There is a bug in the np.einsum code in numpy 1.14 https://github.com/numpy/numpy/pull/10371
That function is used in many different places in marxs, thus I'll freeze the version for testing to 1.13 for now. Once numpy releases a bug fix, testing can continue with current numpy.
Also, numpy 1.14 changes the output format for printing numbers on the screen, so doctests will have to be updated at that point.
It;s probably the easiest to drop any testing with older numpy at that point.
There is a bug in the np.einsum code in numpy 1.14 https://github.com/numpy/numpy/pull/10371 That function is used in many different places in marxs, thus I'll freeze the version for testing to 1.13 for now. Once numpy releases a bug fix, testing can continue with current numpy.
Also, numpy 1.14 changes the output format for printing numbers on the screen, so doctests will have to be updated at that point. It;s probably the easiest to drop any testing with older numpy at that point.