我有一个数据框架"df1"由1245行组成,具有列文本(对象类型)和主题(对象类型)。主题列包含不同的数字,对应于文本标签。下面是一个例子:
text topic
1207 June 2019: The French Facility for Global Envi... 3 12 7
1208 May 2019: Participants from multi-stakeholder ... 8
1209 2 July 2019: UN Member States have reached agr... 1 7
1210 30 June 2019: The G20 Leaders’ Summit and asso... 7 8 9 11 12 13 14 15 17
我想获得一个像这样的热编码形式(也添加一个'S'在列名之前的数字):
text S1 S2 S3 ..... S7 S8 S9 etc.
1207 June 2019: The French Facility for Global Envi... 0 0 1 ..... 1 0 0
1208 May 2019: Participants from multi-stakeholder ... 0 0 0 ...... 0 1 0
1209 2 July 2019: UN Member States have reached agr... 1 0 0 ..... 1 0 0
1210 30 June 2019: The G20 Leaders’ Summit and asso... 0 0 0 ......1 1 1
这里的"困难"是我的文本是多标签的,所以简单的one-hot编码代码不适合我的情况。你知道吗?
如果只使用pandas,您可以这样做:
import pandas as pd
data = [['June 2019: The French Facility for Global Envi...', '3 12 7'],
['May 2019: Participants from multi-stakeholder ...','8'],
['2 July 2019: UN Member States have reached agr...','1 7'],
['30 June 2019: The G20 Leaders’ Summit and asso...','7 8 9 11 12 13 14 15 17']]
df = pd.DataFrame(data , columns=['text', 'topic'])
# creating list of strings where each value is one number out of topic column
unique_values = ' '.join(df['topic'].values.tolist()).split(' ')
# creating new column for each value in unique_values
for number in unique_values:
df[f'S{number}'] = 0
# changing 0 to 1 for every Snumber column where topic contains number
for idx, row in df.iterrows():
for number in row['topic'].split(' '):
df.loc[idx, f'S{number}'] = 1
df.drop('topic', axis=1, inplace=True)
结果:
text S3 S12 S7 S8 S1 S9 S11 S13 S14 S15 S17
0 June 2019: The French Facility for Global Envi... 1 1 1 0 0 0 0 0 0 0 0
1 May 2019: Participants from multi-stakeholder ... 0 0 0 1 0 0 0 0 0 0 0
2 2 July 2019: UN Member States have reached agr... 0 0 1 0 1 0 0 0 0 0 0
3 30 June 2019: The G20 Leaders’ Summit and asso... 0 1 1 1 0 1 1 1 1 1 1
稍微修改一下数据(出于可读性原因…):
from io import StringIO
import pandas as pd
s = """id,text,topic
1207,One,1 2 5
1208,Two,3
1209,Three,1 4
1210,Four,1 2 3"""
df = pd.read_csv(StringIO(s))
df.topic = df.topic.str.split(' ').apply(lambda x: [int(y) for y in x])
b = np.zeros((df.topic.size, max(max(x) for x in df.topic) + 1))
for i in df.index:
b[i, df.topic[i]] = 1
idx = {'id': df.id, 'text': df.text}
idx.update({f'S{i}': b[:, i] for i in range(1, b.shape[1])})
idx
df = pd.DataFrame(idx)
print(df.set_index('id').to_markdown())
这给你:
| id | text | S1 | S2 | S3 | S4 | S5 |
|-----:|:-------|-----:|-----:|-----:|-----:|-----:|
| 1207 | One | 1 | 1 | 0 | 0 | 1 |
| 1208 | Two | 0 | 0 | 1 | 0 | 0 |
| 1209 | Three | 1 | 0 | 0 | 1 | 0 |
| 1210 | Four | 1 | 1 | 1 | 0 | 0 |