python - numpy.nan weird behaviour in hashed object -

i experiencing weird behaviour when using numpy.nan can't understand. here minimal example:

from numpy import nan  def _bool3key(x):   """   defines keys used order list.   allowed values true, false, 1,0 , numpy.nan.   """    return _bool3key.__logic_sort__[x] _bool3key.__logic_sort__ = {0:-1, nan:0 , 1:1}  def and3(*args):    return min(*args,key=_bool3key)   def f(x):   """long function produces in output vector containing 0, nan , 1s.   pass # 

sometimes and3 function fails, despite in vector returned f(x) there 0, nan, or 1 values: reason nan not of type numpy.nan...

for example v = f(x) produced vector [nan,nan]. if try type: v[0] nan false (which causes and3 not work); weird thing though v[1] nan true.

what causing behaviour? how can correctly use nan value in and3 function??

if use

_bool3key.__logic_sort__ = {0:-1, nan:0 , 1:1} 

then 1 problem might run float('nan') not recognized same key np.nan:

in [17]: _bool3key(float('nan')) keyerror: nan 

here workaround:

def _bool3key(x, logic_sort={0: -1, 1: 1}):     """     defines keys used order list.     allowed values true, false, 1,0 , nan.     """     return 0 if np.isnan(x) else logic_sort[x] 

also, attribute lookups slower local variable lookups, you'll better performance defining logic_sort default parameter making function attribute. don't have define outside of function, , little easier read too.