jaxdem.utils.h5#

HDF5 save/load utilities (v2).

Design goals (no API changes):

  • Generic object round-trip for JaxDEM dataclasses and common containers.

  • Skip callables with a warning (user handles them explicitly, e.g. DP containers).

  • Robust schema evolution: warn on unknown fields, warn + default missing fields.

  • Enforce Python types for dataclass fields marked metadata={“static”: True} to keep JAX static hashing happy (e.g. NeighborList.max_neighbors).

This module intentionally does NOT add any top-level file format metadata. It does use minimal per-node tags (e.g. “__kind__”, “__class__”) required to round-trip Python types through HDF5.

Functions

load(path, *[, warn_missing, warn_unknown, ...])

Load an object saved with save().

save(obj, path, *[, overwrite])

jaxdem.utils.h5.save(obj: Any, path: str, *, overwrite: bool = True) → None[source]#
jaxdem.utils.h5.load(path: str, *, warn_missing: bool = True, warn_unknown: bool = True, state_shape: tuple[int, ...] | None = None) → Any[source]#

Load an object saved with save().

state_shape is an optional (N, dim) hint used to bootstrap a System skeleton when the file does not contain the datasets needed to infer it.

On this page