reconstruction

Reconstruction-specific Model class(es).

class graphnet.models.task.reconstruction.AzimuthReconstructionWithKappa(*args, **kwargs)

Bases: StandardLearnedTask

Reconstructs azimuthal angle and associated kappa (1/var).

Construct StandardLearnedTask.

Parameters:

• hidden_size (int) – The number of columns in the output of the last latent layer of Model using this Task. Available through Model.nb_outputs

• args (Any)

• kwargs (Any)

Return type:

object

default_target_labels = ['azimuth']

default_prediction_labels = ['azimuth_pred', 'azimuth_kappa']

nb_inputs = 2

class graphnet.models.task.reconstruction.AzimuthReconstruction(*args, **kwargs)

Bases: AzimuthReconstructionWithKappa

Reconstructs azimuthal angle.

Construct StandardLearnedTask.

Parameters:

• hidden_size (int) – The number of columns in the output of the last latent layer of Model using this Task. Available through Model.nb_outputs

• args (Any)

• kwargs (Any)

Return type:

object

default_target_labels = ['azimuth']

default_prediction_labels = ['azimuth_pred']

nb_inputs = 2

class graphnet.models.task.reconstruction.DirectionReconstructionWithKappa(*args, **kwargs)

Bases: StandardLearnedTask

Reconstructs direction with kappa from the 3D-vMF distribution.

Kappa is the length of the three outputs, so it can only grow as far as the head’s output scale does in training. Trained models hit a kappa ceiling (~1e3 in practice) that bright events exceed, and their predicted distributions come out underconfident. Prefer DirectionReconstructionWithLogKappa, which predicts kappa separately.

Construct StandardLearnedTask.

Parameters:

• hidden_size (int) – The number of columns in the output of the last latent layer of Model using this Task. Available through Model.nb_outputs

• args (Any)

• kwargs (Any)

Return type:

object

default_target_labels = ['direction']

default_prediction_labels = ['dir_x_pred', 'dir_y_pred', 'dir_z_pred', 'direction_kappa']

nb_inputs = 3

class graphnet.models.task.reconstruction.DirectionReconstructionWithLogKappa(*args, **kwargs)

Bases: StandardLearnedTask

Reconstructs direction with log-kappa from the 3D-vMF distribution.

The direction is the normalised first three outputs and kappa = kappa_min + exp(x[:, 3]). The floor kappa_min keeps training from settling on the uniform distribution at initialisation, where the direction’s gradient would otherwise vanish with kappa.

Construct DirectionReconstructionWithLogKappa.

Parameters:

• *args (Any) – Passed to StandardLearnedTask.

• kappa_min (float, default: 1.0) – Lower bound on the predicted kappa; must be positive.

• **kwargs (Any) – Passed to StandardLearnedTask.

Return type:

object

default_target_labels = ['direction']

default_prediction_labels = ['dir_x_pred', 'dir_y_pred', 'dir_z_pred', 'direction_kappa']

nb_inputs = 4

class graphnet.models.task.reconstruction.ZenithReconstruction(*args, **kwargs)

Bases: StandardLearnedTask

Reconstructs zenith angle.

Construct StandardLearnedTask.

Parameters:

• hidden_size (int) – The number of columns in the output of the last latent layer of Model using this Task. Available through Model.nb_outputs

• args (Any)

• kwargs (Any)

Return type:

object

default_target_labels = ['zenith']

default_prediction_labels = ['zenith_pred']

nb_inputs = 1

class graphnet.models.task.reconstruction.ZenithReconstructionWithKappa(*args, **kwargs)

Bases: ZenithReconstruction

Reconstructs zenith angle and associated kappa (1/var).

Construct StandardLearnedTask.

Parameters:

• hidden_size (int) – The number of columns in the output of the last latent layer of Model using this Task. Available through Model.nb_outputs

• args (Any)

• kwargs (Any)

Return type:

object

default_target_labels = ['zenith']

default_prediction_labels = ['zenith_pred', 'zenith_kappa']

nb_inputs = 2

class graphnet.models.task.reconstruction.EnergyReconstruction(*args, **kwargs)

Bases: StandardLearnedTask

Reconstructs energy using stable method.

Construct StandardLearnedTask.

