dymad.models.collections

Module Attributes

LDM(model_config, data_meta[, dtype, device])	Latent dynamics model (LDM), continuous-time.
DLDM(model_config, data_meta[, dtype, device])	LDM, discrete-time.
GLDM(model_config, data_meta[, dtype, device])	LDM with graph autoencoder, continuous-time.
DGLDM(model_config, data_meta[, dtype, device])	LDM with graph autoencoder, discrete-time.
LDMG(model_config, data_meta[, dtype, device])	LDM with graph dynamics, continuous-time.
DLDMG(model_config, data_meta[, dtype, device])	LDM with graph dynamics, discrete-time.
DSDM(model_config, data_meta[, dtype, device])	Sequential dynamics model (SDM), always discrete-time.
DSDMG(model_config, data_meta[, dtype, device])	SDM with graph dynamics, discrete-time.
KBF(model_config, data_meta[, dtype, device])	Koopman bilinear form (KBF), continuous-time.
DKBF(model_config, data_meta[, dtype, device])	KBF, discrete-time.
GKBF(model_config, data_meta[, dtype, device])	KBF with graph autoencoder, continuous-time.
DGKBF(model_config, data_meta[, dtype, device])	KBF with graph autoencoder, discrete-time.
LTI(model_config, data_meta[, dtype, device])	Linear time-invariant (LTI), continuous-time.
DLTI(model_config, data_meta[, dtype, device])	LTI, discrete-time.
GLTI(model_config, data_meta[, dtype, device])	LTI with graph autoencoder, continuous-time.
DGLTI(model_config, data_meta[, dtype, device])	LTI with graph autoencoder, discrete-time.
KM(model_config, data_meta[, dtype, device])	Kernel machine (KM), continuous-time.
KMM(model_config, data_meta[, dtype, device])	Kernel machine on manifold (KMM), continuous-time.
DKM(model_config, data_meta[, dtype, device])	KM, discrete-time.
GKM(model_config, data_meta[, dtype, device])	KM with graph autoencoder, continuous-time.
DGKM(model_config, data_meta[, dtype, device])	KM with graph autoencoder, discrete-time.
DKMSK(model_config, data_meta[, dtype, device])	Kernel machine with skip-connection (KMSK), discrete-time.
DGKMSK(model_config, data_meta[, dtype, device])	KMSK with graph autoencoder, discrete-time.

Classes

PredefinedModel(model_spec)

Compatibility wrapper around one typed predefined model spec.

dymad.models.collections.DGKBF(model_config, data_meta, dtype=None, device=None) = PredefinedModel(model_spec=ModelSpec(recipe=RecipeSpec(kind='lfm', model_cls=<class 'dymad.models.recipes.CD_LFM'>), time_domain='discrete', graph_mode='graph', encoder=EncoderSpec(kind='graph_auto'), feature=FeatureSpec(kind='graph_blin'), dynamics=DynamicsSpec(kind='direct'), decoder=DecoderSpec(kind='graph'), rollout=RolloutSpec(family='default', default_predictor='discrete', allowed_predictors=('discrete', 'discrete_exp'), supports_control_inputs=True), memory=None, name='DGKBF'))

KBF with graph autoencoder, discrete-time.

dymad.models.collections.DGKM(model_config, data_meta, dtype=None, device=None) = PredefinedModel(model_spec=ModelSpec(recipe=RecipeSpec(kind='km', model_cls=<class 'dymad.models.recipes.CD_KM'>), time_domain='discrete', graph_mode='graph', encoder=EncoderSpec(kind='graph_auto'), feature=FeatureSpec(kind='graph_blin'), dynamics=DynamicsSpec(kind='direct'), decoder=DecoderSpec(kind='graph'), rollout=RolloutSpec(family='default', default_predictor='discrete', allowed_predictors=('discrete', 'discrete_exp'), supports_control_inputs=True), memory=None, name='DGKM'))

KM with graph autoencoder, discrete-time.

