Skip to content

Examples

Interactive plots

These are produced by esnfed.viz with its default Plotly backend — pan, zoom and hover. (Generated by examples/plotly_gallery.py.)

Animations

The spectral radius controls the echo state property — eigenvalues must stay inside the unit circle:

Spectral radius sweep

A reservoir's fading response to an input impulse — the "echoes":

Reservoir echoes

Worked example scripts

In the examples/ folder:

  • tutorial_finance.ipynb — an end-to-end federated counterparty-risk walkthrough (also rendered as the Tutorial page).
  • plotly_gallery.py — regenerates the interactive plots above.
  • flower_federated_ridge.py — the Flower integration.
  • make_animations.py — regenerates the GIFs above.
  • privacy_demo.py — differential privacy + secure aggregation of the federated statistics.
  • streaming_demo.py — incremental / streaming ridge and the RLS online readout.
  • classification_demo.py — federated sequence classification (self-contained toy).

Recipes — the newer capabilities

See the Privacy & streaming and Sequence classification guides for the full story.

from esnfed import federated
from esnfed.privacy import PrivacyConfig

cfg = PrivacyConfig(epsilon=1.0, delta=1e-5, clip_state=5.0, clip_target=1.0)
W_dp = federated.federated_ridge_dp(clients, ref, cfg)   # (eps, delta)-DP readout

# the server only ever sees the masked sum of (A_k, B_k); result == exact ridge
W = federated.federated_ridge_secure(clients, ref, seed=0)

from esnfed.streaming import StreamingRidge

acc = StreamingRidge(ref.readout_dim, ref.n_outputs, ridge=1e-6)
for Z_chunk, Y_chunk in stream:        # accumulate as new data arrives
    acc.update(Z_chunk, Y_chunk)
W = acc.readout()                      # exact ridge over everything seen
acc.merge(other_client_acc)            # merging == the federated sum

from esnfed.streaming import RLSReadout

rls = RLSReadout(ref.readout_dim, ref.n_outputs, ridge=1e-6)
for z, y in samples:                   # per-sample Sherman-Morrison updates
    rls.update(z, y)
W = rls.readout()

from esnfed.classification import federated_classifier, predict_labels, accuracy

# client_data: list of (sequences, labels) -- one client per speaker / subject
W_out = federated_classifier(esn, client_data, n_classes)   # exact = centralized
acc = accuracy(y_test, predict_labels(esn, X_test, W_out))

from esnfed import datasets

# align several FRED series into one multivariate forecasting task
u, y = datasets.load_fred_matrix(["TEDRATE", "VIXCLS", "DGS10", "DFF"])

Reproduce the research experiments

pip install -e ".[experiments,viz,reservoirpy,flower]"
python experiments/run_all.py        # synthetic-benchmark figures and tables
python experiments/exp6_finance.py   # federated counterparty-risk study
python experiments/exp7_fedres_prompt.py   # FedResPrompt cost benchmark