dymad.models¶

class dymad.models.DecoderSpec(kind)¶

Bases: object

property family: Literal['auto', 'graph', 'node']¶

kind: Literal['auto', 'graph', 'node']¶

class dymad.models.DynamicsSpec(kind)¶

Bases: object

property family: Literal['direct', 'skip', 'graph_direct', 'graph_skip']¶

kind: Literal['direct', 'skip', 'graph_direct', 'graph_skip']¶

class dymad.models.EncoderSpec(kind)¶

Bases: object

property family: Literal['smpl', 'raw', 'graph', 'node', 'node_raw', 'smpl_auto', 'graph_auto']¶

kind: Literal['smpl', 'raw', 'graph', 'node', 'node_raw', 'smpl_auto', 'graph_auto']¶

class dymad.models.FeatureSpec(kind)¶

Bases: object

property family: Literal['none', 'cat', 'blin', 'graph_cat', 'graph_blin']¶

kind: Literal['none', 'cat', 'blin', 'graph_cat', 'graph_blin']¶

class dymad.models.MemorySpec(family, latent_state, requires_delay_window)¶

Bases: object

family: str¶

latent_state: str¶

requires_delay_window: bool¶

class dymad.models.ModelSpec(recipe, time_domain, graph_mode, encoder, feature, dynamics, decoder, rollout, memory=None, name=None)¶

Bases: object

Authoritative typed model specification for model construction.

property continuous_time: bool¶

decoder: DecoderSpec¶

dynamics: DynamicsSpec¶

encoder: EncoderSpec¶

feature: FeatureSpec¶

graph_mode: Literal['none', 'graph', 'node']¶

memory: MemorySpec | None = None¶

property model_cls: object¶

name: str | None = None¶

recipe: RecipeSpec¶

rollout: RolloutSpec¶

time_domain: Literal['continuous', 'discrete']¶

exception dymad.models.ModelSpecValidationError¶

Bases: ValueError

Raised when a typed model spec cannot be resolved safely.

class dymad.models.PredefinedModel(model_spec)¶

Bases: object

Compatibility wrapper around one typed predefined model spec.

model_spec: ModelSpec¶

typed_spec()¶

Return type:: ModelSpec

class dymad.models.RecipeSpec(kind, model_cls)¶

Bases: object

kind: Literal['ldm', 'sdm', 'lfm', 'km', 'kmsk', 'kmm']¶

model_cls: object¶

class dymad.models.ResolvedModelSpec(model_spec, dims, encoder_key, feature_key, dynamics_key, decoder_key, predictor_key, predictor, input_order, processor_net, graph_mode, linear_mode, continuous_time)¶

Bases: object

Normalized construction plan produced from a ModelSpec.

continuous_time: bool¶

decoder_key: str¶

dims: dict[str, Any]¶

dynamics_key: str¶

encoder_key: str¶

feature_key: str¶

graph_mode: Literal['none', 'graph', 'node']¶

input_order: str | None¶

linear_mode: Literal['smpl', 'graph']¶

model_spec: ModelSpec¶

predictor: Callable[..., Any]¶

predictor_key: Literal['continuous', 'continuous_np', 'continuous_exp', 'continuous_fenc', 'discrete', 'discrete_exp']¶

processor_net: object¶

class dymad.models.RolloutSpec(family, default_predictor, allowed_predictors, supports_control_inputs=True)¶

Bases: object

allowed_predictors: tuple[Literal['continuous', 'continuous_np', 'continuous_exp', 'continuous_fenc', 'discrete', 'discrete_exp'], ...]¶

default_predictor: Literal['continuous', 'continuous_np', 'continuous_exp', 'continuous_fenc', 'discrete', 'discrete_exp']¶

family: str¶

supports_control_inputs: bool = True¶

dymad.models.build_model(model_spec, model_config, data_meta, dtype=None, device=None)¶: Build a model from a typed ModelSpec.

dymad.models.build_model_from_spec(model_spec, model_config, data_meta, dtype=None, device=None)¶: Compatibility alias for the typed-only build path.

dymad.models.get_dims(model_config, data_meta)¶

Determine dimensions used in the model based on configuration and metadata.

This is a generic guess and can be overridden by specific model classes.

Modules

`collections`
`components`
`helpers`
`model_base`
`model_spec`	Typed model-spec contracts for predefined-model construction.
`prediction`
`recipes`
`recipes_corr`
`rollout_engine`	Typed rollout-engine selection for migrated model-spec families.
`runtime_view`	Narrow runtime-view adapter for model helper/component functions.