Funz

TL;DR

Parametric scientific computing ?

Take a "parametrized" input file (contains $x, $y,... replacing some numerical values),

Define values taken by these parameter (or let a dedicated algorithm do),

Run calculations on remote ressources,

Parse results & analyse.

Funz eases this "functional" wrapping of your simulation.
Basic demo in Python: , or R:

Overview

Wrapping

A standardized execution script (properly exiting when calcualtion failed),
A set of reserved characters (like '$') for input files parametrization,
Some regular expression to extract output values of interest.

Features

Parse & compile input files with given parametric syntax,
Launch remote execution of each input combination given (async on dedicated TCP channel),
Call external algorithm to drive calculation parameters values,
Monitor & restart failed calculations (blacklisting unsuitable computers),
Store & parse output results.

Extensibility

Wrapper for any simulation software (holds I/O files, ASCII & common formats: HDF5 NetCDF XML JSON),
Algorithm for design of experiments (supports definition of input values regarding engineering objective).

Examples ?

Funz makes your simulations callable as functions for any of cmd.exe / bash / python / R / Java:

From Excel spreadsheet "sheet.xlsx", get cell commented with "=total" for each 10 different values of cell commented with "$income":
```
Funz.bat Run -m Excel -if sheet.xlsx -iv income=10,20,30,35,40,42,45,47,50,52 -oe total
```

Search convection coefficient which leads to a minimal temperature of 25.5 degrees with OpenModelica PDE solver:

Funz.sh RunDesign -m OpenModelica -if NewtonCooling.mo -iv convection=[0.5,1.0] -oe "min(T)" -d Brent -do ytarget=25.5

Build a response surface of MySimulator for x1 in [-5,5] and x2 in [-5,5]:

# pip install Funz
import Funz
X = {'x1':np.arange(-5, 5, 0.25),'x2':np.arange(-5, 5, 0.25)}
y = Funz.Run(model="MySimulator",input_files=["inputfile1","inputfile2"],input_variables= X ,all_combinations=True)['y']
# mplt.plot_surface(X['x1'], X['x2'], y)

Propagate x1 as uniform random and x2 as normal random uncertainties through MySimulator:

# devtools::install_github("Funz/Funz.R")
library(Funz)
hist( Funz::Run(model="MySimulator",input.files=c("inputfile1","inputfile2"),input.variables= list(x1=runif(100),x2=rnorm(100)) )['y'] )

Chain simulations:

# pip install Funz
import Funz
X = {'x1':..., 'x2':...}
# first simulator, inputfile1 and inputfile2 contain 'x1' and 'x2' variables
y1 = Funz.Run(model="MySimulator1",input_files=["inputfile1","inputfile2"],input_variables= X)['y']
# second simulator, depends on results of first one also, inputfile3 contains 'x1', 'x2' and 'y1' variables
y2 = Funz.Run(model="MySimulator2",input_files=["inputfile3"],input_variables= X.update(y1))['z']
# post-processing in Excel (postrpro.xlsx sheet contains cells commented by '$x1' or '$x2'), returns cell value commented as "=result"
result  = Funz.Run(model="Excel",input_files=["postpro.xlsx"],input_variables= X.update(y2))['result']

Identify the most influent input parameters
Max/minimize simulation output value
Build inverse model

Who we are ?

Early versions of Funz project were funded by IRSN (French institute for radiological protection and nuclear safety).
IRSN open-sourced Funz in 2019 to support safety and design engineering community.

Standardize your simulation as a function.