Question

我注意到基于MultiIndex的合并DataFrames上的pandas非常慢。分配值有时也很慢

import pandas as pd
import numpy as np
from pandas_datareader import data
import datetime
import string
import random

start = datetime.datetime(2002, 1, 1)
end = datetime.datetime(2018, 1, 1)


def id_generator(size=6, chars=string.ascii_uppercase + string.digits):
    return ''.join(random.choice(chars) for _ in range(size))

columns = [id_generator() for i in range(1000)]
dateindex = pd.date_range(start, end)
df = pd.DataFrame(np.random.randint(1, 100, (len(dateindex), len(columns))), columns=columns, index=dateindex)
df.columns = df.columns.rename('Name')
df.index = df.index.rename('Date')

df1 = df.pct_change(1).stack().rename('change1').to_frame()
df2 = df.pct_change(2).stack().rename('change2').to_frame()

df3 = df1.reset_index()
df4 = df2.reset_index()



%timeit pd.merge(df1, df2, left_index=True, right_index=True)
In [11]: 46.7 s ± 656 ms per loop (mean ± std. dev. of 7 runs, 1 loop each)

%timeit pd.merge(df3, df4, on=['Date', 'Name'])
In [12]: 3.17 s ± 168 ms per loop (mean ± std. dev. of 7 runs, 1 loop each)

速度慢了10倍以上。有谁知道发生了什么？是否总是更好地将索引和连接放在列而不是MultiIndex上。

Answer 1

让我们使用join：

%timeit df1.join(df2)

1 loop, best of 3: 647 ms per loop

合并pandas MultiIndex非常慢

1 个答案: