synthval.configs

Module for defining default parameters for network training and testing using PyTorch and pyNeVer.

This module provides default configurations for setting up neural networks and training them using the PyTorch framework, with an emphasis on integrating with pyNeVer strategies.

synthval.configs.DEFAULT_NETWORK_PARAMS

Default parameters for the neural network architecture. The dictionary contains:

• network_idstr

  Identifier for the network architecture.

• num_hidden_neuronslist of int

  Number of neurons for each hidden layer in the network.

Type:

dict

synthval.configs.DEFAULT_TRAINING_PARAMS

Default parameters for training the network using PyTorch. The dictionary contains:

• optimizer_contorch.optim.Optimizer

  Constructor for the optimizer to be used during training.

• opt_paramsdict

  Parameters for the optimizer, such as learning rate (lr).

• n_epochsint

  Number of epochs for training.

• validation_percentagefloat

  Proportion of the data used for validation.

• train_batch_sizeint

  Batch size used for training.

• validation_batch_sizeint

  Batch size used for validation.

• r_splitbool

  Whether to perform a random data split for training/validation.

• scheduler_contorch.optim.lr_scheduler, optional

  Constructor for a learning rate scheduler (default is None).

• sch_paramsdict or None

  Parameters for the learning rate scheduler (default is None).

• precision_metricpynever.strategies.training.PytorchMetrics

  Metric to evaluate training precision, e.g., accuracy or inaccuracy.

• network_transformcallable, optional

  Transformation function to apply to the network (default is None).

• devicestr

  The device on which the training will be performed (e.g., ‘cpu’ or ‘cuda’).

• train_patienceint

  Number of epochs to wait before early stopping if no improvement in validation loss.

• checkpoints_rootstr

  Path to the directory where model checkpoints will be saved.

• verbose_rateint

  Frequency (in epochs) for printing training progress.

Type:

dict

synthval.configs.DEFAULT_TESTING_PARAMS

Default parameters for testing the trained network. The dictionary contains:

• metricpynever.strategies.training.PytorchMetrics

  Metric used to evaluate the model performance on the test set.

• metric_paramsdict

  Additional parameters for the testing metric.

• test_batch_sizeint

  Batch size for testing.

• devicestr

  The device on which testing will be performed (e.g., ‘cpu’ or ‘cuda’).

Type:

dict