dep-impact is a Python module for uncertainty quantification mainly based on studying the influence of dependencies between random variables of a probabilistic model through a deterministic function.
For the moment, the class ConservativeEstimate creates a probabilistic model with an incomplete description of the dependencies between the input variables. The class can therefore estimates, through a Monte-Carlo sampling, a quantity of interest of a model output distribution. It can also give a conservative estimation of the output quantity by determining a dependence structure that minimize the quantity.
An iterative algorithm is also available.
The package is still in development and is not yet available on PyPi or Anaconda.
Unfortunately, the package needs many python library dependencies:
Also, the software R is needed with the package:
However, we are still working on diminishing the number of dependencies. Especially the dependencies with R, by replacing the VineCopula R package with vinecopulib.
The installation is very straightforward using Anaconda. If you don't have Anaconda, you can find the software here and follow the instruction given on the website. The library is tested on python 2.7 and 3.5.
Install the classical dependencies:
conda install numpy scipy pandas scikit-learn matplotlib
Install OpenTURNS, which is in the conda-forge:
conda install -c conda-forge openturns
Install the other dependencies which are not in Anaconda:
pip install pyDOE scikit-optimize
If you don't already have R, you can install it using conda:
conda install -c R R
You also need rpy2 to communicate between Python and R:
conda install rpy2
Then you can install VineCopula using R
R -e 'install.packages("VineCopula", repos="https://cloud.r-project.org")'
You need a recent version of gcc to install the package. If you don't have a recent one, install it with conda install gcc.
Several notebook examples are available in the directory examples.