在向前填充（填充）值的同时对Pandas数据框进行重新采样

Question

I have a dataframe where a snippet looks like this

    Time                    Temperature
19  2019-01-01 11:48:51     23.798
20  2019-01-01 11:48:53     23.832
21  2019-01-01 11:48:54     NaN
22  2019-01-01 11:48:55     23.817
23  2019-01-01 11:48:56     NaN

I want to resample this to '2S' while making sure that the last measured value will replace any NaNs.

df.resample('2S', on='Time').mean().ffill()

A snippet of the result looks like this

                        Temperature
Time            
2019-01-01 11:48:52     23.832
2019-01-01 11:48:54     23.817
2019-01-01 11:48:56     23.809

Notice the value at timestamp t=54s. What I want is the temperature 23.832 from t=53s, since that is the last recorded value at this timestamp. Instead it fillings with the value from t=55s

Edit 1: After a reply, I tried the following:

df.ffill().resample('2S', on='Time').first()

But this gives the following result, where the new t=52s is equal to the old t=t=53s, which is not the behavior I am after...

                        Temperature
Time            
2019-01-01 11:48:50     23.798
2019-01-01 11:48:52     23.832
2019-01-01 11:48:54     23.832
2019-01-01 11:48:56     23.817

EDIT 2: To make it easier to understand, this is the output I desire. I don't care if it is sampled on odd or even seconds.

                        Temperature
Time            
2019-01-01 11:48:52     23.798
2019-01-01 11:48:54     23.832
2019-01-01 11:48:56     23.817

Answer 1

Edit #3:

idx = df.resample('2S').asfreq().index
df.reindex(df.index.union(idx)).ffill().resample('2S').asfreq()

Output:

                     Temperature
Time                            
2019-01-01 11:48:50          NaN
2019-01-01 11:48:52       23.798
2019-01-01 11:48:54       23.832
2019-01-01 11:48:56       23.817

---

Edit #2:

idx = df.resample('2S').asfreq().index
df.reindex(df.index.union(idx)).bfill().resample('2S').first()

Output:

                     Temperature
Time                            
2019-01-01 11:48:50       23.798
2019-01-01 11:48:52       23.832
2019-01-01 11:48:54       23.817
2019-01-01 11:48:56          NaN

---

EDIT:

df.reindex(df.index.union(df.resample('2S').asfreq().index))\
  .interpolate().resample('2S').asfreq()

Output:

                     Temperature
Time                            
2019-01-01 11:48:50          NaN
2019-01-01 11:48:52      23.8150
2019-01-01 11:48:54      23.8245
2019-01-01 11:48:56      23.8170

---

Do you want to resample two seconds on odd seconds or even seconds?

df.ffill().resample('2S', on='Time', base=1).mean()

Output:

                    Temperature
Time                            
2019-01-01 11:48:51       23.798
2019-01-01 11:48:53       23.832
2019-01-01 11:48:55       23.817

Or just on even number of seconds:

df.ffill().resample('2S', on='Time').mean()

Output:

                     Temperature
Time                            
2019-01-01 11:48:50      23.7980
2019-01-01 11:48:52      23.8320
2019-01-01 11:48:54      23.8245
2019-01-01 11:48:56      23.8170

Answer 2

EDITED to use last, not first. Probably doesn't matter with your sample data, but if you have multiple records in a 2 second period, this will ensure you take the most recent one.

There is an option when resampling to specify what edge of the bin to label the data with. The default for S is left - so the start of the 2 second time period. Changing to right I believe gives what you are after.

df.resample('2S', on='Time', label='right').last().ffill()

Time                Temperature
2019-01-01 11:48:52 23.798
2019-01-01 11:48:54 23.832
2019-01-01 11:48:56 23.817
2019-01-01 11:48:58 23.817