Requirements

In order to run RiskFlow, an nvidia video card with compute capability 6.1 or above (and corresponding drivers) is strongly recommended. Most of the modules listed here can simply be installed via pip but may require a full build environment (with a C/C++ compiler). The Anaconda python distribution already comes with many of the below packages already installed. The required packages are:

Python==3.6

  • Only python 3.6 and above is supported - this is mainly due to all dictionaries now being ordered

Numpy>=1.15

  • Earlier versions could work but have not been tested.

Scipy>=1.1

  • Currently only required for interpolation and numerical integration. Again, earlier versions could work but have not been tested.

Pandas>=0.22

  • Earlier versions could work but have not been tested.

tensorflow=1.14

  • This is the computational library that evaluates tensors either on CPU or GPU. Only tensorflow 1.14 has been tested and tensorflow 2.x is not yet supported.

pyparsing

  • Required for simple parsing of time grids.

Optional requirements

NVIDIA CUDA drivers and SDK

  • Needed for GPU code execution. This is needed by Tensorflow if GPU computation is required

Matplotlib>=2.0.0

  • Needed for generating plots of risk factors and simualted paths

mkdocs>=0.16

  • Needed for building this documentation. Note that the math formatting is done via python-markdown-math and also needs to be installed.