dymad.models.collections.DGKMSK(model_config, data_meta, dtype=None, device=None) = PredefinedModel(model_spec=ModelSpec(recipe=RecipeSpec(kind='kmsk', model_cls=<class 'dymad.models.recipes.CD_KMSK'>), time_domain='discrete', graph_mode='graph', encoder=EncoderSpec(kind='graph_auto'), feature=FeatureSpec(kind='graph_blin'), dynamics=DynamicsSpec(kind='skip'), decoder=DecoderSpec(kind='graph'), rollout=RolloutSpec(family='default', default_predictor='discrete', allowed_predictors=('discrete', 'discrete_exp'), supports_control_inputs=True), memory=None, name='DGKMSK'))

KMSK with graph autoencoder, discrete-time.

dymad.models.collections.DGLDM(model_config, data_meta, dtype=None, device=None) = PredefinedModel(model_spec=ModelSpec(recipe=RecipeSpec(kind='ldm', model_cls=<class 'dymad.models.recipes.CD_LDM'>), time_domain='discrete', graph_mode='graph', encoder=EncoderSpec(kind='graph'), feature=FeatureSpec(kind='none'), dynamics=DynamicsSpec(kind='direct'), decoder=DecoderSpec(kind='graph'), rollout=RolloutSpec(family='default', default_predictor='discrete', allowed_predictors=('discrete', 'discrete_exp'), supports_control_inputs=True), memory=None, name='DGLDM'))

LDM with graph autoencoder, discrete-time.

dymad.models.collections.DGLTI(model_config, data_meta, dtype=None, device=None) = PredefinedModel(model_spec=ModelSpec(recipe=RecipeSpec(kind='lfm', model_cls=<class 'dymad.models.recipes.CD_LFM'>), time_domain='discrete', graph_mode='graph', encoder=EncoderSpec(kind='graph_auto'), feature=FeatureSpec(kind='graph_cat'), dynamics=DynamicsSpec(kind='direct'), decoder=DecoderSpec(kind='graph'), rollout=RolloutSpec(family='lti', default_predictor='discrete', allowed_predictors=('discrete', 'discrete_exp'), supports_control_inputs=True), memory=MemorySpec(family='concat-latent-control', latent_state='cat', requires_delay_window=True), name='DGLTI'))

LTI with graph autoencoder, discrete-time.

dymad.models.collections.DKBF(model_config, data_meta, dtype=None, device=None) = PredefinedModel(model_spec=ModelSpec(recipe=RecipeSpec(kind='lfm', model_cls=<class 'dymad.models.recipes.CD_LFM'>), time_domain='discrete', graph_mode='none', encoder=EncoderSpec(kind='smpl_auto'), feature=FeatureSpec(kind='blin'), dynamics=DynamicsSpec(kind='direct'), decoder=DecoderSpec(kind='auto'), rollout=RolloutSpec(family='default', default_predictor='discrete', allowed_predictors=('discrete', 'discrete_exp'), supports_control_inputs=True), memory=None, name='DKBF'))

KBF, discrete-time.

dymad.models.collections.DKM(model_config, data_meta, dtype=None, device=None) = PredefinedModel(model_spec=ModelSpec(recipe=RecipeSpec(kind='km', model_cls=<class 'dymad.models.recipes.CD_KM'>), time_domain='discrete', graph_mode='none', encoder=EncoderSpec(kind='smpl_auto'), feature=FeatureSpec(kind='blin'), dynamics=DynamicsSpec(kind='direct'), decoder=DecoderSpec(kind='auto'), rollout=RolloutSpec(family='default', default_predictor='discrete', allowed_predictors=('discrete', 'discrete_exp'), supports_control_inputs=True), memory=None, name='DKM'))

KM, discrete-time.

