Skip to content

Latest commit

 

History

History
114 lines (91 loc) · 12.2 KB

CHANGELOG.md

File metadata and controls

114 lines (91 loc) · 12.2 KB
Raw

Optima Changelog

Projects before 2.11.4 should be able to be migrated to version 2.11.4 without changing the calibration. Versions 2.12.0 and 2.12.1 both change the calibration (sometimes only slightly).

Versions 2.11.4, 2.12.0 and later all can be run using the branch main. A project will automatically update to the earliest supported version (currently 2.11.4), but updating a project to the latest version can be done using the FE or op.migrate(P, 'latest')

Revision 6

  • Reset the fromdata attribute of the numcirc parameter (to 1.0) to allow it to be updated (older version migration did not ensure this)

Revision 5

  • Update random number generator seeding in runsim by seeding a list of default_rng objects that are passed through to parameters that need to be sampled within each individual makesimpars to both ensure consistency and to avoid all parameters being sampled based on the same seed (e.g. all low or all high)
  • When sampling, each parameter will use its own generator seeded by the global seed + a hash of the parameter short name - this means that changing which parameters are sampled will still be consistent for other parameters for the same seed, resulting in more consistent outputs

Revision 4

  • Update methods of transmission tracking (advancedtracking=True) so numinciallmethods is split by regular, casual, commercial acts. numincimethods remains the same.
    • This causes advancedtracking=True to run ~20-25% slower as there is now 8 methods of transmission instead of 4.
    • numincimethods now more accurately sums to numinci (tested to be less than 1 part in 1 million total error)
    • The intersection of multiple methods per transmission is estimated differently: instead of considering intersections as being a part of the higher probability method, all the intersections are now split according to the ratio of the method probabilities.
  • P.makespreadsheet() can now take different datastart, dataend than the current data to extend or shrink the databook years.
  • Updated defaults settings.now = 2023.0 and settings.dataend = 2040.0

Revision 3

Fix small unpickling bug and FE raises BadFileFormatError when uploading project that it cannot unpickle.

Revision 2

Fix bug when deep-copying a Parameterset, the pars from a different Parameterset would get copied in certain cases (make sure to pass memodict along when deep-copying).

Revision 1

Can run multiple different supported versions with the same branch! Currently supported: [2.11.4, 2.12.0]

Migration to a new version is done on load P = get_latest_project("example", migrateversion='supported') with options of 'supported' (default) which updates to the earliest supported version, 'minor' which updates to the latest within the same 2.X , 'major' = 'latest' = True which updates to the latest version, False which does not update (will give a big warning if the version that the project is on is not supported).

  • P.revision added and first migration added which:
    • Converts P.parsets, P.progsets, parset.pars, progset.programs into odict_customs which:
    • odict_custom is a new class that calls a custom function func every time a value(s) gets set in the odict
    • In the case of P.parsets, P.progsets, parset.pars, progset.programs, the odict_custom.func = checkpropagateversion which checks the projectversion of the object that gets added (that it is equal) and if it doesn't have a version, then it adds it. So Parameterset, Programset, Par, Program now all get .projectversion added
    • They also get a getprojectversion() that checks that the different projectversions match / gets the project version from the projectref if need be
    • The odict_custom.func is checkpropagateversion which checks that P.version is changed then that gets propagated automatically
  • If P.version is changed then that gets propagated automatically using __setattr__ in Project, Parameterset, Programset
  • Added projectversion or version as a argument to many functions, which defaults to finding it itself (eg in Par.interp, the Par has self.projectversion which should be set automatically) eg. model, makepars, makesimpars which are all version-dependent functions
  • Custom __copy__ and __deepcopy__ functions for objects that have odict_customs to set the odict_custom.func to the new object.checkpropagateversion (the odict_custom.func gets set to None when copying)
  • loadspreadsheet and makespreadsheet have version / projectversion as arguments to check compatibility and make / load spreadsheet in the version-correct behaviour
  • When loading a spreadsheet, the matrices get their dimensions checked properly
  • Speed up plotting by ~30-50% (only plot actual data points not nan and faster way of setting axis label fontsize)
  • runscenarios can be run in parallel (creates a copy of the project for each scenario with only the parset and progset it needs)
  • addblankdata in loadtools.py now adds the correctly shaped blank data to P.data for new parameters, fixing an error downloading spreadsheets of migrated projects

Revision 0 and earlier

  • runsim() now has a argument parallel to which, when True, runs the creation of simpars and the model in parallel.
  • MTCT is now attributed to PLHIV instead of susceptible states, but it is still an approximation which could be improved.
  • Emigration is split from background death in the results, meaning new results results.other['numemi'] and results.other['numemiplhiv'] to match immigration.
  • __init__.py is now structured to import the submodules without underscores eg op.project used to be op._project since pickling was throwing an error. So don't confuse op.project (submodule) with op.Project (object).
  • Outcomes optimization now stop at after a maxtime properly.
  • Speed-ups to both running the model and optimizations.
  • Budget and coverage scenarios need to have at least 1 year for the programs to start applying from - otherwise an error is thrown.
  • Update numcirc to default to "general" not "percentage" format in databooks (and to load correctly)

FE changes

  • An error will show if making a program with the same long or short name as another one.
  • The Optimization constraints are now proportional to the default budget (latest) for that progset (proporigconstraints). This means a 100% min constraint actually means the min budget is the latest budget - as opposed to having to scale it up or down depending on the total budget. If an Optim also has contraints and absconstraints those will also be followed and reflected in the constraints shown. But if you change the optimization then the constraints and absconstraints will be removed.

