kiwi.trainers package¶

Submodules¶

kiwi.trainers.callbacks module¶

class kiwi.trainers.callbacks.Checkpoint(output_dir, checkpoint_save=False, checkpoint_keep_only_best=0, checkpoint_early_stop_patience=0, checkpoint_validation_steps=0)[source]¶

Bases: object

Class for determining whether to evaluate / save the model.

best_iteration_path()[source]¶

best_model_path()[source]¶

best_stats()[source]¶

best_stats_and_path()[source]¶

check_in(trainer, stats, epoch=None, step=None)[source]¶: Saves stat summary and handles checkpoint saving.

check_out()[source]¶

static copy_best_model(model_dir, output_dir)[source]¶

early_stopping()[source]¶

last_model_path()[source]¶: Generates the path where the latest model should be saved.

make_output_path(epoch=None, step=None)[source]¶

must_eval(epoch=None, step=None)[source]¶

must_save_best(stats)[source]¶

push_to_heap(stats, output_path)[source]¶: Push stats and output path to the heap.

remove_snapshot(path_to_remove, event=None)[source]¶: Remove snapshot locally and in MLFlow.

save_latest(trainer, saved_best=False)[source]¶: Saves latest checkpoint of the current model. In case a model was just saved due to being the best validation, saves a pointer instead of the full model. Returns path of saved model.

worst_stats()[source]¶

exception kiwi.trainers.callbacks.EarlyStopException(*args, **kwargs)[source]¶: Bases: StopIteration

kiwi.trainers.linear_word_qe_trainer module¶

class kiwi.trainers.linear_word_qe_trainer.LinearWordQETrainer(model, optimizer_name, regularization_constant, checkpointer)[source]¶

Bases: kiwi.models.linear.linear_trainer.LinearTrainer

model¶

kiwi.trainers.trainer module¶

class kiwi.trainers.trainer.Trainer(model, optimizer, checkpointer, log_interval=100, scheduler=None)[source]¶