MLJ中XGBoost回归器调参报错：Unsupported range #2 求助

错误原因 #

报错Unsupported range #2指向第二个超参数max_depth的范围定义问题：你使用了原生Julia的2:10（UnitRange类型），但MLJTuning要求所有超参数范围必须是MLJ提供的Range类型对象，而非原生范围类型。

修正后的代码 #

将max_depth的范围定义替换为MLJ的range函数生成的对象即可，修改后的完整可运行脚本如下：

using MLJ, MLJXGBoostInterface, MLJTuning, Random, XGBoost

# Load the XGBoost Regressor model
XGBoostRegressor = @load XGBoostRegressor 

# Generate synthetic regression dataset
X, y = MLJ.make_regression(200, 5; noise=0.3, rng=1234)
train, test = partition(eachindex(y), 0.8, shuffle=true, rng=1234)

# Define the base model with placeholders for hyperparameters
model = XGBoostRegressor(
    objective="reg:squarederror",
    eta=0.1,        # Learning rate
    max_depth=6,    # Tree depth
    num_round=100,  # Boosting rounds
)

# Define the hyperparameter tuning ranges
tuned_model = TunedModel(
    model=model,
    tuning=RandomSearch(),  
    resampling=CV(nfolds=5, shuffle=true, rng=1234),  
    range=[
        :eta => range(0.01, 0.3, length=5),
        :max_depth => range(2, 10, step=1),  # 替换为MLJ规范的range对象
        :num_round => range(50, 200, length=4)
    ],
    measure=rms,  
    n=10,  
    acceleration=CPUThreads()
);

# Wrap the model into a machine
mach = machine(tuned_model, X, y);

# Train the tuned model
fit!(mach, rows=train);

补充说明 #

MLJTuning的range参数仅接受MLJ定义的范围对象，无论超参数是连续型还是离散型，都需要用range()函数构造。对于整数离散范围，通过step=1可以确保生成的是整数取值。

内容的提问来源于stack exchange，提问作者Andrea