本文将指导你探索 Azure 机器学习服务的主要功能。在这里,你将学习如何创建、注册并发布模型。此教程旨在让你深入了解 Azure 机器学习的基础知识和常用操作。

关注TechLead,分享AI全维度知识。作者拥有10+年互联网服务架构、AI产品研发经验、团队管理经验,同济本复旦硕,复旦机器人智能实验室成员,阿里云认证的资深架构师,项目管理专业人士,上亿营收AI产品研发负责人。

一、开始之前的准备

要深入 Azure 机器学习,首先确保你有一个工作区。如果你还未设置工作区,那么请按照指引,完成必要的资源配置来搭建你的工作区,并了解其基本操作。

请登录到Azure工作室,并选择你的工作区(如果它还未被激活)。

接下来,在工作区内,你可以选择启动或新建一个笔记本:

  • 如果你打算把代码复制到某个单元,那么请新建一个笔记本。
  • 作为另一种选择,你可以在工作室的“示例”区域找到 tutorials/get-started-notebooks/quickstart.ipynb。打开后点击“克隆”,这样这个笔记本就会被保存到你的“文件”里。

二、配置内核

当你打开笔记本时,可以在顶部的工具栏中找到并设定一个计算实例(前提是你之前还没有设立过)。

如果发现计算实例处于暂停状态,请点击“启动计算”并耐心等待其启动完成。

当出现提示横幅,要求你完成身份验证时,请点击“身份验证”进行操作。

三、建立工作区连接

在开始编写代码之前,我们要确保有办法正确引用工作区。工作区是 Azure 机器学习的核心资源,它为你在 Azure 机器学习上创建的所有项目提供了统一的管理点。

你会为这个工作区连接创建名为 ml_client 的句柄。之后,你可以利用 ml_client 来统筹各种资源和任务。

请在下方的代码单元格里输入你的订阅ID、资源组名以及工作区名。要找到这些信息的方法如下:

  1. 在 Azure 机器学习工作室界面的右上角,点击你的工作区名称。
  2. 从显示的信息中复制工作区、资源组和订阅ID。
  3. 一次复制一个信息,粘贴到代码中后再返回继续复制下一个。

from azure.ai.ml import MLClientfrom azure.identity import DefaultAzureCredential# authenticatecredential = DefaultAzureCredential()# Get a handle to the workspaceml_client = MLClient(credential=credential,subscription_id="",resource_group_name="",workspace_name="",)

四、编写训练代码

首先,我们需要制定训练代码并保存为 Python 文件,命名为 main.py。
开始时,为这个脚本设置一个专门的源代码目录。

import ostrain_src_dir = "./src"os.makedirs(train_src_dir, exist_ok=True)

这段代码负责数据预处理,对数据进行训练和测试的划分。接着,脚本将利用这些数据来培训一个基于树的机器学习模型,并输出该模型。
在整个管道运行过程中,我们会利用 MLFlow 来记录相关参数和性能指标。
接下来的代码单元将使用 IPython magic 命令,把训练脚本保存到你刚刚设定的目录中。

%%writefile {train_src_dir}/main.pyimport osimport argparseimport pandas as pdimport mlflowimport mlflow.sklearnfrom sklearn.ensemble import GradientBoostingClassifierfrom sklearn.metrics import classification_reportfrom sklearn.model_selection import train_test_splitdef main():"""Main function of the script."""# input and output argumentsparser = argparse.ArgumentParser()parser.add_argument("--data", type=str, help="path to input data")parser.add_argument("--test_train_ratio", type=float, required=False, default=0.25)parser.add_argument("--n_estimators", required=False, default=100, type=int)parser.add_argument("--learning_rate", required=False, default=0.1, type=float)parser.add_argument("--registered_model_name", type=str, help="model name")args = parser.parse_args() # Start Loggingmlflow.start_run()# enable autologgingmlflow.sklearn.autolog()#######################################print(" ".join(f"{k}={v}" for k, v in vars(args).items()))print("input data:", args.data)credit_df = pd.read_csv(args.data, header=1, index_col=0)mlflow.log_metric("num_samples", credit_df.shape[0])mlflow.log_metric("num_features", credit_df.shape[1] - 1)train_df, test_df = train_test_split(credit_df,test_size=args.test_train_ratio,)############################################################################### Extracting the label columny_train = train_df.pop("default payment next month")# convert the dataframe values to arrayX_train = train_df.values# Extracting the label columny_test = test_df.pop("default payment next month")# convert the dataframe values to arrayX_test = test_df.valuesprint(f"Training with data of shape {X_train.shape}")clf = GradientBoostingClassifier(n_estimators=args.n_estimators, learning_rate=args.learning_rate)clf.fit(X_train, y_train)y_pred = clf.predict(X_test)print(classification_report(y_test, y_pred))############################################################################################# Registering the model to the workspaceprint("Registering the model via MLFlow")mlflow.sklearn.log_model(sk_model=clf,registered_model_name=args.registered_model_name,artifact_path=args.registered_model_name,)# Saving the model to a filemlflow.sklearn.save_model(sk_model=clf,path=os.path.join(args.registered_model_name, "trained_model"),)######################################################## Stop Loggingmlflow.end_run()if __name__ == "__main__":main()

