Consider school openings (in low density areas)

[loladekhuijzen]

For example, by adding an add heuristic after the drop heuristic.

[loladekhuijzen]

Can we rank by potential travel cost reduction and filter out schools in dense areas to prune the search space?

[jipclaassens]

search space verkleinen:

• per nuts3
• max potentiele bereikbaarheid nemen Deel grid van studentenbereikbaarheid met Chris #40

Open/sluiten van scholen geeft frictiekosten (parameter maken), bv. 80% van de overprovisie kosten van twee bijna lege scholen.

[cjacobscrisioni]

I have made density plots of potential access to primary school children and the presence of primary schools in Romania:

Access to primary school students in the x-axis, density of occurence on the y-axis. Red line is for grid cells with no schools, blue line is for grid cells with at least one school. The gap between exp(5) and exp(7.5) is intrigueing, seems as if Romania hardly has grid cells with intermediate accessibility levels?

Thinking on where to put boundaries on our search space... First guess, I would restrict the search space so that it only includes locations with access > exp(1) and access < exp(10). Access below exp(2) is a very unlikely location, access above exp(10) is plausibly indicative of high demand densities where we do not need to shift schools around.

For Romania this would lead to ~77,523 extra locations, which, granted is still massive. Together with existing school locations that makes ~85k nodes. @loladekhuijzen , would a search space with ~85k potential locations work, or do we need to be more restrictive? We can maybe go to < exp(9) instead of < exp(10) to weed out more suburban areas.

Results of the proposed access-based selection criterion:

Feedback warmly welcomed :)

[cjacobscrisioni]

I have added a criterion (census pop 2011 > 0) and reduced the search space to ~37,000 cells in RO (~3.9x the number of school locations). See latest push. @loladekhuijzen , @MaartenHilferink , is that a sufficiently constrained search space?