mifrost Documentation

mifrost is a high-performance graph encoding library for planning states and transitions. It provides native BatchEncoding outputs and optional conversion to PyTorch Geometric (Data/HeteroData).

Start Here

• New users: Installation
• First usable graph: First Encoding
• Throughput workflows: Batching and Streaming

Choose an Encoder Quickly

• Use HGraphEncoder for state-centric heterogeneous encodings (goals/actions/history optional).
• Use FlatRelationEncoder for compact packed state/goal/action/history encodings.
• Use HorizonEncoder for root state + lookahead DAG candidate structure.
• Use FlatHorizonEncoder for compact packed lookahead/state-target encodings.
• Use TransitionHGraphEncoder for full current-to-successor transition structure.
• Use FlatTransitionEncoder for compact full current-to-successor transition structure.
• Use TransitionEffectsHGraphEncoder for delta/effects-focused transition features.
• Use FlatTransitionEffectsEncoder for compact delta/effects-focused transition features.
• Use ColorEncoder for compact homogeneous graph baselines.
• Use ILGEncoder for Python-native ILG topology/features.

See How to Choose an Encoder for scenario-specific guidance.

What Is Covered

• Native-first API and conversion model
• Native hetero/homo tensor facades (as_hetero, as_homo) for model-boundary use
• Single, batch, append-only stream, and mutable stream workflows
• Dynamic graph fields (GraphFieldSpec) and batching semantics
• Adapter registration for custom wrapper types