Darkopt API Reference

Integration

class darkopt.ChainerTrigger(score_key, known_best_score, stop_trigger, maximize=False, test_trigger=(5, 'epoch'), pruning_prob_thresh=0.05, learning_curve_predictor=None)

The trigger class for Chainer to prune with learning curve prediction.

info()

Returns trial information (e.g., the number of iterations before pruning).

Returns:A dict that contains the information.

class darkopt.XGBoostCallback(known_best_score, score_key=None, pruning_prob_thresh=0.05, maximize=False, learning_curve_predictor=None, min_iters_before_prune=10, test_interval=10)

The callback class for XGBoost to prune with learning curve prediction.

info()

Returns trial information (e.g., the number of iterations before pruning).

Returns:A dict that contains the information.

Optimization

class darkopt.Optimizer(target_func, param_space, engine='random_search', maximize=False, prune=True)

Hyper-parameter optimizer based on random search algorithm with pruning by learning curve prediction.

optimize(max_n_calls=None)

Invoke the optimization.

Parameters:max_n_calls – The maximum number to call the target_func. Returns:A TrialResult object that describes the best trial.