dymad.models.collections.DKMSK(model_config, data_meta, dtype=None, device=None) = PredefinedModel(model_spec=ModelSpec(recipe=RecipeSpec(kind='kmsk', model_cls=<class 'dymad.models.recipes.CD_KMSK'>), time_domain='discrete', graph_mode='none', encoder=EncoderSpec(kind='smpl_auto'), feature=FeatureSpec(kind='blin'), dynamics=DynamicsSpec(kind='skip'), decoder=DecoderSpec(kind='auto'), rollout=RolloutSpec(family='default', default_predictor='discrete', allowed_predictors=('discrete', 'discrete_exp'), supports_control_inputs=True), memory=None, name='DKMSK'))

Kernel machine with skip-connection (KMSK), discrete-time.

dymad.models.collections.DLDM(model_config, data_meta, dtype=None, device=None) = PredefinedModel(model_spec=ModelSpec(recipe=RecipeSpec(kind='ldm', model_cls=<class 'dymad.models.recipes.CD_LDM'>), time_domain='discrete', graph_mode='none', encoder=EncoderSpec(kind='smpl'), feature=FeatureSpec(kind='none'), dynamics=DynamicsSpec(kind='direct'), decoder=DecoderSpec(kind='auto'), rollout=RolloutSpec(family='default', default_predictor='discrete', allowed_predictors=('discrete', 'discrete_exp'), supports_control_inputs=True), memory=None, name='DLDM'))

LDM, discrete-time.

dymad.models.collections.DLDMG(model_config, data_meta, dtype=None, device=None) = PredefinedModel(model_spec=ModelSpec(recipe=RecipeSpec(kind='ldm', model_cls=<class 'dymad.models.recipes.CD_LDM'>), time_domain='discrete', graph_mode='node', encoder=EncoderSpec(kind='node'), feature=FeatureSpec(kind='none'), dynamics=DynamicsSpec(kind='graph_direct'), decoder=DecoderSpec(kind='node'), rollout=RolloutSpec(family='default', default_predictor='discrete', allowed_predictors=('discrete', 'discrete_exp'), supports_control_inputs=True), memory=None, name='DLDMG'))

LDM with graph dynamics, discrete-time.

dymad.models.collections.DLTI(model_config, data_meta, dtype=None, device=None) = PredefinedModel(model_spec=ModelSpec(recipe=RecipeSpec(kind='lfm', model_cls=<class 'dymad.models.recipes.CD_LFM'>), time_domain='discrete', graph_mode='none', encoder=EncoderSpec(kind='smpl_auto'), feature=FeatureSpec(kind='cat'), dynamics=DynamicsSpec(kind='direct'), decoder=DecoderSpec(kind='auto'), rollout=RolloutSpec(family='lti', default_predictor='discrete', allowed_predictors=('discrete', 'discrete_exp'), supports_control_inputs=True), memory=MemorySpec(family='concat-latent-control', latent_state='cat', requires_delay_window=True), name='DLTI'))

LTI, discrete-time.

dymad.models.collections.DSDM(model_config, data_meta, dtype=None, device=None) = PredefinedModel(model_spec=ModelSpec(recipe=RecipeSpec(kind='sdm', model_cls=<class 'dymad.models.recipes.CD_SDM'>), time_domain='discrete', graph_mode='none', encoder=EncoderSpec(kind='raw'), feature=FeatureSpec(kind='none'), dynamics=DynamicsSpec(kind='direct'), decoder=DecoderSpec(kind='auto'), rollout=RolloutSpec(family='default', default_predictor='discrete', allowed_predictors=('discrete', 'discrete_exp'), supports_control_inputs=True), memory=None, name='DSDM'))

Sequential dynamics model (SDM), always discrete-time.

dymad.models.collections.DSDMG(model_config, data_meta, dtype=None, device=None) = PredefinedModel(model_spec=ModelSpec(recipe=RecipeSpec(kind='sdm', model_cls=<class 'dymad.models.recipes.CD_SDM'>), time_domain='discrete', graph_mode='node', encoder=EncoderSpec(kind='node_raw'), feature=FeatureSpec(kind='none'), dynamics=DynamicsSpec(kind='graph_direct'), decoder=DecoderSpec(kind='node'), rollout=RolloutSpec(family='default', default_predictor='discrete', allowed_predictors=('discrete', 'discrete_exp'), supports_control_inputs=True), memory=None, name='DSDMG'))

