On Friday, April 21, 2017 at 9:11:37 AM UTC+1, Marc Mezzarobba wrote:
>
> Dima Pasechnik wrote:
> > but it looks as if it might be a coercion problem.
> > Any ideas where to look?
>
> Not really, but it does look like the common parent
> discovered by the coercion system is incorrect:
>
> sage: import numpy as np
> sage: a = np.float('1.5')
> sage: b = np.float32('1.5')
>
> sage: get_coercion_model().common_parent(b, polygen(RR))
> Univariate Polynomial Ring in x over Real Field with 53 bits of
> precision
>
> sage: RR.coerce(a)
> 1.50000000000000
>
> sage: RR.coerce(b)
> ...
> TypeError: no canonical coercion from <type 'numpy.float32'> to Real
> Field with 53 bits of precision
>
> sage: get_coercion_model().common_parent(b, RR)
> <type 'numpy.float32'>
>
This does not look like this is the root cause of the problem.
Namely, note that also
sage: a128 = np.float128('1.5')
sage: RR.coerce(a128)
# throws the same as above TypeError, but
sage: a128*x
# does not print the numpy warning.
what is also somewhat puzzling is
that while
sage: np.float32('1.5')*x
# does print the numpy warning
sage: x*np.float32('1.5')
# just works (i.e. does not print the numpy warning)
So this is a subtle combination of an apparent bug in numpy (or in clang)
with
some Sage coercion weirdness.
Apparently numpy's floats are coerced into RDF for the purpose of
dealing with polynomials, see py_scalar_to_element in structure/coerce.pyx,
but apparently this only happens
for
x*np.float32('1.5'), and not for np.float32('1.5')*x ?
Dima
> --
> Marc
>
>
--
You received this message because you are subscribed to the Google Groups
"sage-devel" group.
To unsubscribe from this group and stop receiving emails from it, send an email
to sage-devel+unsubscr...@googlegroups.com.
To post to this group, send email to sage-devel@googlegroups.com.
Visit this group at https://groups.google.com/group/sage-devel.
For more options, visit https://groups.google.com/d/optout.
