Use the Nth underscore from the right as the separator to split a column into two columns

Suppose I have a dataframe df as follows:

    id          value_type_and_model_name
0    1                       actual_value
1    2             fitted_value_RUT_ARIMA
2    3       fitted_lower_value_RUT_ARIMA
3    4       fitted_upper_value_RUT_ARIMA
4    5          predicted_value_RUT_ARIMA
5    6    predicted_lower_value_RUT_ARIMA
6    7    predicted_upper_value_RUT_ARIMA
7    8                fitted_value_RUT_ES
8    9          fitted_lower_value_RUT_ES
9   10          fitted_upper_value_RUT_ES
10  11             predicted_value_RUT_ES
11  12       predicted_lower_value_RUT_ES
12  13       predicted_upper_value_RUT_ES
13  14           fitted_value_RUT_SARIMAX
14  15     fitted_lower_value_RUT_SARIMAX
15  16     fitted_upper_value_RUT_SARIMAX
16  17        predicted_value_RUT_SARIMAX
17  18  predicted_lower_value_RUT_SARIMAX
18  19  predicted_upper_value_RUT_SARIMAX

I want to split this column into two columns (except 'actual_value') using the second underscore from the right as the delimiter.
The expected results are as follows:

    id             value_type   model_name
0    1           actual_value          NaN
1    2           fitted_value    RUT_ARIMA
2    3     fitted_lower_value    RUT_ARIMA
3    4     fitted_upper_value    RUT_ARIMA
4    5        predicted_value    RUT_ARIMA
5    6  predicted_lower_value    RUT_ARIMA
6    7  predicted_upper_value    RUT_ARIMA
7    8           fitted_value        UT_ES
8    9     fitted_lower_value        UT_ES
9   10     fitted_upper_value        UT_ES
10  11        predicted_value        UT_ES
11  12  predicted_lower_value        UT_ES
12  13  predicted_upper_value        UT_ES
13  14           fitted_value  RUT_SARIMAX
14  15     fitted_lower_value  RUT_SARIMAX
15  16     fitted_upper_value  RUT_SARIMAX
16  17        predicted_value  RUT_SARIMAX
17  18  predicted_lower_value  RUT_SARIMAX
18  19  predicted_upper_value  RUT_SARIMAX

How to achieve this? Thanks. I try with code: df['value_type_and_model_name'].str.rsplit('_', n=2, expand=True), but it’s not working out.

>Solution :

Since you "only" have one underscore in the second chunk, the easiest is to craft a regex for that specific case:

out = (df['value_type_and_model_name']
       .str.extract(r'(?P<value_type>.*)_(?P<model_name>[^_]*_[^_]*)$')
       .fillna({'value_type': df['value_type_and_model_name']})
      )

regex demo (extract)

Output:

               value_type   model_name
0            actual_value          NaN
1            fitted_value    RUT_ARIMA
2      fitted_lower_value    RUT_ARIMA
3      fitted_upper_value    RUT_ARIMA
4         predicted_value    RUT_ARIMA
5   predicted_lower_value    RUT_ARIMA
6   predicted_upper_value    RUT_ARIMA
7            fitted_value       RUT_ES
8      fitted_lower_value       RUT_ES
9      fitted_upper_value       RUT_ES
10        predicted_value       RUT_ES
11  predicted_lower_value       RUT_ES
12  predicted_upper_value       RUT_ES
13           fitted_value  RUT_SARIMAX
14     fitted_lower_value  RUT_SARIMAX
15     fitted_upper_value  RUT_SARIMAX
16        predicted_value  RUT_SARIMAX
17  predicted_lower_value  RUT_SARIMAX
18  predicted_upper_value  RUT_SARIMAX

If you really want to split:

df['value_type_and_model_name'].str.split('_(?=[^_]*_[^_]*$)', expand=True)

regex demo (split)

If you want, you can assign to the original dataframe:

s = df.pop('value_type_and_model_name')
df[['value_type', 'model_name']] = (s.str.extract(r'(.*)_([^_]*_[^_]*)$')
                                     .fillna({0: s})
                                   )

Output:

    id             value_type        model
0    1           actual_value          NaN
1    2           fitted_value    RUT_ARIMA
2    3     fitted_lower_value    RUT_ARIMA
3    4     fitted_upper_value    RUT_ARIMA
4    5        predicted_value    RUT_ARIMA
5    6  predicted_lower_value    RUT_ARIMA
6    7  predicted_upper_value    RUT_ARIMA
7    8           fitted_value       RUT_ES
8    9     fitted_lower_value       RUT_ES
9   10     fitted_upper_value       RUT_ES
10  11        predicted_value       RUT_ES
11  12  predicted_lower_value       RUT_ES
12  13  predicted_upper_value       RUT_ES
13  14           fitted_value  RUT_SARIMAX
14  15     fitted_lower_value  RUT_SARIMAX
15  16     fitted_upper_value  RUT_SARIMAX
16  17        predicted_value  RUT_SARIMAX
17  18  predicted_lower_value  RUT_SARIMAX
18  19  predicted_upper_value  RUT_SARIMAX