aqua-blue 0.2.19

Lightweight and basic reservoir computing library

aqua-blue

Lightweight and basic reservoir computing library

🌊 What is aqua-blue?

aqua-blue is a lightweight python library for reservoir computing (specifically echo state networks) depending only on numpy. aqua-blue's namesake comes from:

• A blue ocean of data, aka a reservoir 💧
• A very fancy cat named Blue 🐾

🐜 Found a bug?

Please open an issue here if you found a bug! The easier it is to reproduce the bug, the faster we will find a solution to the problem. Please consider including the following info in your issue:

• Steps to reproduce
• Expected and actual behavior
• Version info, OS, etc.

🔧 Contributing

Please see CONTRIBUTING.md for instructions on how to contribute to aqua-blue ☺

📥 Installation

aqua-blue is on PyPI, and can therefore be installed with pip:

pip install aqua-blue

📝 Quickstart

import numpy as np
import aqua_blue

# generate arbitrary two-dimensional time series
# y_1(t) = cos(t), y_2(t) = sin(t)
# resulting dependent variable has shape (number of timesteps, 2)
t = np.linspace(0, 4.0 * np.pi, 10_000)
y = np.vstack((2.0 * np.cos(t) + 1, 5.0 * np.sin(t) - 1)).T

# create time series object to feed into echo state network
time_series = aqua_blue.time_series.TimeSeries(dependent_variable=y, times=t)

# normalize
normalizer = aqua_blue.utilities.Normalizer()
time_series = normalizer.normalize(time_series)

# make model and train
model = aqua_blue.models.Model(
    reservoir=aqua_blue.reservoirs.DynamicalReservoir(
        reservoir_dimensionality=100,
        input_dimensionality=2
    ),
    readout=aqua_blue.readouts.LinearReadout()
)
model.train(time_series)

# predict and denormalize
prediction = model.predict(horizon=1_000)
prediction = normalizer.denormalize(prediction)

📃 License

aqua-blue is released under the MIT License.

---

Blue, the cat behind aqua-blue.