SDM with graph dynamics, discrete-time.

dymad.models.collections.GKBF(model_config, data_meta, dtype=None, device=None) = PredefinedModel(model_spec=ModelSpec(recipe=RecipeSpec(kind='lfm', model_cls=<class 'dymad.models.recipes.CD_LFM'>), time_domain='continuous', graph_mode='graph', encoder=EncoderSpec(kind='graph_auto'), feature=FeatureSpec(kind='graph_blin'), dynamics=DynamicsSpec(kind='direct'), decoder=DecoderSpec(kind='graph'), rollout=RolloutSpec(family='default', default_predictor='continuous', allowed_predictors=('continuous', 'continuous_np', 'continuous_exp'), supports_control_inputs=True), memory=None, name='GKBF'))

KBF with graph autoencoder, continuous-time.

dymad.models.collections.GKM(model_config, data_meta, dtype=None, device=None) = PredefinedModel(model_spec=ModelSpec(recipe=RecipeSpec(kind='km', model_cls=<class 'dymad.models.recipes.CD_KM'>), time_domain='continuous', graph_mode='graph', encoder=EncoderSpec(kind='graph_auto'), feature=FeatureSpec(kind='graph_blin'), dynamics=DynamicsSpec(kind='direct'), decoder=DecoderSpec(kind='graph'), rollout=RolloutSpec(family='default', default_predictor='continuous', allowed_predictors=('continuous', 'continuous_np', 'continuous_exp'), supports_control_inputs=True), memory=None, name='GKM'))

KM with graph autoencoder, continuous-time.

dymad.models.collections.GLDM(model_config, data_meta, dtype=None, device=None) = PredefinedModel(model_spec=ModelSpec(recipe=RecipeSpec(kind='ldm', model_cls=<class 'dymad.models.recipes.CD_LDM'>), time_domain='continuous', graph_mode='graph', encoder=EncoderSpec(kind='graph'), feature=FeatureSpec(kind='none'), dynamics=DynamicsSpec(kind='direct'), decoder=DecoderSpec(kind='graph'), rollout=RolloutSpec(family='default', default_predictor='continuous', allowed_predictors=('continuous', 'continuous_np', 'continuous_exp'), supports_control_inputs=True), memory=None, name='GLDM'))

LDM with graph autoencoder, continuous-time.

dymad.models.collections.GLTI(model_config, data_meta, dtype=None, device=None) = PredefinedModel(model_spec=ModelSpec(recipe=RecipeSpec(kind='lfm', model_cls=<class 'dymad.models.recipes.CD_LFM'>), time_domain='continuous', graph_mode='graph', encoder=EncoderSpec(kind='graph_auto'), feature=FeatureSpec(kind='graph_cat'), dynamics=DynamicsSpec(kind='direct'), decoder=DecoderSpec(kind='graph'), rollout=RolloutSpec(family='lti', default_predictor='continuous', allowed_predictors=('continuous', 'continuous_np', 'continuous_exp'), supports_control_inputs=True), memory=MemorySpec(family='concat-latent-control', latent_state='cat', requires_delay_window=True), name='GLTI'))

LTI with graph autoencoder, continuous-time.

dymad.models.collections.KBF(model_config, data_meta, dtype=None, device=None) = PredefinedModel(model_spec=ModelSpec(recipe=RecipeSpec(kind='lfm', model_cls=<class 'dymad.models.recipes.CD_LFM'>), time_domain='continuous', graph_mode='none', encoder=EncoderSpec(kind='smpl_auto'), feature=FeatureSpec(kind='blin'), dynamics=DynamicsSpec(kind='direct'), decoder=DecoderSpec(kind='auto'), rollout=RolloutSpec(family='default', default_predictor='continuous', allowed_predictors=('continuous', 'continuous_np', 'continuous_exp'), supports_control_inputs=True), memory=None, name='KBF'))

Koopman bilinear form (KBF), continuous-time.

