Metadata-Version: 2.1
Name: lerpnet
Version: 1.0.0
Summary: Lightweight learnable interpolation using compact MLPs (edge & ESP32 ready)
Home-page: 
Author: Somendra Seth
Author-email: akhilresearch07@gmail.com
License: MIT
Keywords: machine-learning,interpolation,mlp,edge-ai,embedded,esp32,tinyml
Classifier: Development Status :: 5 - Production/Stable
Classifier: Intended Audience :: Developers
Classifier: Intended Audience :: Education
Classifier: License :: OSI Approved :: MIT License
Classifier: Operating System :: OS Independent
Classifier: Programming Language :: Python :: 3
Classifier: Programming Language :: Python :: 3 :: Only
Classifier: Topic :: Scientific/Engineering :: Artificial Intelligence
Classifier: Topic :: Software Development :: Embedded Systems
Requires-Python: >=3.8
Description-Content-Type: text/markdown
License-File: LICENSE.txt
Requires-Dist: numpy
Requires-Dist: matplotlib

LerpNet (Learnable Interpolation Network)

LerpNet is a lightweight neural interpolation library designed for fast regression and function approximation, with export-ready weights for embedded systems like ESP32 (C++).

It trains a small MLP, learns smooth interpolation from data, and exports results in JSON and PKL formats for easy deployment.

Features
Learnable interpolation using a compact MLP
Extremely lightweight (no deep learning frameworks)

Outputs:
.pkl for Python inference
.json for C++ / ESP32 inference

Built-in:
Data normalization
Learning rate finder
Training metrics (RMSE, MAPE, R²)
Designed for edge & embedded deployment

Installation
pip install numpy matplotlib
(No PyTorch / TensorFlow required)




Quick Start
Train a model

from lerpnet import create_and_train
xs = [[1], [2], [3], [4], [5]]
ys = [2, 4, 6, 8, 10]
create_and_train(
    xs,
    ys,
    output_dir="output",
    architecture=[4, 4, 1],
    training_epochs=3000
)


This generates:
model.pkl
norm_constants.json
model_weights.json (for C++ / ESP32)
Training plots & CSV


Load & Predict (Python)

from lerpnet import load_and_predict
y = load_and_predict(
    weights_json_path="output/model_weights.json",
    xs=[6]
)
print(y)



🧠 How It Works (High Level)
Learns a smooth interpolation function from data
Uses a compact MLP with tanh activation
Automatically normalizes inputs & outputs
Exports raw weights for framework-free inference


Embedded / ESP32 Use
Use model_weights.json
No Python required on device
Implement forward pass in C/C++

Ideal for:
Sensor calibration
Curve fitting
Control systems
Real-time interpolation
Function approximation
Learnable lookup tables
Edge AI regression
Replacement for hard-coded interpolation formulas



Output Files
File	Purpose
model.pkl       	  Python-only inference
model_weights.json	  Embedded / C++ inference
norm_constants.json	  Input/output scaling
actual_vs_predicted.csv	  Error analysis
.png files	          Training visualizations
