Unexpected behavior when using the n_periods and drop_high/drop_low parameters at the same time

[jbogaardt]

Discussed in #292

^{Originally posted by cdietrich215 April 11, 2022}
First off, thanks a ton for all of the hard work on this package. Its been a huge help for me and my company.

When using the cl.Development() method with n_periods, I would expect the drop_high/drop_low parameters to drop the highest/lowest parameters after the earlier periods have been dropped from each development period. Instead it appears that the highest and lowest periods are dropped first, and then the earlier diagonals get dropped. This can frequently cause cases where only the low (or high) ldf gets dropped, skewing results.

Is there a way you'd recommend accomplishing this in the package's current state, or is this functionality yet to be built in?

For an example try running: cl.Development(drop_high = 1, drop_low=1, preserve = 1, n_periods = 5).fit_transform(cl.load_sample('raa')).age_to_age.heatmap()

[kennethshsu]

Let me try to fix this one.