Running with Python¶

After implementing your MDP in C++, it s possible to interact with DynaPlex through Python. This way, you can run algorithms, collect statistics, and benchmark the built-in DynaPlex algorithm with existing algorithms via Python libraries, e.g., benchmark DCL with PPO. To do this, we will need to build PyBindings. Follow the below steps. We assume you already installed a suitable Python IDE, e.g., PyCharm or VSCode. Note that we provide several example Python scripts in the python folder which show you how to run DCL and PPO (using the tianshou library) from Python.

1. Virtual environment creation

First, in your Python IDE or in your terminal set up a Conda environment for compiling the Python bindings and running Python scripts that load those bindings. We provide an environment.yml file (in the DynaPlex/python folder) that enables the creation of a python/anaconda environment (envDP) that has all packages for compiling the dynaplex python binding:

conda env create -f path/to/environment.yml

module load 2022
module load  ...

2. Adapt your CMakeUserPresets.json

Adapt the paths linking to your Conda virtual environment. Do this for WinPB and/or LinPB, depending if your work on Windows or Linux.

3. Compile PyBindings

In your C++ IDE, compile the PyBindings. For instance, in Visual Studio, you can select WinPB from the drop-down build configurations.

4. Activate your virtual environment

In your Python IDE or terminal, activate your virtual environment

conda activate envDP

5. Install DynaPlex in the envDP environment

In your Python IDE or terminal, navigate to the DynaPlex/python folder. Next, run:

pip install -e .

After this step, you should be able to import DynaPlex as a library in Python:

from dp import dynaplex

On Linux (Snellius), you can simply call, which should be enough:

conda activate EnvDP