hyperparameter-tuning

# 使用 PyTorch Lightning 进行超参数调优 ## 与调优框架的集成 Lightning 与流行的超参数调优库无缝集成。 ### 1. Ray Tune 集成 **安装**: bash pip install ray[tune] pip install lightning **Ray Tune 基础示例**: python import lightning as L from ray import tune from ray.tune.integration.pytorch_lightning import TuneReportCallback class LitModel(L.LightningModule): def __init__(self, lr, batch_size): super().__init__() self.lr = lr self.batch_size = batch_size self.model = nn.Sequential(nn.Linear(10, 128), nn.ReLU(), nn.Linear(128, 1)) def training_step(self, batch, batch_idx): loss = self.model(batch).mean() self.log('train_loss', loss) return loss def validation_step(self, batch, batch_idx): val_loss = self.model(batch).mean() self.log('val_loss', val_loss) def configure_optimizers(self): return torch.optim.Adam(self.parameters(), lr=self.lr) def train_fn(config): """Ray Tune 的训练函数。""" model = LitModel(lr=config["lr"], batch_size=config["batch_size"]) # 添加回调以向 Tune 报告指标 trainer = L.Trainer( max_epochs=10, callbacks=[TuneReportCallback({"loss": "val_loss"}, on="validation_end")] ) trainer.fit(model, train_loader, val_loader) # 定义搜索空间 config = { "lr": tune.loguniform(1e-5, 1e-1), "batch_size": tune.choice([16, 32, 64, 128]) } # 运行超参数搜索 analysis = tune.run( train_fn, config=config, num_samples=20, # 20 次试验 resources_per_trial={"gpu": 1} ) # 最佳超参数 best_config = analysis.get_best_config(metric="loss", mode="min") print(f"最佳配置: {best_config}") **高级：基于种群的训练 (PBT)**: python from ray.tune.schedulers import PopulationBasedTraining # PBT 调度器 scheduler = PopulationBasedTraining( time_attr='training_iteration', metric='val_loss', mode='min', perturbation_interval=5, # 每 5 个 epoch 扰动一次 hyperparam_mutations={ "lr": tune.loguniform(1e-5, 1e-1), "batch_size": [16, 32, 64, 128] } ) analysis = tune.run( train_fn, config=config, num_samples=8, # 种群大小 scheduler=scheduler, resources_per_trial={"gpu": 1} ) ### 2. Optuna 集成 **安装**: bash pip install optuna pip install optuna-integratio