It looks like some stricter type casting conversion rules for ufuncs in numpy 1.10 has broken a whole bunch of code which uses np.linalg.solve and np.einsum. Some objects which I thought were of dtype float were apparently actually generic objects getting implicitly converted. It will take some time to fix.
It looks like some stricter type casting conversion rules for ufuncs in numpy 1.10 has broken a whole bunch of code which uses
np.linalg.solveandnp.einsum. Some objects which I thought were of dtype float were apparently actually generic objects getting implicitly converted. It will take some time to fix.