Numpyでゼロ要素がある配列のEntropyを計算

配列に0要素がある時は普通にやると値がNanになってしまう。

>>> import numpy as np
>>> a = np.array([0.1,0.3,0,0.05,0.15,0.6,0])
>>> np.log(a)
array([-2.30258509, -1.2039728 ,        -inf, -2.99573227, -1.89711998,
       -0.51082562,        -inf])
>>> a*np.log(a)
array([-0.23025851, -0.36119184,         nan, -0.14978661, -0.284568  ,
       -0.30649537,         nan])
>>> -sum(a*np.log(a))
nan

そういう時はmasked arraysを使う。

>>> import numpy as np
>>> a = np.array([0.1,0.3,0,0.05,0.15,0.6,0])
>>> np.ma.log(a)
masked_array(data = [-2.3025850929940455 -1.2039728043259361 -- -2.995732273553991
 -1.8971199848858813 -0.5108256237659907 --],
             mask = [False False  True False False False  True],
       fill_value = 1e+20)

>>> a*np.ma.log(a)
masked_array(data = [-0.23025850929940456 -0.3611918412977808 -- -0.14978661367769955
 -0.28456799773288216 -0.30649537425959444 --],
             mask = [False False  True False False False  True],
       fill_value = 1e+20)

>>> -(a*np.ma.log(a)).sum()
1.3323003362673613

ちなみに単純にリスト内包でやる時はこう。

>>> import numpy as np
>>> a = np.array([0.1,0.3,0,0.05,0.15,0.6,0])
>>> -sum([v*math.log(v) if v > 0 else 0 for v in a])
1.3323003362673613

どっちがいいんだろ。