Sex splitting below root node

[garland-culbreth]

Does at_cascade have a configuration option, in its CSV interface or API, which allows a user to specify the node(s) at which the cascade should split on sex, i.e., splitting after a specified node rather than after the root node?

This was prompted by learning about some diseases for which data is collected almost exclusively for one sex, e.g., female, but for which there is a known mechanism connecting male and female disease rates. For such a disease, a model would likely be best run by using data from male and female to inform the cascade deeper than the root node, since with very limited data splitting sex at the root node could well result in one branch failing to fit any child nodes. I'm curious id at_cascade could handle extremely asymmetric data like this, maybe by fitting both-sex data down to a specified set of nodes, then splitting. Or alternately, maybe fit both-sex all the way down the cascade and use a sex covariate to adjust the predictions for female vs male. Or maybe another alternative.

Raising as an issue because I'm interested in how functionality for achieving this might be implemented as a new feature, if it isn't already built in, or how it might be achieved with the existing feature set. I'm also curious whether the problem could be re-framed and use existing features to achieve the effect, e.g., by using a customized node set for the cascade perhaps.

Tagging @ntemiq as on this discussion as well since we discussed it briefly and were both curious how this might be accomplished.

[bradbell]

The node_split table,
https://at-cascade.readthedocs.io/node_split_table.html
and the mulcov_freeze table
https://at-cascade.readthedocs.io/mulcov_freeze_table.html
are currently fixed in the csv.fit interface.

It should not be hard to add either of them to the csv.fit input files.