seisnn.model.GAN_trainer.GeneratorTrainer

class seisnn.model.GAN_trainer.GeneratorTrainer(database=None, model=<tensorflow.python.keras.engine.functional.Functional object>, optimizer=<tensorflow.python.keras.optimizer_v2.adam.Adam object>, loss=<tensorflow.python.keras.losses.BinaryCrossentropy object>)

Bases: seisnn.model.GAN_trainer.BaseTrainer

Trainer class.

__init__(database=None, model=<tensorflow.python.keras.engine.functional.Functional object>, optimizer=<tensorflow.python.keras.optimizer_v2.adam.Adam object>, loss=<tensorflow.python.keras.losses.BinaryCrossentropy object>)

Initialize the trainer.

Parameters

• database – SQL database.

• model – keras model.

• optimizer – keras optimizer.

• loss – keras loss.

Methods

__init__([database, model, optimizer, loss])	Initialize the trainer.
get_dataset_length([database, tfr_list])
get_model_dir(model_instance[, remove])
train_loop(tfr_list, model_name[, epochs, …])	Main training loop.
train_step(train, val)	Training step.

train_loop(tfr_list, model_name, epochs=1, batch_size=1, log_step=100, plot=False, remove=False)

Main training loop.

Parameters

• tfr_list – List of TFRecord path.

• model_name (str) – Model directory name.

• epochs (int) – Epoch number.

• batch_size (int) – Batch size.

• log_step (int) – Logging step interval.

• plot (bool) – Plot training validation, False save fig, True show fig.

• remove (bool) – If True, remove model folder before training.

Returns

train_step(train, val)

Training step.

Parameters

• train – Training data.

• val – Validation data.

Return type

float

Returns

predict loss, validation loss