ncem.estimators.EstimatorInteractions.train

EstimatorInteractions.train(epochs: int = 1000, epochs_warmup: int = 0, max_steps_per_epoch: Optional[int] = 20, batch_size: int = 16, validation_batch_size: int = 16, max_validation_steps: Optional[int] = 10, shuffle_buffer_size: Optional[int] = 10000, patience: int = 20, lr_schedule_min_lr: float = 1e-05, lr_schedule_factor: float = 0.2, lr_schedule_patience: int = 5, initial_epoch: int = 0, monitor_partition: str = 'val', monitor_metric: str = 'loss', log_dir: Optional[str] = None, callbacks: Optional[list] = None, early_stopping: bool = True, reduce_lr_plateau: bool = True, pretrain_decoder: bool = False, decoder_epochs: int = 1000, decoder_patience: int = 20, decoder_callbacks: Optional[list] = None, aggressive: bool = False, aggressive_enc_patience: int = 10, aggressive_epochs: int = 5, seed: int = 1234, **kwargs)

Train model.

Use validation loss and maximum number of epochs as termination criteria.

Parameters

• epochs (int) – Integer number of times to iterate over the training data arrays. If unspecified, it will default to 1000.

• epochs_warmup (int) – Integer number of times to iterate over the training data arrays in warm up (without early stopping). If unspecified, it will default to 0.

• max_steps_per_epoch (int, optional) – Maximal steps per epoch. If unspecified, it will default to 20.

• batch_size (int) – Number of samples per gradient update. If unspecified, it will default to 16.

• validation_batch_size (int) – Number of samples in validation. If unspecified, it will default to 16.

• max_validation_steps (int) – Maximal steps per validation. If unspecified, it will default to 10.

• shuffle_buffer_size (int, optional) – Shuffle buffer size. If unspecified, it will default to 1e4.

• patience (int) – Number of epochs with no improvement. If unspecified, it will default to 20.

• lr_schedule_min_lr (float) – Lower bound on the learning rate. If unspecified, it will default to 1e-5.

• lr_schedule_factor (float) – Factor by which the learning rate will be reduced. new_lr = lr * factor. If unspecified, it will default to 0.2.

• lr_schedule_patience (int) – Number of epochs with no improvement after which learning rate will be reduced. If unspecified, it will default to 5.

• initial_epoch (int) – Epoch at which to start training (useful for resuming a previous training run). If unspecified, it will default to 0.

• monitor_partition (str) – Monitor partition.

• monitor_metric (str) – Monitor metric.

• log_dir (str, optional) – Logging directory.

• callbacks (list, optional) – List of callbacks to be called during training.

• early_stopping (bool) – Whether to activate early stopping.

• reduce_lr_plateau (bool) – Whether to reduce learning rate on plateau.

• pretrain_decoder (bool) – Whether to pretrain the decoder model.

• decoder_epochs (int) – Integer number of times to iterate over the training data arrays in decoder pretraining. If unspecified, it will default to 1000.

• decoder_patience (int) – Number of epochs with no improvement in decoder pretraining. If unspecified, it will default to 20.

• decoder_callbacks (list, optional) – List of callbacks to be called during decoder pretraining.

• aggressive (bool) – Whether to train aggressive.

• aggressive_enc_patience (int) – Number of epochs with no improvement in aggressive training. If unspecified, it will default to 10.

• aggressive_epochs (int) – Integer number of times to iterate over the training data arrays in aggressive training. If unspecified, it will default to 5.

• seed (int) – Random seed for reproduability.

• kwargs – Arbitrary keyword arguments.