OSJL-py / PRCpy

Simple modular python package for physical reservoir computing. Use your own experimental data.

PRCpy: A Python Library for Physical Reservoir Computing

PRCpy is a Python package designed to ease experimental data processing for physical reservoir computing.

Features

• Data handling and preprocessing.
• Customizable data processing pipelines for various research needs.

Installation

PRCpy requires Python 3.9 or later.

Using pip

pip install prcpy

Using Poetry

poetry add prcpy

Note

Latest release is always recommended.

PIP: pip install prcpy --upgrade
POERTY: poetry update prcpy

General usage overview

1. Define data path
2. Define pre-processing parameters
3. Create RC pipeline
4. Define target and add to pipeline
5. Define model for training
6. Define RC parameters
7. Run RC

Example:

import PRCpy

from prcpy.RC import Pipeline
from prcpy.TrainingModels.RegressionModels import define_Ridge
from prcpy.Maths.Target_functions import get_mackey_glass, get_square_waves

Define data directory and processing parameters

Note: Data files must contain "scan" in their file names. See examples/data for example data files.

data_dir_path = "your/data/path"
process_params = {
    "Xs": "Frequency",
    "Readouts": "Spectra",
    "remove_bg": True,
    "bg_fname": "background_data.txt",
    "smooth": False,
    "smooth_win": 51,
    "smooth_rank": 4,
    "cut_xs": False,
    "x1": 2,
    "x2": 5,
    "normalize": False,
    "sample": True,
    "sample_rate": 13
}

Create RC pipeline

rc_pipeline = Pipeline(data_dir_path, process_params)

Target generation

Transformation

period = 10
sample_spacing = rc_pipeline.get_sample_spacing(period)
target_values = get_square_waves(sample_spacing, period, norm=True)

Forecasting

mg_path = "mackey_glass_t17.npy"
target_values = get_npy_data(mg_path, norm=True)

Add target to pipeline

rc_pipeline.define_target(target_values)

Define model

model_params = {
        "alpha": 1e-3,
        "fit_intercept": True,
        "copy_X": True,
        "max_iter": None,
        "tol": 0.0001,
        "solver": "auto",
        "positive": False,
        "random_state": None,
    }
model = define_Ridge(model_params)

Define RC parameters

Set "tau": 0 for transformation.

rc_params = {
        "model": model,
        "tau": 10,
        "test_size": 0.3,
        "error_type": "MSE"
    }

Run RC

rc_pipeline.run()

Get results

results = rc_pipeline.get_rc_results()

Authors & Maintainers

We are a neuromorphic computing division within the UCL Spintronics Group at London Centre for Nanotechnology, University College London, UK. For any queries about PRCpy, please contact Dr. Oscar Lee.

Research enquries

For collaborations or research enquires, please contact Prof. Hide Kurebayashi.

Find out more on PRC

PRCpy

• Refer to PRCpy documentation [TBA] for detailed package documentation.
• Refer to the PRCpy tutorial paper [TBA] for detailed tutorial use of PRCpy.

RC publications from the group

Research articles

• Task-adaptive physical reservoir computing, O. Lee, et al., Nature Materials (2024).
• Neuromorphic Few-Shot Learning: Generalization in Multilayer Physical Neural Networks, K. Stenning, et al., arXiv (2023).
• Reconfigurable training and reservoir computing in an artificial spin-vortex ice via spin-wave fingerprinting, J. Gartside, et al., Nature Nanotechnology (2022).

Review/perspectives

• Perspective on unconventional computing using magnetic skyrmions, O. Lee, et al., Applied Physics Letters (2023).
• Memristive, Spintronic, and 2D-Materials-Based Devices to Improve and Complement Computing Hardware, D. Joksas, et al., Advanced Intelligent Systems (2022).

Outreach

• Physical reservoir computers that can adapt to perform different tasks, H. Kurebayashi, O. Lee, Nature Materials Research Briefing (2024).

Recent PRC publications

TBA.

Contributing

Any community contributions are welcome. Please refer to the project's GitHub repository for contribution guidelines.