ml-property-predictor

Official

Train property heads on pretrained MLIPs

Authorlearningmatter-mit
Version1.0.0
Installs0

System Documentation

What problem does it solve?

This Skill trains a custom property predictor by reusing pretrained MLIP GNN representations, so you can learn new scalar targets (e.g., bulk modulus, bandgap, energies) without building a full model from scratch.

Core Features & Use Cases

  • Trainable readout head on top of an MLIP backbone (MACE or MatGL/M3GNet) for custom scalar regression targets.
  • Supports intensive vs extensive targets, including practical handling differences between MACE and MatGL for scaling and readout behavior.
  • Script-driven workflow that converts structure datasets (.json with Pymatgen dicts or .xyz/.extxyz with ASE/Pymatgen-compatible metadata) into the training format expected by each ML framework.

Quick Start

Ask your agent to train an intensive bulk modulus head using MACE from a JSON dataset of structures and labels by running the MACE training script with the dataset path, target property key, property type, and output directory.

Dependency Matrix

Required Modules

pymatgenasetorchnumpymatgldgltqdmdgl

Components

scripts

💻 Claude Code Installation

Recommended: Let Claude install automatically. Simply copy and paste the text below to Claude Code.

Please help me install this Skill:
Name: ml-property-predictor
Download link: https://github.com/learningmatter-mit/AtomisticSkills/archive/main.zip#ml-property-predictor

Please download this .zip file, extract it, and install it in the .claude/skills/ directory.
View Source Repository

Agent Skills Search Helper

Install a tiny helper to your Agent, search and equip skill from 471,000+ vetted skills library on demand.