How to resample a Pandas dataframe of mixed type?

Question

I generate a mixed type (floats and strings) Pandas DataFrame df3 with the following Python code:

df1 = pd.DataFrame(np.random.randn(dates.shape[0],2),index=dates,columns=list('AB'))
df1['C'] = 'A'
df1['D'] = 'Pickles'
df2 = pd.DataFrame(np.random.randn(dates.shape[0], 2),index=dates,columns=list('AB'))
df2['C'] = 'B'
df2['D'] = 'Ham'
df3 = pd.concat([df1, df2], axis=0)

When I resample df3 to a higher frequency I don't get the frame resampled to a higher rate but the how is ignored and I just get missing values:

df4 = df3.groupby(['C']).resample('M',  how={'A': 'mean', 'B': 'mean',  'D': 'ffill'})
df4.head()

Result:

                      B          A        D
C                                          
A 2014-03-31 -0.4640906 -0.2435414  Pickles
  2014-04-30        NaN        NaN      NaN
  2014-05-31        NaN        NaN      NaN
  2014-06-30 -0.5626360  0.6679614  Pickles
  2014-07-31        NaN        NaN      NaN

When I resample df3 to a lower frequency I don't get any resampling at all:

df5 = df3.groupby(['C']).resample('A',  how={'A': np.mean, 'B': np.mean,  'D': 'ffill'})
df5.head()

Result:

                      B          A        D
C                                          
A 2014-03-31        NaN        NaN  Pickles
  2014-06-30        NaN        NaN  Pickles
  2014-09-30        NaN        NaN  Pickles
  2014-12-31 -0.7429617 -0.1065645  Pickles
  2015-03-31        NaN        NaN  Pickles

I'm pretty sure that this has something to do with the mixed types because if I redo the annual down-sampling with just numerical columns everything works as expected:

df5b = df3[['A', 'B', 'C']].groupby(['C']).resample('A',  how={'A': np.mean, 'B': np.mean})
df5b.head()

Result:

                     B          A
  C                                 
  A 2014-12-31 -0.7429617 -0.1065645
    2015-12-31 -0.6245030 -0.3101057
  B 2014-12-31  0.4213621 -0.0708263
    2015-12-31 -0.0607028  0.0110456

But even when I switch to numerical types the resampling to higher frequency still doesn't work as I expected:

df4b = df3[['A', 'B', 'C']].groupby(['C']).resample('M',  how={'A': 'mean', 'B': 'mean'})
df4b.head()

Results:

                      B          A
C                                 
A 2014-03-31 -0.4640906 -0.2435414
  2014-04-30        NaN        NaN
  2014-05-31        NaN        NaN
  2014-06-30 -0.5626360  0.6679614
  2014-07-31        NaN        NaN

Which leaves me with two questions:

1. What is the proper way to resample a dataframe of mixed type?
2. When resampling from a lower frequency to a higher frequency what is the proper way to do the resampling so that the new values are interpolated?

Even if you can't provide a full answer to both parts a partial solution or an answer to either question is appreciated.

Answer 1

When resampling from a lower frequency to a higher frequency I realized that I was specifying the how when I wanted to specify the fill_method. When I do so things seem to work.

df4c = df3.groupby(['C']).resample('M',  fill_method='ffill')
df4c.head()
                     A          B        D
C                                          
A 2014-03-31 -0.2435414 -0.4640906  Pickles
  2014-04-30 -0.2435414 -0.4640906  Pickles
  2014-05-31 -0.2435414 -0.4640906  Pickles
  2014-06-30  0.6679614 -0.5626360  Pickles
  2014-07-31  0.6679614 -0.5626360  Pickles

You get a much more limited set of interpolation choices but it does handle the mixed types.

When resampling to a lower frequency using no how option (I believe it defaults to mean) the down-sampling does work:

   df5c =df3.groupby(['C']).resample('A')
   df5c.head()
                  A          B
C                                 
A 2014-12-31 -0.1065645 -0.7429617
  2015-12-31 -0.3101057 -0.6245030
B 2014-12-31 -0.0708263  0.4213621
  2015-12-31  0.0110456 -0.0607028

Therefore it seems the problem seems to be with passing a dictionary of how options or one of the option choices, presumably ffill, but I'm not sure.

Answer 2

Use resample and agg

Since pandas-1.0.0, the how and fill_method keywords no longer exist. Besides, the resample method now returns a Resampler object.

The solution is to define an aggregation rule using functions or function names associated to each column.

df.resample(period).agg(aggregation_rule)

More examples on aggregation rules in the documentation.

Working example

Prepare test data:

import numpy as np
import pandas as pd

dates = pd.date_range("2021-02-09", "2021-04-09", freq="1D")
df1 = pd.DataFrame(np.random.randn(dates.shape[0],2), index=dates, columns=list('AB'))
df1['C'] = 'A'
df1['D'] = 'Pickles'
df2 = pd.DataFrame(np.random.randn(dates.shape[0], 2), index=dates, columns=list('AB'))
df2['C'] = 'B'
df2['D'] = 'Ham'
df3 = pd.concat([df1, df2], axis=0)
print(df3)

Output:

                   A         B  C        D
2021-02-09  2.591285  2.455686  A  Pickles
2021-02-10  0.753461 -0.072643  A  Pickles
2021-02-11 -0.351667 -0.025511  A  Pickles
2021-02-12 -0.896730  0.004512  A  Pickles
2021-02-13 -0.493139 -0.770514  A  Pickles
...              ...       ... ..      ...
2021-04-05  1.615935  1.152517  B      Ham
2021-04-06 -0.067654 -0.858186  B      Ham
2021-04-07  0.085587 -0.848542  B      Ham
2021-04-08 -0.371983  0.088441  B      Ham
2021-04-09  0.681501  0.235328  B      Ham

[120 rows x 4 columns]

Resample per month:

agg_rules = { "A": "mean", "B": "sum", "C": "first", "D": "last",}
df4 = df3.resample("M").agg(agg_rules)
print(df4)

Output:

                   A         B  C    D
2021-02-28  0.025987  3.886781  A  Ham
2021-03-31  0.081423 -5.492928  A  Ham
2021-04-30  0.239309 -3.344334  A  Ham