Hyperopt Spark 3.0 问题

Question

我正在运行运行时 8.1（包括 Apache Spark 3.1.1、Scala 2.12），试图让 hyperopt 按定义工作

https://docs.databricks.com/applications/machine-learning/automl-hyperparam-tuning/hyperopt-spark-mlflow-integration.html ​

py4j.Py4JException: Method maxNumConcurrentTasks([]) does not exist

当我尝试

spark_trials = SparkTrials()

我需要做什么特别的事情才能使其正常工作吗？

这是我正在使用的集群

{
    "autoscale": {
        "min_workers": 1,
        "max_workers": 2
    },
    "cluster_name": "mlops_tiny_ml",
    "spark_version": "8.2.x-cpu-ml-scala2.12",
    "spark_conf": {},
    "aws_attributes": {
        "first_on_demand": 1,
        "availability": "SPOT_WITH_FALLBACK",
        "zone_id": "us-west-2b",
        "instance_profile_arn": "arn:aws:iam::112437402463:instance-profile/databricks_instance_role_s3",
        "spot_bid_price_percent": 100,
        "ebs_volume_type": "GENERAL_PURPOSE_SSD",
        "ebs_volume_count": 3,
        "ebs_volume_size": 100
    },
    "node_type_id": "m4.large",
    "driver_node_type_id": "m4.large",
    "ssh_public_keys": [],
    "custom_tags": {},
    "spark_env_vars": {},
    "autotermination_minutes": 120,
    "enable_elastic_disk": false,
    "cluster_source": "UI",
    "init_scripts": [],
    "cluster_id": "0xxxxxt404"
}

这是我正在使用的代码 https://docs.databricks.com/applications/machine-learning/automl-hyperparam-tuning/hyperopt-model-selection.html

Answer 1

Hyperopt 仅包含在 DBR ML 运行时中，而不包含在库存运行时中。您可以通过查看每个运行时的发行说明来检查它：DBR 8.1与DBR 8.1 ML。

从文档中：

用于机器学习的 Databricks Runtime结合了 MLflow 和 Hyperopt 这两个开源工具，可以自动执行模型选择和超参数调整的过程。