I noticed that numpy.sum produces large enough round-off errors for the tests to fail with as ... def kahan_sum(a, axis=0): s = numpy.zeros(a.shape[:axis] + ...
確定! 回上一頁