multiml.agent.keras package

Submodules

• multiml.agent.keras.callback module
  • AlphaDumperCallback
    • AlphaDumperCallback.alpha_history
    • AlphaDumperCallback.model
    • AlphaDumperCallback.__init__()
    • AlphaDumperCallback.formatting()
    • AlphaDumperCallback.on_train_begin()
    • AlphaDumperCallback.on_epoch_end()
    • AlphaDumperCallback.on_train_end()
    • AlphaDumperCallback.get_alpha_history()
  • EachLossDumperCallback
    • EachLossDumperCallback.loss_history
    • EachLossDumperCallback.model
    • EachLossDumperCallback.__init__()
    • EachLossDumperCallback.on_epoch_end()
    • EachLossDumperCallback.get_loss_history()
  • NaNKillerCallback
    • NaNKillerCallback.on_epoch_end()
• multiml.agent.keras.keras_connection_grid_search module
  • KerasConnectionGridSearchAgent
• multiml.agent.keras.keras_connection_random_search module
  • KerasConnectionRandomSearchAgent
    • KerasConnectionRandomSearchAgent.ModelConnectionTask
      • KerasConnectionRandomSearchAgent.ModelConnectionTask.build_model()
• multiml.agent.keras.keras_darts module
  • KerasDartsAgent
    • KerasDartsAgent.__init__()
• multiml.agent.keras.keras_ensemble module
  • KerasEnsembleAgent
    • KerasEnsembleAgent.__init__()
    • KerasEnsembleAgent.execute()

Module contents

class multiml.agent.keras.KerasConnectionRandomSearchAgent(freeze_model_weights=False, do_pretraining=True, connectiontask_name=None, connectiontask_args={}, **kwargs)

Bases: ConnectionRandomSearchAgent

Keras implementation for ConnectionRandomSearchAgent.

class ModelConnectionTask(subtasks, loss_weights=None, variable_mapping=None, **kwargs)

Bases: ModelConnectionTask, KerasBaseTask

Keras implementation of ModelConnectionTask.

build_model()

Build model.

class multiml.agent.keras.KerasEnsembleAgent(ensembletask_args={}, **kwargs)

Bases: KerasConnectionRandomSearchAgent

Agent packing subtasks using Keras EnsembelModel.

__init__(ensembletask_args={}, **kwargs)

Parameters:

• ensembletask_args (dict) – args for EnsembleTask

• **kwargs – Arbitrary keyword arguments

execute()

Execute.

class multiml.agent.keras.KerasDartsAgent(select_one_models=True, use_original_darts_optimization=True, dartstask_args={}, **kwargs)

Bases: KerasEnsembleAgent

Model selection agent inspired by Differential Architecture Search (DARTS)

__init__(select_one_models=True, use_original_darts_optimization=True, dartstask_args={}, **kwargs)

Parameters:

• select_one_models (bool) – (Expert option) If false, all models are kept for the final prediction

• use_original_darts_optimization (bool) – (Expert option) If false, both model parameter and alpha parameters are simultaneously trained.

• dartstask_args (dict) – args for DARTSTask

• **kwargs – Arbitrary keyword arguments