Parameters:

• hidden_size (int) – The number of columns in the output of the last latent layer of Model using this Task. Available through Model.nb_outputs

• args (Any)

• kwargs (Any)

Return type:

object

default_target_labels = ['energy']

default_prediction_labels = ['energy_pred']

nb_inputs = 1

class graphnet.models.task.reconstruction.EnergyReconstructionWithPower(*args, **kwargs)

Bases: StandardLearnedTask

Reconstructs energy.

Construct StandardLearnedTask.

Parameters:

• hidden_size (int) – The number of columns in the output of the last latent layer of Model using this Task. Available through Model.nb_outputs

• args (Any)

• kwargs (Any)

Return type:

object

default_target_labels = ['energy']

default_prediction_labels = ['energy_pred']

nb_inputs = 1

class graphnet.models.task.reconstruction.EnergyTCReconstruction(*args, **kwargs)

Bases: StandardLearnedTask

Reconstructs track and cascade energies using stable method.

Construct StandardLearnedTask.

Parameters:

• hidden_size (int) – The number of columns in the output of the last latent layer of Model using this Task. Available through Model.nb_outputs

• args (Any)

• kwargs (Any)

Return type:

object

default_target_labels = ['energy_track', 'energy_cascade']

default_prediction_labels = ['energy_track_pred', 'energy_cascade_pred']

nb_inputs = 2

class graphnet.models.task.reconstruction.EnergyReconstructionWithUncertainty(*args, **kwargs)

Bases: EnergyReconstruction

Reconstructs energy and associated uncertainty (log(var)).

Construct StandardLearnedTask.

Parameters:

• hidden_size (int) – The number of columns in the output of the last latent layer of Model using this Task. Available through Model.nb_outputs

• args (Any)

• kwargs (Any)

Return type:

object

default_target_labels = ['energy']

default_prediction_labels = ['energy_pred', 'energy_sigma']

nb_inputs = 2

class graphnet.models.task.reconstruction.VertexReconstruction(*args, **kwargs)

Bases: StandardLearnedTask

Reconstructs vertex position and time.

Construct StandardLearnedTask.

Parameters:

• hidden_size (int) – The number of columns in the output of the last latent layer of Model using this Task. Available through Model.nb_outputs

• args (Any)

• kwargs (Any)

Return type:

object

default_target_labels = ['vertex']

default_prediction_labels = ['position_x_pred', 'position_y_pred', 'position_z_pred', 'interaction_time_pred']

nb_inputs = 4

class graphnet.models.task.reconstruction.PositionReconstruction(*args, **kwargs)

Bases: StandardLearnedTask

Reconstructs vertex position.

Construct StandardLearnedTask.

Parameters:

• hidden_size (int) – The number of columns in the output of the last latent layer of Model using this Task. Available through Model.nb_outputs

• args (Any)

• kwargs (Any)

Return type:

object

default_target_labels = ['position']

default_prediction_labels = ['position_x_pred', 'position_y_pred', 'position_z_pred']

nb_inputs = 3

class graphnet.models.task.reconstruction.TimeReconstruction(*args, **kwargs)

Bases: StandardLearnedTask

Reconstructs time.

Construct StandardLearnedTask.

Parameters:

• hidden_size (int) – The number of columns in the output of the last latent layer of Model using this Task. Available through Model.nb_outputs

• args (Any)

• kwargs (Any)

Return type:

object

default_target_labels = ['interaction_time']

default_prediction_labels = ['interaction_time_pred']

nb_inputs = 1

class graphnet.models.task.reconstruction.InelasticityReconstruction(*args, **kwargs)

Bases: StandardLearnedTask

Reconstructs interaction inelasticity.

That is, 1-(track energy / hadronic energy).

Construct StandardLearnedTask.

Parameters:

• hidden_size (int) – The number of columns in the output of the last latent layer of Model using this Task. Available through Model.nb_outputs

• args (Any)

• kwargs (Any)

Return type:

object

default_target_labels = ['inelasticity']

default_prediction_labels = ['inelasticity_pred']

nb_inputs = 1

class graphnet.models.task.reconstruction.VisibleInelasticityReconstruction(*args, **kwargs)

Bases: StandardLearnedTask

Reconstructs interaction visible inelasticity.

That is, 1-(visible track energy / visible hadronic energy).

Construct StandardLearnedTask.

Parameters:

• hidden_size (int) – The number of columns in the output of the last latent layer of Model using this Task. Available through Model.nb_outputs

• args (Any)

• kwargs (Any)

Return type:

object

default_target_labels = ['visible_inelasticity']

default_prediction_labels = ['visible_inelasticity_pred']

nb_inputs = 1