如何使用 H2o 库获取 GBM 模型的数据

Question

我正在使用 GBM 模型，我想与其他机器学习方法进行比较。我跑了5折。据我所知，他们会将数据分成 5 份，并选择其中一份进行测试，另一份进行训练。如何从 H2o lib 的 gbm 中获取 5 折数据？

我用 Python 语言运行它。

folds = 5
cars_gbm = H2OGradientBoostingEstimator(nfolds = folds, seed = 1234)

Answer 1

有两种方法：

1. 您可以手动创建和指定折叠。
2. 您可以要求 H2O 保存折叠索引（对于每一行，它属于哪个折叠 ID？）并通过设置 keep_cross_validation_fold_assignment=True 将它们作为单列数据返回。

以下是一些代码示例：

import h2o
from h2o.estimators import *

h2o.init()

# Import cars dataset
cars = h2o.import_file("https://s3.amazonaws.com/h2o-public-test-data/smalldata/junit/cars_20mpg.csv")
cars["economy_20mpg"] = cars["economy_20mpg"].asfactor()
x = ["displacement","power","weight","acceleration","year"]
y = "economy_20mpg"
nfolds = 5

第一种方式：

# Create a k-fold column and append to the cars dataset
# Or you can use an existing fold id column
cars["fold_id"] = cars.kfold_column(n_folds=nfolds, seed=1)

# Train a GBM
cars_gbm = H2OGradientBoostingEstimator(seed=1, fold_column = "fold_id",
              keep_cross_validation_fold_assignment=True)
cars_gbm.train(x=x, y=y, training_frame=cars)

# View the fold ids (identical to cars["fold_id"])
print(cars_gbm.cross_validation_fold_assignment())

第二种方式：

# Train a GBM & save fold IDs
cars_gbm = H2OGradientBoostingEstimator(seed=1, nfolds=nfolds,
              keep_cross_validation_fold_assignment=True)
cars_gbm.train(x=x, y=y, training_frame=cars)

# View the fold ids
print(cars_gbm.cross_validation_fold_assignment())