[2.12.3] -

ALL PLANNED / TODO:

  • !! TB/HIV co-infection mortality
  • Change acts to be better balanced
  • Change acts to either be 0%, 50% or 100% insertive or receptive for a population, not based on the numbers in the Partnerships tab - these numbers are used to calibrate the split of a population's acts with the other populations, not the relative insertive/receptive split.
  • Multiresultset.parsetname and Multiresultset.progsetname is now an odict, with a key and value for each result.
  • Change forcepopsize to not affect the number of PLHIV. The previous assumption was to remove (or add) people from (or into) the susceptible and the "not on ART" states. Now people only are removed from (or added into) the susceptible states.

[2.12.2] - 2024-07-17

  • Clean up MTCT code and fix problems:
    • MTCT of people on ART was being double-counted
    • Changed who PMTCT goes to based on their diagnosis / treatment state:
      • Previously went randomly to anyone diagnosed, including people diagnosed but not in care. This meant some pregnant people on ART were not getting PMTCT but they were getting the low probability of transmission from being on ART.
      • eg: 100 preg people on ART, 100 preg people who are dx but not on ART, data PMTCT: 100 on PMTCT -> 50 dx (not on ART people) on PMTCT, 50 people on ART also on PMTCT, 50 people on ART but not PMTCT = 150 births at the lower probability of being on PMTCT
      • Now: first give PMTCT to anyone on ART, then if there is more PMTCT spots, put any diagnosed people not on ART onto PMTCT. Then if there is more PMTCT spots, try to diagnose people to put onto PMTCT (they can go onto ART next time step if there is spots).
      • eg: 100 preg people on ART, 100 preg people who are dx but not on ART, data PMTCT: 100 on PMTCT -> Only 100 on ART also PMTCT = 100 births on PMTCT
      • NOTE: this can mean that populations which have more people on ART get more PMTCT if there is only enough spots for the number of people on ART
  • Fix rare negative people issue when FOI is very high (when probability of infection for a population is >1) - make FOI = min(FOI, 1)

[2.12.1] - 2024-05-27

  • Fix numcirc being set to 0 in the Parameterset upon loading from data - meaning running with just a parset had no VMMC. Running scenarios or programs affecting numcirc (eg. with VMMC program) were still working, just not the calibration.
    • The migration reloads the numcirc in every parset from the P.data which is from the latest loaded spreadsheet (for the parsets that have numcirc.y = 0).

[2.12.0] - 2023-

New changes - most of these change the calibration (some only slightly):

M F MSM1 MSM2 TGW
M 1 1
F
MSM1 1 10 5 1
MSM2 1 1 1
TGW
  • Sexual partnerships are now properly direction - in the Partnerships tab insertive is on the left, receptive on top - example above. eg.
    • This means TGW receive the proper (higher) risk associated with only having receptive acts.
    • If there are multiple MSM populations which both have both receptive and insertive risk then there should be a number in both spots in the matrix. The insertive acts will be split proportionally to the numbers in each spot. eg. in this case MSM2 has 5 receptive acts with MSM1 for every 1 insertive act.
      • (Note this was changed in 2.12.1 to be always 50/50 insertive/receptive if there is a non-zero entry in each spot).
    • If one population is M and the other F, then the acts will always be insertive for the M population and receptive for the F, regardless of the direction in the databook.
    • Note that an old project will keep the old acts behaviour until it is updated from a databook on >2.12.0
    • Behind the scenes, the pars['actsreg'] etc, now only stores the insertive acts, and is "per insertive population" so the calculation of the corresponding receptive acts happens during creation of the simpars['actsreginsertive'] and simpars['actsregreceptive'] etc which the model uses. pars['actsreg'].insertiveonly = True etc is for the new behaviour, a migration means older projects are False until updated from a databook.
  • Condom use now properly stored only once per partnership in the parset, preferring (M,F) and sorting (M,M) partnerships alphabetically. This means programs only need to target condom use for (MSM1, MSM2) not also (MSM2, MSM1)
  • ANC testing for HIV+ mothers happens automatically based on the number of mothers on PMTCT - if there are not enough diagnosed HIV+ pregnant mothers, then a proportion of each population's undiagnosed HIV+ pregnant women will be diagnosed to fill the numpmtct (or based on proppmtct).
  • Only PLHIV who have been diagnosed get pregnant and give birth at the reduced rate relhivbirth, whereas previously it was for all PLHIV.
  • proppmtct is now the proportion of HIV+ pregnant women who are on PMTCT, whereas the model used to interpret it as the proportion of diagnosed HIV+ mothers who are on PMTCT.
  • Fixed bug where if getting population size for only one timestep, then the exponential curve would be inappropriately applied. Most noticeable in the Cost-Coverage plots in the FE.
  • Other small bug fixes in the model while also speeding it up (like popsize for a single timestep - which affected Cost-Cov graphs, and a small number of people ageing two steps at once)