gluonts.nursery.tsbench.src.tsbench.gluonts.callbacks package#

class gluonts.nursery.tsbench.src.tsbench.gluonts.callbacks.Callback[source]#

Bases: object

A stripped-down callback which is focused on batches rather than epochs.

on_network_initialization_end(network: mxnet.gluon.block.HybridBlock) → None[source]#

Hook called once the network is initialized.

on_train_batch_end(network: mxnet.gluon.block.HybridBlock, time_elapsed: float) → None[source]#

Hook called after every training batch.

on_train_start(trainer: mxnet.gluon.trainer.Trainer) → None[source]#

Hook called before training is run.

on_validation_epoch_end(loss: float) → None[source]#

Hook called after every validation epoch.

class gluonts.nursery.tsbench.src.tsbench.gluonts.callbacks.CallbackList(callbacks: List[gluonts.nursery.tsbench.src.tsbench.gluonts.callbacks.base.Callback])[source]#

Bases: gluonts.nursery.tsbench.src.tsbench.gluonts.callbacks.base.Callback

Wrapper class for a list of callbacks.

on_network_initialization_end(network: mxnet.gluon.block.HybridBlock) → None[source]#

Hook called once the network is initialized.

on_train_batch_end(network: mxnet.gluon.block.HybridBlock, time_elapsed: float) → None[source]#

Hook called after every training batch.

on_train_start(trainer: mxnet.gluon.trainer.Trainer) → None[source]#

Hook called before training is run.

on_validation_epoch_end(loss: float) → None[source]#

Hook called after every validation epoch.

class gluonts.nursery.tsbench.src.tsbench.gluonts.callbacks.LearningRateScheduleCallback(milestones: List[float], decay: float = 0.5)[source]#

Bases: gluonts.nursery.tsbench.src.tsbench.gluonts.callbacks.base.Callback

The learning rate schedule callback decreases the learning rate by a predefined factor after each of the provided milestones (after x seconds during training).

on_train_batch_end(network: mxnet.gluon.block.HybridBlock, time_elapsed: float) → None[source]#

Hook called after every training batch.

on_train_start(trainer: mxnet.gluon.trainer.Trainer) → None[source]#

Hook called before training is run.

class gluonts.nursery.tsbench.src.tsbench.gluonts.callbacks.ModelSaverCallback(directory: pathlib.Path, milestones: List[float])[source]#

Bases: gluonts.nursery.tsbench.src.tsbench.gluonts.callbacks.base.Callback

The model saver callback saves the model during training at exponential frequency.

network#

The network that was trained. Not available prior to training.

saved_parameters#

The parameters saved for the different milestones. Should only be accessed after training has finished and should not be modified.

training_times#

The training times in seconds for the different milestones.

num_gradient_updates#

The number of gradient updates for the different milestones.

on_network_initialization_end(network: mxnet.gluon.block.HybridBlock) → None[source]#

Hook called once the network is initialized.

on_train_batch_end(network: mxnet.gluon.block.HybridBlock, time_elapsed: float) → None[source]#

Hook called after every training batch.

on_train_start(trainer: mxnet.gluon.trainer.Trainer) → None[source]#

Hook called before training is run.

class gluonts.nursery.tsbench.src.tsbench.gluonts.callbacks.ParameterCountCallback[source]#

Bases: gluonts.nursery.tsbench.src.tsbench.gluonts.callbacks.base.Callback

This callback allows counting model parameters during training.

num_parameters#

The number of parameters of the model. This attribute should only be accessed after training.

on_network_initialization_end(network: mxnet.gluon.block.HybridBlock) → None[source]#

Hook called once the network is initialized.