lerpnet 1.0.3

Lightweight learnable interpolation using compact MLPs (edge & ESP32 ready)

LerpNet (Learnable Interpolation Network)

LerpNet is a lightweight neural interpolation library designed for fast regression and function approximation. It trains a compact MLP, learns smooth interpolation from data, and exports results in Python-friendly .pkl and embedded-friendly .json formats for deployment on systems like ESP32.

No heavy deep learning frameworks are required—perfect for edge AI applications.

---

Features

• Learnable interpolation using a compact MLP
• Extremely lightweight (no PyTorch / TensorFlow required)
• Outputs:
  • .pkl for Python inference
  • .json for C++ / ESP32 inference
• Built-in utilities:
  • Data normalization
  • Learning rate finder
  • Training metrics (RMSE, MAPE, R²)
• Designed for edge & embedded deployment
• Ideal for:
  • Sensor calibration
  • Curve fitting
  • Control systems
  • Real-time interpolation
  • Function approximation
  • Learnable lookup tables
  • Replacement for hard-coded formulas

---

Generated files:

1. model.pkl — Python inference
2. norm_constants.json — Input/output normalization
3. model_weights.json — Embedded / C++ inference
4. Training plots & CSV (actual_vs_predicted.csv)

Installation

pip install lerpnet