AtiyahElsheikh
commented
1 year ago Summary Report II (13. March 2023)
This report corresponds to the new releases
| Package | Release |
|---|---|
| SocioEconmics.jl | milestone V0.4 |
| LPM.jl | V0.4 (14th Mar.) |
| MultiAgents.jl | (V0.5 5th Mar.) |
| MiniDemographABM.jl | (V1.0 End Feb.) |
I. Integration of Agents.jl-extended space type
Agents.jl is the most popular Julia library used within the Julia community with an extensive well-written documentation . Its integration allows employing so many ready-to-use code s.a. results collections and visualisation among others.
In order to integrate Agents.jl, the space concept of Agents.jl has to be adopted. This has been accomplished as follows:
- Inspired by MiniDemographicABM.jl (see below), a new Agents.jl-conform type
DemographicSpacehas been implemented. This type extends Agents.jl built-in space types - A new release of MultiAgents.jl V0.5 has been extended with additional capabilities imported from Agents.jl .
- The previous step represents a starting phase for merging Agents.jl into MultiAgents.jl so that many capabilities of Agents.jl will be directly usable. MultiAgents will then rather specialises at functionalities needed for large-scale socio-economic multi-agents based models. More on that can be consulted from issues of MultiAgents.jl
cf. issue #52 for further technical details
II. MiniDemographicABM.jl package
In order to learn how to integrate Agents.jl concepts into SEs library, a toy model MiniDemographicABM.j has been established to explore Agents.jl utilisation and facilities.
- about ca. 1600 LOC where as SocioEconomics library ~4300 LOC
- The library has been implemented in an incremental way with unit tests being implemented in parallel with the functionalities. The main model was constructed at last. This way ensures robustness and reduces time for bugs corrections and code development and modifications. There are about +70 unit tests as well as others for examining runtime and memory performance. This is helpful to detect and reside bottlenecks and instabilities at an early stage of the development
- Various clock rates (monthly, daily, hourly) exist.
- It is aimed to make MiniDemographicABM.jl import from SocioEconomics.jl one day if level of details concepts is going to be realized, cf. #32. In that case, the corresponding LOC are expected to decrease
- The MiniDemographicABM.jl package is helpful in learning
- how to establish a new space type adequate for SocioEconomics library. Namely, houses and towns are no longer agents as before, but rather they constitute a space on which agents (i.e. people) are operating,
- how to exploit useful ideas and code to be adopted in SE and MulitAgents libraries, cf. Issue #58 and issue number 26 of MA.jl
- The MiniDemographicABM.jl package is helpful in learning
Future potentials and further details can be consulted from the library issues
III. Fixing significant set of bugs (initial population level)
Many of the functionalities in the SocioEconomics library are not sufficiently tested for correctness. Consequently, constructed models were expected to have inconsistencies and semantics errors (i.e. errors that are not raised and hence not detected at runtime).
Issue #39 recommends a method to improve the quality of the library and hence the constructed models. This is done by challenging constructed populations and verify their consistency, for example kinship relationship and housing consistency. For example, if person x is the father of person y, then y has to be the child of person x.
This method has been now started in this release and a significant set of bugs have been identified. It has been found that:
- initial population contains homeless
- children without mothers
- a father might be not older enough than his child
- a house with only a child / children as occupants
- children are not in the same household as of their parents
All of the above bugs have been corrected, cf. #59. Some consistency checks did not result in any errors, cf. #39.
Therefore, it makes sense to further apply this method for verifying the consistency of the population as the simulation advances. This is hopefully to be accomplished in milestone V0.5)
IV. Blue style coding convention
There have been difficulities in previous version of understanding the code, what an identifier refers to and from where it comes from etc. In order to reduce such difficulities, the code has been transformed / renamed to follow coding guidelines that are widely approved by the Julia community.
At the top of the ReadMe in the library website, a blue-grey badge can be noticed. This is to refer that the library is now (largely) following blue-style coding conventions. These conventions prescribe the coding guidelines that need to be obeyed to improve code clarity and readability, not only by others but also by the developer who conducted the code. These conventions are invented from a large community and it is useful to stick to these conventions , e.g.
- it provides guidelines how future naming conventions should obey
- when using code from other blue-styled packages the same coding conventions are followed in own code
- a direct result from the transformation, identifiers and what they refer to can better recognised (e.g. internal/global functions, types, variables, global variables, etc.
- For example (no
foo_x()&fooX()under the same portion of source file or_foo()refers to an internal function that is not accessed outside the source file etc. - ...
V. Model-based function operation
As a result of starting constructing a model based on Agents.jl library, i.e. enabling many ways to construct and simulate the same or a similar model, it is further recognised that so many functions that are applied on model , i.e.foo(model) are splintered around the code. It is desired to collect such functions in one place to enhance code reuse within the library and make them accessible to the client. A new moduleAPI.ModelOphas been started and to be enhanced along next releases. More details, cf. #50 (Milestone V0.6)
VI. Enabling very large-scale simulations
It is now possible to simulate models with large initial population size , 100,000 (ca. 3 minutes ) - 1,000,000 (ca. 5 hours) cf. Performance Table. Potentials for further up-scaling of memory and runtime performance exist, cf. #31
Summary Report I (11th Jan. 2023)
The unified simulation interface is done ( Issue 28 & issue 29). Consequently,
select_people(*), selectedfor(*), verbose(), init_return, progress_return, xParameters(model)Before starting this library, execution of the simulation program of the LoneParents Model package demanded a huge memory allocations and consumption. This indicated that the julia code had serious problems that one should have paid attention at. This was not unfortunately the case. Significant Performance improvement (Since V0) have been achieved partially via the tasks highlighted in this (Issue 31). In numbers, the following has been achieved:
Worth to mention is that
The most recently implemented simulation functions were implemented in a careless way. The more checking code I place, I notice inconsistencies (issue 39 and others with the bug badge). But I believe that this will be subject to improvement through the next phases anyway.