optuna 保存训练过程

Optuna提供了多种方法来保存训练过程，以下是其中几种：

1.使用Optuna内置的JSON格式记录器：

import optuna

def objective(trial):
    ...

study = optuna.create_study(study_name='example_study', storage='sqlite:///example.db')
study.optimize(objective, n_trials=100)

# 保存JSON格式文件
optuna.visualization.plot_optimization_history(study).write_image('optimization_history.png')
study.trials_dataframe().to_json('trials.json')

2.使用Optuna内置的CSV格式记录器：

import optuna

def objective(trial):
    ...

study = optuna.create_study(study_name='example_study', storage='sqlite:///example.db', pruner=optuna.pruners.MedianPruner())
study.optimize(objective, n_trials=100)

# 保存CSV格式文件
optuna.visualization.plot_optimization_history(study).write_image('optimization_history.png')
study.trials_dataframe().to_csv('trials.csv')

3.自定义记录器，保存更多的训练信息：

import optuna
import pandas as pd

class CustomRecorder(optuna.study.Study):

    def __init__(self, *args, **kwargs):
        super().__init__(*args, **kwargs)
        self._results = []

    def _log(self, trial_id, values):
        trial = self._storage.get_trial(trial_id)
        self._results.append((trial.datetime_start, trial.value, trial.params, trial.user_attrs, trial.system_attrs))

    def save_results(self, filename):
        df = pd.DataFrame(self._results, columns=['datetime_start', 'value', 'params', 'user_attrs', 'system_attrs'])
        df.to_pickle(filename)

def objective(trial):
    ...

study = optuna.create_study(study_name='example_study', storage='sqlite:///example.db', pruner=optuna.pruners.MedianPruner())
study.optimize(objective, n_trials=100)

# 保存自定义格式文件
recorder = CustomRecorder(study_name='example_study', storage='sqlite:///example.db')
recorder.save_results('results.pkl')