Enforcing compatibility between numpy 1.8 and 1.9 nansum? -

Enforcing compatibility between numpy 1.8 and 1.9 nansum? -

i have code needs behave identically independent of numpy version, underlying np.nansum function has changed behavior such np.nansum([np.nan,np.nan]) 0.0 in 1.9 , nan in 1.8. <=1.8 behavior 1 prefer, more of import thing code robust against numpy version.

the tricky thing is, code applies arbitrary numpy function (generally, np.nan[something] function) ndarray. there way forcefulness new or old numpy nan[something] functions conform old or new behavior shy of monkeypatching them?

a possible solution can think of outarr[np.allnan(inarr, axis=axis)] = np.nan, there no np.allnan function - if best solution, best implementation np.all(np.isnan(arr), axis=axis) (which require supporting np>=1.7, that's ok)?

in numpy 1.8, nansum defined as:

class="lang-py prettyprint-override">a, mask = _replace_nan(a, 0) if mask none: homecoming np.sum(a, axis=axis, dtype=dtype, out=out, keepdims=keepdims) mask = np.all(mask, axis=axis, keepdims=keepdims) tot = np.sum(a, axis=axis, dtype=dtype, out=out, keepdims=keepdims) if np.any(mask): tot = _copyto(tot, np.nan, mask) warnings.warn("in numpy 1.9 sum along empty slices zero.", futurewarning) homecoming tot

in numpy 1.9, is:

class="lang-py prettyprint-override">a, mask = _replace_nan(a, 0) homecoming np.sum(a, axis=axis, dtype=dtype, out=out, keepdims=keepdims)

i don't think there way create new nansum behave old way, given original nansum code isn't long, can include re-create of code (without warning) if care preserving pre-1.8 behavior?

note _copyto can imported numpy.lib.nanfunctions

numpy