Pandas 用概率填充缺失值

假设给一个列水果，我有57个香蕉，54个苹果和其他空值。 现在我想用

fillna

填充空值，其中57/（57+54）概率作为香蕉，54/（57+54）概率作为苹果，我应该怎么做

Fruit
------
None
Banana
Fruit
Banana
....(with 57 banana, 54 apple, 10 None)

设置

fruit = pd.Series(['banana'] * 57 + ['apple'] * 54 + [None] * 10, name='fruit')

nullfruit = fruit.isnull()
fruit.loc[nullfruit] = fruit.dropna().sample(nullfruit.sum()).values

nullfruit = fruit.isnull().values
u, c = np.unique(fruit.values[~nullfruit], return_counts=1)

fruit.loc[nullfruit] = np.random.choice(u, nullfruit.sum(), p=c / c.sum())

---

使用

pd.Series.sample

fruit = pd.Series(['banana'] * 57 + ['apple'] * 54 + [None] * 10, name='fruit')

nullfruit = fruit.isnull()
fruit.loc[nullfruit] = fruit.dropna().sample(nullfruit.sum()).values

nullfruit = fruit.isnull().values
u, c = np.unique(fruit.values[~nullfruit], return_counts=1)

fruit.loc[nullfruit] = np.random.choice(u, nullfruit.sum(), p=c / c.sum())

使用

np.random.choice

和

np.unique

fruit = pd.Series(['banana'] * 57 + ['apple'] * 54 + [None] * 10, name='fruit')

nullfruit = fruit.isnull()
fruit.loc[nullfruit] = fruit.dropna().sample(nullfruit.sum()).values

nullfruit = fruit.isnull().values
u, c = np.unique(fruit.values[~nullfruit], return_counts=1)

fruit.loc[nullfruit] = np.random.choice(u, nullfruit.sum(), p=c / c.sum())

设置

fruit = pd.Series(['banana'] * 57 + ['apple'] * 54 + [None] * 10, name='fruit')

nullfruit = fruit.isnull()
fruit.loc[nullfruit] = fruit.dropna().sample(nullfruit.sum()).values

nullfruit = fruit.isnull().values
u, c = np.unique(fruit.values[~nullfruit], return_counts=1)

fruit.loc[nullfruit] = np.random.choice(u, nullfruit.sum(), p=c / c.sum())

---

使用

pd.Series.sample

fruit = pd.Series(['banana'] * 57 + ['apple'] * 54 + [None] * 10, name='fruit')

nullfruit = fruit.isnull()
fruit.loc[nullfruit] = fruit.dropna().sample(nullfruit.sum()).values

nullfruit = fruit.isnull().values
u, c = np.unique(fruit.values[~nullfruit], return_counts=1)

fruit.loc[nullfruit] = np.random.choice(u, nullfruit.sum(), p=c / c.sum())

使用

np.random.choice

和

np.unique

fruit = pd.Series(['banana'] * 57 + ['apple'] * 54 + [None] * 10, name='fruit')

nullfruit = fruit.isnull()
fruit.loc[nullfruit] = fruit.dropna().sample(nullfruit.sum()).values

nullfruit = fruit.isnull().values
u, c = np.unique(fruit.values[~nullfruit], return_counts=1)

fruit.loc[nullfruit] = np.random.choice(u, nullfruit.sum(), p=c / c.sum())

---

您能添加样本和期望输出吗？您能添加样本和期望输出吗？