sad.task.training package

Submodules

sad.task.training.training module

class TrainingTask(config: Dict, input_dir: Optional[str] = None, output_dir: Optional[str] = None)[source]

Bases: sad.task.base.TaskBase

A concrete task class that will be responsible to train a model.

This class inherits all existing properties in sed.task.base.TaskBase.

create_generator(model: sad.model.base.ModelBase) → sad.generator.base.GeneratorBase[source]

Instance method to create a generator for training.

Parameters

model (sad.model.ModelBase) – An instance of model that will be associated with the generator.

Returns

An instance of generator class that will be used to train the model in a trainer, an instance of sad.trainer.TrainerBase.

Return type

sad.generator.GeneratorBase

Raises

RuntimeError – When a generator instance is not able to create from configuration in self.generator_config.

create_model() → sad.model.base.ModelBase[source]

Instance method to initialize a model for training.

Returns

An instance of model class that will be trained in current task.

Return type

sad.model.ModelBase

Raises

RuntimeError – When a model instance is not able to initialize from configuration in self.model_config.

create_trainer(model: sad.model.base.ModelBase, generator: sad.generator.base.GeneratorBase) → sad.trainer.base.TrainerBase[source]

Instance method to create a trainer for training. Require an instance of sad.model.ModelBase and a sad.generator.GeneratorBase.

Parameters

  • model (sad.model.ModelBase) – An instance of model that will be associated with the trainer.

  • generator (sad.generator.GeneratorBase) – An instance of generator that will be used by trainer.

Returns

An instance of trainer class that will be used in current task.

Return type

sad.trainer.TrainerBase

Raises

RuntimeError – When a trainer instance is not able to initialize from configuration in self.trainer_config.

property filename: str

A relative path pointing to where user-item interaction data are located. The path is relative to self.input_dir.

property generator_config: Dict

A dictionary read from configuration of the task. It specifies the configuration to initialize a generator of type sad.generator.GeneratorBase. Will read directly from "generator" field from self.config.

An example is shown below:

name: ImplicitFeedbackGenerator
spec:
  u_batch: 50
  i_batch: 100
property model_config: Dict

A dictionary read from configuration of the task. It specifies the configuration to initialize a model of type sad.model.ModelBase. Will read directly from "model" field from self.config.

An example is shown below:

name: SADModel
spec:
  n: 200
  m: 500
  k: 100
property model_id: str

A string that uniquely identifies a trained model. It is usually set to "model_{self.task_id}".

run()[source]

Run training task.

property trainer_config: Dict

A dictionary read from configuration of the task. It specifies the configuration to initialize a trainer of type sad.trainer.TrainerBase. Will read directly from "trainer" field from self.config.

An example is shown below:

name: SGDTrainer
spec:
  n_iters: 50
  u_idxs: [0, 1, 2, 3, 4, 5]
  w_l1: 0.01
  w_l2: 0.01

  callbacks:
  - name: "MetricsLoggingCallback"
    spec:
      every_iter: 1
      every_step: 2
run_task(config_file: Optional[str] = None)[source]

Main function that will be called when running a processing task.

Parameters

config_file (str) – A yml file that contains configurations for running the processing task. Optional, when None a default file at ./ppgflow/tasks/processing/config.yml will be used.

Module contents