熊猫分位数功能的日期？

Question

I have a dataframe of donation amounts and dates. I would like to see how long it took a certain proportion of the donations to come in (at what point did we have 25% of donations?, 75% ?). It looked like the Pandas quantile function would do what I want. However it seems to only want numbers, not dates. Is there a function that would do the same with dates ?

http://pandas.pydata.org/pandas-docs/dev/generated/pandas.core.groupby.DataFrameGroupBy.quantile.html#pandas.core.groupby.DataFrameGroupBy.quantile

Answer 1

就像Evert所说的那样，您可以将其临时转换为int 64计算，然后转换回datetime

YOUR_DATAFRAME.YOUR_DATE.astype('int64').quantile([.25,.5,.75]).astype('datetime64[ns]')

Answer 2

I had the same problem, in my case to split a timeseries for a machine learning problem.

I wrote the following based on the above answers by evert and steboc , and added the case where the dates might be written as strings:

def get_split_date(df, date_column, quantile): 

    """ Get the date on which to split a dataframe for timeseries splitting """ 

    # 1. convert date_column to datetime (useful in case it is a string) 
    # 2. convert into int (for sorting) 
    # 3. get the quantile 
    # 4. get the corresponding date
    # 5. return, pray that it works 

    quantile_date = pd.to_datetime(df[date_column], coerce = True).astype('int64').quantile(q=quantile).astype('datetime64[ns]')

    return quantile_date