Installation¶

DEEM is a small, dependency-light Python package. It requires:

Python 3.9 or newer
NumPy
SciPy (used for the Cauchy distribution, low-discrepancy sequences and the surrogate in DEEMI)

The benchmark examples additionally use opfunu to provide the CEC test functions.

Get the code¶

DEEM is distributed as a source package. Clone the repository and place its root directory on your PYTHONPATH:

git clone https://github.com/j-machacek/DEEM.git
cd DEEM

The package layout is:

DEEM/                  # repository root (put this on PYTHONPATH)
└─ DEEM/               # the importable package
   ├─ __init__.py
   ├─ DEEM.py          # base algorithm  -> class DEEM
   ├─ DEEMI.py         # improved variant -> class DEEM
   ├─ population.py
   ├─ evaluation.py
   ├─ boundary_conditions.py
   ├─ sampling.py
   ├─ toolbox.py
   └─ surrogate.py     # optional surrogate model used by DEEMI

Because both algorithms live in the DEEM package, the base optimiser is imported as from DEEM.DEEM import DEEM and the improved variant as from DEEM.DEEMI import DEEM.

Install the dependencies¶

pipconda

python -m venv .venv
source .venv/bin/activate          # Windows: .venv\Scripts\activate
pip install numpy scipy
pip install opfunu                 # only needed for the benchmark examples

conda create -n deem python=3.12 numpy scipy
conda activate deem
pip install opfunu                 # only needed for the benchmark examples

Verify the installation¶

From the repository root:

from DEEM.DEEM import DEEM
import numpy as np

opt = DEEM(lambda x: float(np.sum(np.square(x))),
           lower_bound=np.array([-1.0, -1.0]),
           upper_bound=np.array([ 1.0,  1.0]),
           nparticles_max=20, nparticles_min=20,
           nswarm_max=4, nswarm_min=4, maxiter=50)
opt.update()
print("best:", opt.FBEST)

If this prints a small number close to zero, you are ready to go. Continue with the Quickstart.