Can't get sympy.functions.atan2 to work with _eval_evalf
Ville Bergholm
I've been trying to use sympy.Symbol to implement a system where the numerical value of a symbolic quantity is only fixed when it is evaluated using evalf().To accomplish this I've redefined _eval_evalf() to return the aforementioned value (a fixed 5.0 in the toy example below).
This works fine with basic algebraic operations and most SymPy functions. However, for some reason atan2 seems to behave differently and _eval_evalf is never called.Is this a bug, or have I misunderstood the purpose of _eval_evalf?
import sympy
import sympy.functions as syf
class A(sympy.Symbol):
def _eval_evalf(self, prec):
print('*')
return sympy.Float(5.0)
x = A('X')
y = A('Y')
print('polynomial:')
print((2*x-4+y).evalf()) # works fine
print('atan:')
print(syf.atan(y / x).evalf()) # works fine
print('beta:')
print(syf.beta(y, x).evalf()) # works fine
print('atan2:')
print(syf.atan2(y, x).evalf()) # prints "atan2(Y, X)", does not call A._eval_evalf
Oscar Benjamin
$ git diff
diff --git a/sympy/functions/elementary/trigonometric.py
b/sympy/functions/elementary/trigonometric.py
index 635e7f2..b7a0ad1 100644
--- a/sympy/functions/elementary/trigonometric.py
+++ b/sympy/functions/elementary/trigonometric.py
@@ -3070,4 +3070,4 @@ def fdiff(self, argindex):
def _eval_evalf(self, prec):
y, x = self.args
if x.is_extended_real and y.is_extended_real:
- super(atan2, self)._eval_evalf(prec)
+ return super(atan2, self)._eval_evalf(prec)
Also it checks if the args are real so it looks like you might need to
declare your variables a and y with real=True. If you do that then the
result is caculated as: 0.785398163397448
Can you open an issue on Github for this?
> You received this message because you are subscribed to the Google Groups "sympy" group.
> To unsubscribe from this group and stop receiving emails from it, send an email to sympy+un...@googlegroups.com.
> To view this discussion on the web visit https://groups.google.com/d/msgid/sympy/4480befb-c8d3-48f1-a1be-b5e8bd55a0be%40googlegroups.com.
Kalevi Suominen
Ville Bergholm
On Tuesday, August 20, 2019 at 5:43:40 PM UTC+3, Oscar wrote:
Can you open an issue on Github for this?
David Bailey
It seems as though the symbolic integration process fails when it can't find a solution:
integrate(sin(x**n),x)
Traceback (most recent call last):
File "<stdin>", line 1, in <module>
File
"C:\SymPyWorkbook\lib\site-packages\sympy\integrals\integrals.py",
line 1477, in integrate
return integral.doit(**doit_flags)
File
"C:\SymPyWorkbook\lib\site-packages\sympy\integrals\integrals.py",
line 541, in doit
function, xab[0], **eval_kwargs)
File
"C:\SymPyWorkbook\lib\site-packages\sympy\integrals\integrals.py",
line 1012, in _eval_integral
h = meijerint_indefinite(g, x)
File
"C:\SymPyWorkbook\lib\site-packages\sympy\integrals\meijerint.py",
line 1621, in meijerint_indefinite
res = _meijerint_indefinite_1(f.subs(x, x + a), x)
File
"C:\SymPyWorkbook\lib\site-packages\sympy\integrals\meijerint.py",
line 1684, in _meijerint_indefinite_1
if b < 0 or f.subs(x, 0).has(nan, zoo):
File
"C:\SymPyWorkbook\lib\site-packages\sympy\core\relational.py",
line 304, in __nonzero__
raise TypeError("cannot determine truth value of Relational")
TypeError: cannot determine truth value of Relational
I suppose I expected it to return an unevaluated integral - integral(sin(x**n),x)
David
Aaron Meurer
indicates a bug in SymPy. And yes, integrate() should always return
unevaluated when it can't compute the integral.
Aaron Meurer
> You received this message because you are subscribed to the Google Groups "sympy" group.
> To unsubscribe from this group and stop receiving emails from it, send an email to sympy+un...@googlegroups.com.
Oscar Benjamin
Can you open an issue on Github please. That's a bug that should be fixed.
Oscar
David Bailey
Hi David, Can you open an issue on Github please. That's a bug that should be fixed.
The problem is, I tried to join Github a while back, and then needed a password reset. It said it had sent me an email to reset my password, but it never appeared (and I looked in junk). After that happened several times, I gave up.
However, if that problem is already solved, that's fine, I was really just reporting it - I suspect the integral doesn't have a closed form for variable n. Do you release beta tests of the next version of sympy?
Or did you mean 'should' in the sense of 'needs to be'?
David
Kalevi Suominen
if b < 0 or f.subs(x, 0).has(nan, zoo):
David Bailey
"TypeError: cannot determine truth value of Relational" generally indicates a bug in SymPy. And yes, integrate() should always return unevaluated when it can't compute the integral.
I imagine that symbolic integration may be littered by tricky problems of this sort, and I wonder if the integrate code should be wrapped in some sort of exception handler so that it exits and returns unevaluated.
David
Oscar Benjamin
exception and fix the code. Maybe after that check (if it didn't
raise) something more useful could be done than return unevaluated.
Oscar
Oscar Benjamin
Aaron Meurer
it. If you run Python from within the repo directory, it will import
the git version. Since SymPy is pure Python there are no builds
required.
Aaron Meurer
>
> Or did you mean 'should' in the sense of 'needs to be'?
>
> David
>
> You received this message because you are subscribed to the Google Groups "sympy" group.
> To unsubscribe from this group and stop receiving emails from it, send an email to sympy+un...@googlegroups.com.
Oscar Gustafsson
To view this discussion on the web visit https://groups.google.com/d/msgid/sympy/CAKgW%3D6Lc%2BZ%2Bd-8dawVzH3h8NwpZvFvyothm1p01vSZidL%3DQbxA%40mail.gmail.com.