On 1 April 2010 01:40, Charles R Harris <[email protected]> wrote: > > > On Wed, Mar 31, 2010 at 11:25 PM, <[email protected]> wrote: >> >> On Thu, Apr 1, 2010 at 1:22 AM, <[email protected]> wrote: >> > On Thu, Apr 1, 2010 at 1:17 AM, Charles R Harris >> > <[email protected]> wrote: >> >> >> >> >> >> On Wed, Mar 31, 2010 at 6:08 PM, <[email protected]> wrote: >> >>> >> >>> On Wed, Mar 31, 2010 at 7:37 PM, Warren Weckesser >> >>> <[email protected]> wrote: >> >>> > T J wrote: >> >>> >> On Wed, Mar 31, 2010 at 1:21 PM, Charles R Harris >> >>> >> <[email protected]> wrote: >> >>> >> >> >>> >>> Looks like roundoff error. >> >>> >>> >> >>> >>> >> >>> >> >> >>> >> So this is "expected" behavior? >> >>> >> >> >>> >> In [1]: np.logaddexp2(-1.5849625007211563, -53.584962500721154) >> >>> >> Out[1]: -1.5849625007211561 >> >>> >> >> >>> >> In [2]: np.logaddexp2(-0.5849625007211563, -53.584962500721154) >> >>> >> Out[2]: nan >> >>> >> >> >>> > >> >>> > Is any able to reproduce this? I don't get 'nan' in either 1.4.0 or >> >>> > 2.0.0.dev8313 (32 bit Mac OSX). In an earlier email T J reported >> >>> > using >> >>> > 1.5.0.dev8106. >> >>> >> >>> >> >>> >> >>> >>> np.logaddexp2(-0.5849625007211563, -53.584962500721154) >> >>> nan >> >>> >>> np.logaddexp2(-1.5849625007211563, -53.584962500721154) >> >>> -1.5849625007211561 >> >>> >> >>> >>> np.version.version >> >>> '1.4.0' >> >>> >> >>> WindowsXP 32 >> >>> >> >> >> >> What compiler? Mingw? >> > >> > yes, mingw 3.4.5. , official binaries release 1.4.0 by David >> >> sse2 Pentium M >> > > Can you try the exp2/log2 functions with the problem data and see if > something goes wrong?
Works fine for me. If it helps clarify things, the difference between the two problem input values is exactly 53 (and that's what logaddexp2 does an exp2 of); so I can provide a simpler example: In [23]: np.logaddexp2(0, -53) Out[23]: nan Of course, for me it fails in both orders. Anne _______________________________________________ NumPy-Discussion mailing list [email protected] http://mail.scipy.org/mailman/listinfo/numpy-discussion
