Digital communications decision module #82
Conversation
…nformation. Now blocked by smif#333
|
After discussion with Ed:
|
willu47
changed the title
|
Getting more believable results now:
|
|
@tomalrussell - this is ready for re-review. Happier with rollout now (just checking with pre-specified planning though). Will chat through with @edwardoughton at 3pm. |
|
Current issue with three pre-specified planning interventions in 2019. Unexpected regressions/downgrades added in later years on the same exchanges.
|
|
Intervention regressions above are caused because multiple technologies are not supported by the digital comms model yet. This will be resolved by updated the cost and benefits calculations within the digital comms model code to take account of multiple technologies. |
|
Requires nismod/digital_comms#87 |
|
Now blocked by merging into master of nismod/smif develop branch |
There was a problem hiding this comment.
Runs okay 👍 and principles look good.
Still gives unexpected results with FTTDP before FFTC in one exchange area:
2021,exchange_STBNMTH_fttdp
2023,exchange_STBNMTH_fttc
$ cat results/digital_comms_test/state_2030_decision_13.csv | sort -n
build_year,name
2019,exchange_NEILB_fttc
2020,exchange_NEILB_fttdp
2021,exchange_STBNMTH_fttdp
2021,exchange_WEWPRI_fttc
2022,exchange_EACAM_fttc
2023,exchange_EACAM_fttdp
2023,exchange_EACOM_fttc
2023,exchange_MYCHA_fttc
2023,exchange_STBNMTH_fttc
2023,exchange_WEWPRI_fttdp
2024,exchange_EACOM_fttdp
2024,exchange_MYCHA_fttdp
2027,exchange_STBNMTH_fttp
2029,exchange_EACAM_fttp
Overview
In this PR, I implement a decision module which extracts the decision logic from the
digital_comms/interventionsmodule.Further progress is currently blocked by nismod/smif#333 and nismod/smif#328
How it works
timestep = 2019iteration = 1Decision metrics (
'rollout_costs', 'rollout_bcr', 'total_potential_benefit', 'total_potential_bcr'), which are attributes on the Exchange objects from the digital communications model are computed and saved to smif.iteration += 1The decision module reads these in for the previous iteration, and then ranks interventions by descending
rollout_bcr(and ascendingrollout_costs) and chooses the subset of interventions that fit within a currently hardcodedannual_budget.These interventions are returned and written to a state file which looks like this:
state_2019_decision_2.csvThe state file is subsequently read in by the digital comms model wrapper simulate method and used to upgrade the digital comms network.
Updated results are produced and the cycle continues until the decision module is satisfied (at present, it chooses as many interventions as can be squeezed into the budget and is automatically satisfied to progress to the next timestep).
if satisfied == True: timestep += 1; iteration +=1; else: iteration +=1Outstanding issues and questions
@edwardoughton - can you answer any of these?
rollout_bcrresults are zero in each iteration, while rollout costs do update.update_adoption_desirabilitymethod effects the system - it seems to cause each of the nested assets to update and recompute their calculated attributes, but it is unclear when this should be triggered