用pickle-python序列化一个h2o模型

Question

我目前正在尝试将 h2o gb 模型序列化为 pickle 对象并重用它。由于一些限制，我不能使用http://docs.h2o.ai/h2o/latest-stable/h2o-docs/productionizing.html 给出的默认方法或 POJO 和 MOJO。模型被腌制，但是在 unpickling(pickle.loads) 时，出现以下错误 -

__new__() missing 1 required positional argument: 'keyvals'

以下代码供参考-

import h2o as h2o
import pickle as pickle
from h2o.estimators.gbm import H2OGradientBoostingEstimator
h2o.init()

csv_url = "https://h2o-public-test-data.s3.amazonaws.com/smalldata/wisc/wisc-diag-breast-cancer-shuffled.csv"
data = h2o.import_file(csv_url)
y = 'diagnosis'
x = data.columns
del x[0:1]
train, test = data.split_frame(ratios=[0.75], seed=1)


model = H2OGradientBoostingEstimator(distribution='bernoulli',
                                ntrees=100,
                                max_depth=4,
                                learn_rate=0.1)
model.train(x=x, y=y, training_frame=train, validation_frame=test)

loaded_model = pickle.loads(saved_model)
perf = loaded_model.model_performance(test)
perf.auc()

我试图了解 pickle 模块并进行一些更改，但没有成功。任何解决方法/帮助将不胜感激。谢谢。

Answer 1

你不用pickle，h2o提供了自己的持久化方法：http://docs.h2o.ai/h2o/latest-stable/h2o-docs/save-and-load-model.html

# build the model
model = H2ODeepLearningEstimator(params)
model.train(params)

# save the model
model_path = h2o.save_model(model=model, path="/tmp/mymodel", force=True)

print(model_path)
# outputs: /tmp/mymodel/DeepLearning_model_python_1441838096933

# load the model
saved_model = h2o.load_model(model_path)

但请注意，即使是最细微的版本更改，持久化模型也不兼容，即如果您在 3.18.0.1 中训练并保存模型，您将不会能够在 3.18.0.2 中加载它