pandas grouper vs time grouper

Question

The new pandas version deprecates the TimeGrouper, so we should use the regular Grouper.

The old code:

df['column_name'].groupby(pd.TimeGrouper("M")).mean().plot()

works fine in the old version of pandas. However, none of:

df.groupby(pd.Grouper(key='column_name', freq="M")).mean().plot()
df['column_name'].groupby(pd.Grouper(freq="M")).mean().plot()

works in the new version. Eiter the key is considered to be missing, or pandas complains about:

Only valid with DatetimeIndex, TimedeltaIndex or PeriodIndex, but got an instance of 'Float64Index'

edit

import pandas as pd

df = pd.DataFrame({'column_name':['2017-01-01', '2017-01-02'],
                  'column_value':[1,3]})

df

df.index = pd.DatetimeIndex(df.column_name)

df.index

# old version
df['column_value'].groupby(pd.TimeGrouper("M")).mean().plot()

# new version
df.groupby(pd.Grouper(key='column_value', freq="M")).mean().plot()

If the key column is not a datetime then grouper with freq attribute wont work . — Bharath M Shetty, Oct 30 '17 at 13:00
But that was already true for the old version. So the code snippet should not have worked in the first place? assuming that the index would be a regular range index. — Georg Heiler, Oct 30 '17 at 13:01
Oh sorry I agree. Can you add example data so we try to reproduce the same. — Bharath M Shetty, Oct 30 '17 at 13:02
You are not converting column_name to be datetime so grouper wont work. Its not about the index its about the key column. — Bharath M Shetty, Oct 30 '17 at 13:13

score 15 · Accepted Answer · answered Oct 30 '17 at 13:12

15

As I said in the comment key should be datetime in grouper. Timegrouper by default converts it to datetime so use

df['column_name'] = pd.to_datetime(df['column_name'])
# new version
df.groupby(pd.Grouper(key='column_name', freq="M")).mean().plot()

answered Oct 30 '17 at 13:12

Bharath M Shetty

30,075
6
57
108

pandas grouper vs time grouper

edit

1 Answers1