PyPIELM

A Unified and Reproducible Framework for Physics-Informed Extreme Learning Machines

Overview

PyPIELM is an open-source PyTorch-native library that provides a unified implementation of 26+ Physics-Informed Extreme Learning Machine (PIELM) variants alongside PINN baselines for solving partial differential equations (PDEs). It exposes a scikit-learn-style fit / predict / score API so that any PIELM or PINN variant can be used in three lines of code.

Key Features

Installation

pip install pypielm                  # core only
pip install "pypielm[viz]"           # + matplotlib
pip install "pypielm[viz,bench]"     # + memory profiling
pip install "pypielm[dev]"           # full dev environment

From source:

git clone https://github.com/KStruniawski/pypielm.git
cd pypielm
pip install -e ".[dev]"

Quickstart

import pypielm
from pypielm.data import auto_load
from pypielm.models import CorePIELM
from pypielm.pde.operators import AnalyticLaplacian

# Load data (CSV, NPZ, PINNacle .dat, …)
ds = auto_load("data/poisson_classic.dat", source="pinnacle")

# Train model
model = CorePIELM(hidden_dim=300, ridge_lambda=1e-8)
model.fit(ds, pde_operator=AnalyticLaplacian())

# Evaluate
print(model.score(ds.X_test, ds.y_test))  # relative L² error

Documentation

Full API reference and tutorials: pypielm.readthedocs.io

Citation

If you use PyPIELM in your research, please cite:

@software{struniawski2026pypielm,
  author  = {Struniawski, Karol},
  title   = {{PyPIELM}: A Unified and Reproducible Framework for
             Physics-Informed Extreme Learning Machines},
  year    = {2026},
  url     = {https://github.com/KStruniawski/pypielm},
}

Related

• PyPIELM App — Streamlit web UI for training, benchmarking and exporting models: github.com/KStruniawski/pypielm-app

License

CC BY-NC-ND 4.0 © Karol Struniawski