dymad.models.collections.KM(model_config, data_meta, dtype=None, device=None) = PredefinedModel(model_spec=ModelSpec(recipe=RecipeSpec(kind='km', model_cls=<class 'dymad.models.recipes.CD_KM'>), time_domain='continuous', graph_mode='none', encoder=EncoderSpec(kind='smpl_auto'), feature=FeatureSpec(kind='blin'), dynamics=DynamicsSpec(kind='direct'), decoder=DecoderSpec(kind='auto'), rollout=RolloutSpec(family='default', default_predictor='continuous', allowed_predictors=('continuous', 'continuous_np', 'continuous_exp'), supports_control_inputs=True), memory=None, name='KM'))

Kernel machine (KM), continuous-time.

dymad.models.collections.KMM(model_config, data_meta, dtype=None, device=None) = PredefinedModel(model_spec=ModelSpec(recipe=RecipeSpec(kind='kmm', model_cls=<class 'dymad.models.recipes.CD_KMM'>), time_domain='continuous', graph_mode='none', encoder=EncoderSpec(kind='smpl_auto'), feature=FeatureSpec(kind='blin'), dynamics=DynamicsSpec(kind='direct'), decoder=DecoderSpec(kind='auto'), rollout=RolloutSpec(family='kmm', default_predictor='continuous_fenc', allowed_predictors=('continuous_fenc',), supports_control_inputs=True), memory=None, name='KMM'))

Kernel machine on manifold (KMM), continuous-time.

dymad.models.collections.LDM(model_config, data_meta, dtype=None, device=None) = PredefinedModel(model_spec=ModelSpec(recipe=RecipeSpec(kind='ldm', model_cls=<class 'dymad.models.recipes.CD_LDM'>), time_domain='continuous', graph_mode='none', encoder=EncoderSpec(kind='smpl'), feature=FeatureSpec(kind='none'), dynamics=DynamicsSpec(kind='direct'), decoder=DecoderSpec(kind='auto'), rollout=RolloutSpec(family='default', default_predictor='continuous', allowed_predictors=('continuous', 'continuous_np', 'continuous_exp'), supports_control_inputs=True), memory=None, name='LDM'))

Latent dynamics model (LDM), continuous-time.

dymad.models.collections.LDMG(model_config, data_meta, dtype=None, device=None) = PredefinedModel(model_spec=ModelSpec(recipe=RecipeSpec(kind='ldm', model_cls=<class 'dymad.models.recipes.CD_LDM'>), time_domain='continuous', graph_mode='node', encoder=EncoderSpec(kind='node'), feature=FeatureSpec(kind='none'), dynamics=DynamicsSpec(kind='graph_direct'), decoder=DecoderSpec(kind='node'), rollout=RolloutSpec(family='default', default_predictor='continuous', allowed_predictors=('continuous', 'continuous_np', 'continuous_exp'), supports_control_inputs=True), memory=None, name='LDMG'))

LDM with graph dynamics, continuous-time.

dymad.models.collections.LTI(model_config, data_meta, dtype=None, device=None) = PredefinedModel(model_spec=ModelSpec(recipe=RecipeSpec(kind='lfm', model_cls=<class 'dymad.models.recipes.CD_LFM'>), time_domain='continuous', graph_mode='none', encoder=EncoderSpec(kind='smpl_auto'), feature=FeatureSpec(kind='cat'), dynamics=DynamicsSpec(kind='direct'), decoder=DecoderSpec(kind='auto'), rollout=RolloutSpec(family='lti', default_predictor='continuous', allowed_predictors=('continuous', 'continuous_np', 'continuous_exp'), supports_control_inputs=True), memory=MemorySpec(family='concat-latent-control', latent_state='cat', requires_delay_window=True), name='LTI'))

Linear time-invariant (LTI), continuous-time.

class dymad.models.collections.PredefinedModel(model_spec)

Bases: object

Compatibility wrapper around one typed predefined model spec.

model_spec: ModelSpec

typed_spec()

Return type:

ModelSpec