henrydingliu
henrydingliu
confirmed this also affects Benktander. Using the example from the tutorial, set n_iters = 0. the next diagonal goes weird.
quick clarification, i'm doing .full_triangle_.cum_to_incr(). The next _incremental_ diagonal is acting weird.
allocating IBNR using conditional dev distribution will 1. coincidence with standard LDF based approach on the standard chainladder; 2. smooth out the bottom triangle for any special cases of BK....
quick example of conditional dev distribution. say selected CDF is 3, 2, 1,6, 1.4, 1.3 etc. the uncondtitional dev distribution is .33, .17, .125, .089, etc. conditional dev distribution at...
In the code, drop and drop_high/drop_low are applied in parallel. (interestingly enough, drop_above/drop_below are applied after.) Not a difficult thing to put them in any order, but I wonder if...
pull request #370 created with fix to support 4D drop hi/lo. @jbogaardt I did some quick tests. Also ran your assertions from the top (pass on both). seems to work....
pull request #375 created with fix to support 4D drop above/below this assertion fails without this fix ``` import chainladder as cl clrd = cl.load_sample('clrd').groupby('LOB')['CumPaidLoss'].sum() assert cl.Development(drop_above=1.3).fit(clrd).ldf_.loc['wkcomp'] == cl.Development(drop_above=1.3).fit(clrd.loc['wkcomp']).ldf_ ```
@jbogaardt thanks for reviewing and approving. do you want to talk through pipeline support architecture on this issue or open a new one?
with 4D support, do we need to modify the existing [weight_ property](https://github.com/casact/chainladder-python/blob/4766369bdd3d987619d63e4e425d3d4e480004ec/chainladder/development/development.py#L216) as well? also, should weight_ be in base instead of development?