Optimize memory and performance impacts

[nikhilwoodruff]

This model wasn't written with memory usage at the forefront of its targets, but with the tax-benefit system now written with its complexity, a smaller performance and memory footprint would be hugely beneficial. I can think of one way: finding the subset of variables involved in the net_income computation tree.

[MaxGhenis]

As in, using the baseline values for all variables except those affected by the reform (directly or indirectly)? Could this go in core?

[nikhilwoodruff]

I think so, just investigating the tracer tool in Core which shows the breakdown of computation time for a full run:

Not sure exactly how to interpret this, because some of the variable lengths are due to the order of calculation rather than the cost. For example, because income tax is first calculated to get the Housing Benefit applicable income, its duration appears under housing benefit and then just needs to consult the cache when actually calculating income tax.

[MaxGhenis]

If this is still in progress, could we make the issue title more specific?