如何检测和过滤时间序列数据中的峰值？

Question

I have a pandas dataframe of user logins like this:

    id     datetime_login 
    646  2017-03-15 15:30:25
    611  2017-04-14 11:38:30
    611  2017-05-15 08:49:01
    651  2017-03-15 15:30:25
    611  2017-03-15 15:30:25
    652  2017-03-08 14:03:56
    652  2017-03-08 14:03:56
    652  2017-03-15 15:30:25
    654  2017-03-15 15:30:25
    649  2017-03-15 15:30:25
    902  2017-09-09 15:00:00
    902  2017-02-13 16:39:53
    902  2017-11-15 12:00:00
    902  2017-11-15 12:00:00
    902  2017-09-09 15:00:00
    902  2017-05-15 08:48:47
    902  2017-11-15 12:00:00

After plotting the logins:

df.datetime_login = df.datetime_login.apply(lambda x: str(x)[:10])
df.datetime_login = df.datetime_login.apply(lambda x: date(int(x[:4]), int(x[5:7]), int(x[8:10])))


fig, ax = subplots()
df.datetime_login.value_counts().sort_index().plot(figsize=(25,10), colormap='jet',fontsize=20)

1. How can I detect in my plot the peaks in the time series data?

2. How can I filter into an array the peaks in my time series data?

I tried to:

import peakutils
indices = peakutils.indexes(df, thres=0.4, min_dist=1000)
print(indices) 

However, I got:

TypeError: unsupported operand type(s) for -: 'datetime.date' and 'int'

However, I got:

Answer 1

Where df.datetime_login.value_counts().sort_index().plot(figsize=(25,10), colormap='jet',fontsize=20) plots:

Let's try the following, you need to use the series returned by value_counts instead of your original df, peakutils.indexes :

df_counts = df.datetime_login.value_counts().sort_index()
df_counts[peakutils.indexes(df_counts, thres=0.4, min_dist=1000)]

Output:

2017-03-15 15:30:25    6
Name: datetime_login, dtype: int64