Variable grouping of results

[jacobvjk]

depends on RMI-PACTA/pacta.multi.loanbook.analysis#34
depends on RMI-PACTA/pacta.multi.loanbook.plot#30

• updates calls of pacta.multi.loanbook.* functions to adjust to the new variable interface
• gains a parameter BY_GROUP that can be used to generate aggregate metrics by any user defined variable, provided the variable exists in the matched_prioritized data set
• in the given example, the aggregate alignment metric is calculated for:
  • an aggregate loan book across the six fake input loan books, giving a meta loan book view
  • grouped by loan book, as indicated by "group_id"
  • grouped by another random user-provided variable above the loan book level, as indicated by "foo"
  • grouped by a combination of "group_id" and "foo", giving a sub loan book view
• at this point, calculating the aggregations at the level of multiple combined dimensions is only done for the calculation of results. Plots are currently limited to one level of aggregation, because a combination across variables is not in all cases straight-forward

NOTE:

• calculation of the aggregate metric for corporate benchmarks is temporarily suspended, as it turns out too resource intense to calculate these on standard machines and we have not seen heavy use of the benchmark in the aggregate metric at this point

EXAMPLE OUTPUTS with new grouping functionality

Sankey plot at the aggregate level:

Sankey plot calculated based on the foo split:

Sankey plot calculated based on the group_id split:

Scatter plot alignment by exposure based on foo split:

Scatter plot alignment by exposure based on group_id split:

[jdhoffa]

Consider renaming by_groups to by_group otherwise LGTM (i think, it's a pretty mammoth PR so I may have missed something)

[MonikaFu]

I run it with test data provided and using the two versions of the supporting packages mentioned. It seems to work correctly in general. For some plots the axis labels overlap with the title but since this is only a demo I am not sure if it is worth it to spend time on it. I guess you would need to play around with figure size when saving. Also - the sankey plot per company is rather unreadable which is to be expected with a big number of companies.

[jacobvjk]

I run it with test data provided and using the two versions of the supporting packages mentioned. It seems to work correctly in general. For some plots the axis labels overlap with the title but since this is only a demo I am not sure if it is worth it to spend time on it. I guess you would need to play around with figure size when saving. Also - the sankey plot per company is rather unreadable which is to be expected with a big number of companies.

I agree, especially re sankey plot with companies. In the end we need to decide what we want to maintain there. At the same time, we could just as well show examples using other variables with less categories. In any case, this seems like a topic for discussing the standardized P4S offering