SvF

SvF stands for "Simplicity vs Fitting" and is a short title of a method of balanced regularized identification of mathematical models by experimental data.

The SvF-technology proposes a promising area of applied mathematics, combining methods of structural mathematical modeling, cross-validation by subsets of experimental data, adaptive Tykhonov regularization, optimization and distributed computing.

The author of the SvF-technology is Alexander Sokolov, sashasok.

The technology and underlying mathematical algorithms are described in the following articles:

1. Sokolov A. V., Voloshinov V. V. Model Selection by Balanced Identification: the Interplay of Optimization and Distributed Computing // Open Computer Science, 2020, 10 — p. 283–295. DOI: 10.1515/comp-2020-0116
2. Соколов, А.В.; Волошинов, В.В. Выбор математической модели: баланс между сложностью и близостью к измерениям. International Journal of Open Information Technologies, 2018, 6(9) C. 33-41, PDF

How to cite

Please, cite the first of the above articles if you'll use the technology.

References

This mathematical technology has been successfully used in various fields of applied research. Here you can find a list of references to publications concerning application of SvF-technology.

How to install

Current implementation of the SvF-technolosy is based on Everest Python API and SSOP Everest Application, which have to be cloned from their Git-repos. So, use the following command for correct cloning

$ git clone --recurse-submodules https://github.com/distcomp/SvF.git

or, if you have public key attached to your GITHUB account, then you can use another command:

$ git clone --recurse-submodules git@github.com:distcomp/SvF.git

!!!===================== Recommendation ==============================!!!
We recommend to update all submodules ... Just in case...
Open SvF folder in your system console and run
$ git submodule update --recursive --remote --merge
=====================================================================

If you know Russian read the Section 1 (software requirements) of User Manual :

1. OS Linux is the default recommendation
2. Python 3.7.4+ (to save disk space, it is recommended to use the Miniconda Python environment
3. Basic Python packages are (depending on the actual set of packages of your Python environment some other packages may be missed and should be installed according to ModuleNotFoundError messages)
   • numpy 1.6.+
   • matplotlib 1.5.+
   • Pyomo 6.+, Installation
4. For solving NLP problems (Nonlinear Mathematical Programming Problems with continuous variables and differentiable functions) you need Ipopt solver.
   • For regular installation see native Ipopt documentation.
   • Full functional build with additional Linear Algebra libraries may be found here https://gitlab.com/ssmir/solver-build-scripts (contact with this installation pack developers for disclosure of unclear details)
   • For demonstrative or testing purposes "light" Ipopt build may be istalled as Python package:
     $ conda install -c conda-forge ipopt
     see Pyomo documentation
5. For large-scale calculations, it is desirable to use the Everest platform, in particular, the SSOP application, which allows you to solve in parallel a set of optimization problems on computing resources connected to the Everest Optimization Portal Everest Opt. To do this, you will need to register on the site https://everest.distcomp.org or https://optmod.distcomp.org.

Test run

1. Open Bash-script runSvF30.sh in any text editor and set correct value to the system environment variable SVFLIBPATH (a path to SvF/Lib30 folder at your system)
2. Open in your system console some of subfolders in Examples folder, e.g.
   SvF/Examples/3-ThermalConductivity/MSD(Dreg11x11)Curv(T)M0,
   and run the command
   $ bash ../../../runSvF30.sh

Try local or remote solver

All description of mathematical model, references to experimental data (e.g. in text or Spreadshet formats), settings and options of SvF-algorithm are presented in a special Task-file. You can find examples of these files in "terminal" subfolders of SvF/Examples folder. Possible extensions of task-files may be .mng (text format) or .odt (Libre/OpenOffice Writer). Among other options there is one, which tells SvF-system how to solve optimization problems arisen: "locally" (by solver installed at the system where Python SvF-application is running) or "remotely" (by Everest optimization service).

These options are: Runmode or RunSolver. Not going in deep details for beginners you may set these option either to
P&P - to solve all problems by local solver
or
S&S - to solve all problems by remote solvers.

E.g. in task-file MSD(Dreg11x11)+Curv(T):M=0.odt in example model
SvF/Examples/3-ThermalConductivity/MSD(Dreg11x11)Curv(T)M0
the Runmode option is in the first line. You can chage its value and run example by the command (from this folder)
$ bash ../../../runSvF30.sh

Try remote solver

To try remote solvers you can set the proper value of Runmode option (in *.odt task-file)
Runmode = 'S&S'

Then you must to get special token-file which is required to use Everest-services by Everest Python API.

To get the token you must be a registered user of sites https://everest.distcomp.org or https://optmod.distcomp.org.
If so, you can get a standard (7 days valid token) by the following command (you will be asked to enter your Everest password):

$ python everest.py get-token -server_uri https://optmod.distcomp.org -u YOUR_EVEREST_LOGIN -l ssop | tee .token

Run that command in SvF/pyomo-everest/python-api folder and .token file will appear.

After that switch to SvF/Examples/3-ThermalConductivity/MSD... folder and try to run SvF-application
$ bash ../../../runSvF30.sh