Python solution for using several machine learning methods to search a candidate solution space for the optimal population models in NONMEM.
Visit pyDarwin Documentation to learn more.
- Windows 10
- Windows Server 2018, 2019
- CentOS8/RHEL8
- Ubuntu >= 18.04
- Sun Grid Engine (SGE)
- Python >= 3.10
- NONMEM >= 7.4.3
- R >= 4.0.0 (optional)
Note: Requirements are Python and NONMEM installation with nmfe.bat available. R is required if using post-run R penalty function.
First, create a new virtual environment:
python -m venv .venv
This will create a virtual environment in the folder .venv
Next, use pip to install the pyDarwin package from the Certara managed PyPi repo:
pip install pyDarwin-Certara --index-url https://certara.jfrog.io/artifactory/api/pypi/certara-pypi-release-public/simple --extra-index-url https://pypi.python.org/simple/
pip install pyDarwin-Certara --index-url https://certara.jfrog.io/artifactory/api/pypi/certara-pypi-develop-local/simple --extra-index-url https://pypi.python.org/simple/
python -m darwin.run_search <template_path> <tokens_path> <options_path>
To execute, call the run_search function from the darwin module and provide the following file paths as arguments:
- Template file (e.g., template.txt) - basic shell for NONMEM control files
- Tokens file (e.g., tokens.json) - json file describing the dimensions of the search space and the options in each dimension
- Options file (e.g., options.json) - json file describing algorithm, run options, and post-run penalty code configurations.
After cloning https://github.com/certara/pyDarwin from GitHub, navigate to one of the example folders e.g.,
cd .\pyDarwin\examples\user\Example1
Then execute:
python -m darwin.run_search template.txt tokens.json options.json
Note: Both absolute and relative file paths are supported.