hyperiax.Tree

class hyperiax.Tree(topology, schema, data)

Bases: object

A topology + a typed dict of (N, *trailing) arrays.

Construction is always functional: every “mutator” returns a new Tree sharing the same Topology and (usually) Schema, with replaced data.

Data fields are reachable two ways: tree['value'] (dict-style, the safe form when the field name is dynamic or collides with a reserved name) and tree.value (attribute-style, the canonical form mirroring node.value / children.value inside a sweep).

Parameters:

• topology (Topology)

• schema (Schema)

• data (dict)

__init__(topology, schema, data)

Parameters:

• topology (Topology)

• schema (Schema)

• data (dict)

Return type:

None

Methods

__init__(topology, schema, data)
drop(*names)	Drop fields; returns a Tree with reduced schema and data.
empty(topology, schema)	Allocate a Tree of zeros with the declared schema.
from_data(topology, data)	Build a Tree from an existing data dict; schema is inferred from arrays.
set(**fields)	Replace whole fields.
update(**fields)	Add brand-new fields.

Attributes

at	tree.at[indices].set(field=values).
size
topology
schema
data

topology: Topology

schema: Schema

data: dict

classmethod empty(topology, schema)

Allocate a Tree of zeros with the declared schema.

Return type:

Tree

Parameters:

• topology (Topology)

• schema (Schema | Mapping[str, tuple | FieldSpec | None])

classmethod from_data(topology, data)

Build a Tree from an existing data dict; schema is inferred from arrays.

Return type:

Tree

Parameters:

• topology (Topology)

• data (Mapping[str, Array])

property size: int

set(**fields)

Replace whole fields. Each value must already match (N, *spec.shape).

Return type:

Tree

Parameters:

fields (Array)

property at: _AtAccessor

tree.at[indices].set(field=values).

Mirrors jax.numpy.ndarray.at. The returned indexer supports set, add, multiply, min, max (all take fields as keyword arguments and return a new Tree) and get (returns a dict of every field evaluated at indices).

Type:

JAX-style scatter accessor

update(**fields)

Add brand-new fields. Values can be a FieldSpec, a shape tuple, or an array.

Note: this changes the pytree structure and will invalidate any JIT caches keyed on the previous structure. Prefer Tree.empty(topo, full_schema) when you know all the fields up front (the typical case in a sweep pipeline).

Return type:

Tree

drop(*names)

Drop fields; returns a Tree with reduced schema and data.

Return type:

Tree

Parameters:

names (str)