| assert_raises(ValueError, cheb.chebint, [0], .5) | |||
| assert_raises(ValueError, cheb.chebint, [0], -1) | |||
| assert_raises(ValueError, cheb.chebint, [0], 1, [0, 0]) | |||
| assert_raises(TypeError, cheb.chebint, [0], lbnd=[0]) | |||
| assert_raises(TypeError, cheb.chebint, [0], scl=[0]) | |||
| assert_raises(ValueError, cheb.chebint, [0], axis=.5) | |||
There was a problem hiding this comment.
This strikes me as backwards. The first two should be ValueErrors, and the second should be a TypeError.
I don't think there is any precedent in numpy for "wrong number of dimensions" → TypeError, but there absolutely is precedent for isinstance(axis, float) → TypeError
There was a problem hiding this comment.
Floats here are fine as long as they are integer valued, same with cnt. Might not be the way things are done now, but this is very old code. What is done is conversion to int followed by a check for equality against the original value.
If you think the first two should be ValueError, then the original was correct. However, the ndim check is for scalar, so passing a list could be considered a TypeError.
There was a problem hiding this comment.
I see your point - the int(x) == x check is consistent with ValueError, even if neither are consistent with the axis argument in the rest of numpy.
So yes, I prefered the original - sorry for not being clearer there.
numpy/polynomial/polynomial.py
Outdated
| if np.ndim(lbnd) != 0: | ||
| raise TypeError("lbnd must be a scalar.") | ||
| if np.ndim(scl) != 0: | ||
| raise TypeError("scl must be a scalar.") | ||
| if iaxis != axis: | ||
| raise ValueError("The axis must be integer") |
There was a problem hiding this comment.
This should be operator.index, not int(axis) == axis
There was a problem hiding this comment.
In fact, this line and iaxis should just be removed, and the normalize_axis_index below allowed to throw a more useful error.
There was a problem hiding this comment.
But that's not part of this PR, so don't worry about it
06e62db to
050cd17
Compare
|
OK, reverted to original. |
|
I'm guessing you'd prefer the terminating periods to stay, and I don't think it's worth me arguing over. Does this need a release note? If not, LGTM. |
|
Non-scalars should have been causing errors or bad results all along, so either a bug fix or just better error messages. I don't think a release note is needed. |
|
Fair call, thanks! I'll open an issue about the |