Multinomial vector of policies

[RobertClay]

in interacting with WS7 going to create an experimental intervention that combines living wage and child uplift together. Given a finite pot of money will assign p1% to uplift and p2% to each intervention respectively. should create a differentiable objective surface with SF12 nanmean as the loss function. Can be optimised fairly easily? Things like lagrangian multiplier solutions seem to fit this format. Need to ask ws7.

[ld-archer]

Like this idea. Something that would also be useful alongside this is a fully custom intervention. As in, we could specify multiple interventions in the same experiment perhaps through nested JSON branches in config files. Thinking something like specifying a subgroup, an effect, and a start date and end date (i.e. take single parent households, increase disposable income by £X, and run from 2020-2025). Think this would be good as we could chain up these branches, so we could do a similar thing to what is proposed above with the finite pot of money being split across multiple interventions, but with more flexibility. If we do this correctly also, we could specify a sum of money in the config file and do some clever things about who gets how much etc.

[ld-archer]

Bumping this as we have had further discussions about this. Would require the creation of a custom intervention module, which would take parameters from the config file and act on them. Might be something like a function that accepts the parameters listed above and makes the required change to hh_income. This could then run in a loop depending on the number of custom interventions read from the config file.