熊猫datetime64快速转化为str

Question

转换大型datetime64[D] Series (即DataFrame列的900 k行)花费的时间太长。我怎样才能加快速度？

import pandas as pd

df = pd.DataFrame(['2021-10-01']*900000, columns=['date'])   # 0.025286900 seconds
df = df.assign(date=df['date'].astype('datetime64[D]'))      # 0.105065900 

# Why is converting from datetime to str so slow?
df.assign(date=df['date'].dt.strftime('%Y-%m-%d'))           # 5.600835100 seconds. 

txt = str(df)                                                # 0.006202600
# Converting the entire DataFrame to a str is much faster 
# than converting a column directly, despite a similar display format!

有a related question，它询问如何快速地从str转换到datetime。但我的瓶颈是(令人惊讶的)相反的；从datetime[D]到str的转换太慢了。

Answer 1

下面是一个更快的解决方案：

np.datetime_as_string(df['date'], unit='D')   # 0.375799800 seconds.

它仍然比应该花费的时间长(从datetime转换比从str转换慢有什么意义)，但它要快得多。