Installation¶

esnfed runs on Python 3.9+. The core has only two dependencies (NumPy and NetworkX); everything heavier is an optional extra.

pip install esnfed                  # core (numpy + networkx)

Optional extras¶

Extra	Adds	For
`fast`	numba, scipy	optional acceleration (Numba JIT, sparse reservoirs)
`viz`	plotly, matplotlib, seaborn, kaleido	the `esnfed.viz` plots
`experiments`	matplotlib, pandas, scipy, scikit-learn	reproducing the experiments
`reservoirpy`	reservoirpy	designing reservoirs in ReservoirPy
`flower`	flwr	the Flower integration example
`llm`	transformers	the experimental FedResPrompt LLM backend
`dev`	pytest, build, twine	development

pip install "esnfed[fast]"                   # Numba + sparse acceleration
pip install "esnfed[viz]"                    # plotting
pip install "esnfed[viz,experiments]"        # plotting + data-science stack
pip install "esnfed[reservoirpy,flower,llm]" # all integrations

The llm extra also needs a torch build

The FedResPrompt real-LLM backend (TransformersLM) needs torch in addition to transformers. The default surrogate language model has no such requirement.

From source¶

git clone https://github.com/daibeal/esnfed.git
cd esnfed
pip install -e ".[viz,experiments,dev]"
python -m pytest        # 59 tests

Verify¶

import esnfed
print(esnfed.__version__)