使用numpy向量化检查pandas列中的参考列表

问题描述

我有一个参考名单

ref = ['September', 'August', 'July', 'June', 'May', 'April', 'March']

还有一个数据框

df = pd.DataFrame({'Month_List': [['July'], ['August'], ['July', 'June'], ['May', 'April', 'March']]})
df
    Month_List
0   [July]
1   [August]
2   [July, June]
3   [May, April, March]

我想检查引用列表中的哪些元素存在于每一行中，然后转换为二进制列表

I want to check which elements from reference list is present in each row, and convert into binary list

我可以使用apply

def convert_month_to_binary(ref,lst):
    s = pd.Series(ref)
    return s.isin(lst).astype(int).tolist()  

df['Binary_Month_List'] = df['Month_List'].apply(lambda x: convert_month_to_binary(ref, x))
df

    Month_List          Binary_Month_List
0   [July]              [0, 0, 1, 0, 0, 0, 0]
1   [August]            [0, 1, 0, 0, 0, 0, 0]
2   [July, June]        [0, 0, 1, 1, 0, 0, 0]
3   [May, April, March] [0, 0, 0, 0, 1, 1, 1]

但是，在大型数据集上使用apply非常慢，因此我希望使用numpy向量化.如何改善我的表现?

However, using apply on large datasets is very slow and hence I am looking to use numpy vectorization. How can I improve my performance?

扩展名:

我想使用numpy vectorization，因为我现在需要在此列表上应用另一个功能

I wanted to use numpy vectorization because I need to now apply another function on this list

我正在尝试这样，但是性能非常慢.与apply

I am trying like this, but performance is very slow. Similar results with apply

def count_one(lst):
    index = [i for i, e in enumerate(lst) if e != 0] 
    return len(index)

vfunc = np.vectorize(count_one)
df['Value'] = vfunc(df['Binary_Month_List']) 

推荐答案

在熊猫中更好不要这样使用list ，但可以使用 MultiLabelBinarizer a>和 DataFrame.reindex 对于添加的缺少类别，最后将值转换为numpy数组，如果性能很重要，则将其转换为list:

from sklearn.preprocessing import MultiLabelBinarizer

mlb = MultiLabelBinarizer()
df1 = pd.DataFrame(mlb.fit_transform(df['Month_List']),columns=mlb.classes_)
df['Binary_Month_List'] = df1.reindex(columns=ref, fill_value=0).values.tolist()

或使用 Series.str.join ， Series.str.get_dummies 和reindex:

df['Binary_Month_List'] = (df['Month_List'].str.join('|')
                                           .str.get_dummies()
                                           .reindex(columns=ref, fill_value=0)
                                           .values
                                           .tolist())
print (df)
            Month_List      Binary_Month_List
0               [July]  [0, 0, 1, 0, 0, 0, 0]
1             [August]  [0, 1, 0, 0, 0, 0, 0]
2         [July, June]  [0, 0, 1, 1, 0, 0, 0]
3  [May, April, March]  [0, 0, 0, 0, 1, 1, 1]

性能不同:

df = pd.concat([df] * 1000, ignore_index=True)

from sklearn.preprocessing import MultiLabelBinarizer

mlb = MultiLabelBinarizer()

In [338]: %timeit (df['Month_List'].str.join('|').str.get_dummies().reindex(columns=ref, fill_value=0).values.tolist())
31.4 ms ± 1.41 ms per loop (mean ± std. dev. of 7 runs, 10 loops each)

In [339]: %timeit pd.DataFrame(mlb.fit_transform(df['Month_List']),columns=mlb.classes_).reindex(columns=ref, fill_value=0).values.tolist()
5.57 ms ± 94.5 µs per loop (mean ± std. dev. of 7 runs, 100 loops each)

In [340]: %timeit df['Binary_Month_List2'] =df.Month_List.explode().str.get_dummies().sum(level=0).reindex(columns=ref, fill_value=0).values.tolist()
58.6 ms ± 461 µs per loop (mean ± std. dev. of 7 runs, 10 loops each)

这篇关于使用numpy向量化检查pandas列中的参考列表的文章就介绍到这了，希望我们推荐的答案对大家有所帮助，也希望大家多多支持！