正如你将在脚本中看到的,一旦模型训练完毕,它会被保存并在工作区中进行注册。这样,这个已注册的模型就可以被用于推理节点了。

为了在“文件”区域看到新创建的文件夹和脚本,你可能需要点击“刷新”按钮。

五、配置计算集群

为训练任务提供弹性处理能力
虽然你已有一个计算实例来执行笔记本操作,但下一步你需要设置一个计算集群,专门用于处理训练任务。不同于计算实例的单节点,计算集群能够支持单节点或多节点的 Linux 或 Windows 操作系统,甚至是特定的计算配置,如 Spark。

此处,你应当预先设置一个 Linux 计算集群。关于虚拟机的规格和价格,你可以查阅相关资料。

对于本例子,你只需简单的集群配置,选择 Standard_DS3_v2,拥有 2 个 vCPU 核心和 7 GB 的 RAM。

from azure.ai.ml.entities import AmlCompute# Name assigned to the compute clustercpu_compute_target = "cpu-cluster"try:# let's see if the compute target already existscpu_cluster = ml_client.compute.get(cpu_compute_target)print(f"You already have a cluster named {cpu_compute_target}, we'll reuse it as is.")except Exception:print("Creating a new cpu compute target...")# Let's create the Azure Machine Learning compute object with the intended parameters# if you run into an out of quota error, change the size to a comparable VM that is available.\# Learn more on https://azure.microsoft.com/en-us/pricing/details/machine-learning/.cpu_cluster = AmlCompute(name=cpu_compute_target,# Azure Machine Learning Compute is the on-demand VM servicetype="amlcompute",# VM Familysize="STANDARD_DS3_V2",# Minimum running nodes when there is no job runningmin_instances=0,# Nodes in clustermax_instances=4,# How many seconds will the node running after the job terminationidle_time_before_scale_down=180,# Dedicated or LowPriority. The latter is cheaper but there is a chance of job terminationtier="Dedicated",)print(f"AMLCompute with name {cpu_cluster.name} will be created, with compute size {cpu_cluster.size}")# Now, we pass the object to MLClient's create_or_update methodcpu_cluster = ml_client.compute.begin_create_or_update(cpu_cluster)

六、命令设置

既然我们已有了执行任务的脚本和对应的计算集群,接下来你将设置一系列的命令行操作,这些操作或直接调用系统命令,或执行特定脚本。

在这一部分,你需要定义输入变量,比如输入数据、数据拆分比例、学习率以及模型的注册名。你的命令脚本将做以下事情:

利用计算集群执行命令。
使用 Azure 机器学习提供的预设环境来运行训练脚本,这些环境内包含了训练脚本所需的软件和运行时库。后续,在其他教程中,你将了解如何自定义这些环境。
设定命令行操作,例如 python main.py。你可以使用 ${{ … }} 这样的语法在命令中传递输入/输出参数。
在这一示例中,我们将直接从互联网获取数据。

from azure.ai.ml import commandfrom azure.ai.ml import Inputregistered_model_name = "credit_defaults_model"job = command(inputs=dict(data=Input(type="uri_file",path="https://azuremlexamples.blob.core.windows.net/datasets/credit_card/default_of_credit_card_clients.csv",),test_train_ratio=0.2,learning_rate=0.25,registered_model_name=registered_model_name,),code="./src/",# location of source codecommand="python main.py --data ${{inputs.data}} --test_train_ratio ${{inputs.test_train_ratio}} --learning_rate ${{inputs.learning_rate}} --registered_model_name ${{inputs.registered_model_name}}",environment="AzureML-sklearn-1.0-ubuntu20.04-py38-cpu@latest",compute="cpu-cluster", #delete this line to use serverless computedisplay_name="credit_default_prediction",)

七、任务提交

现在,你可以在 Azure 机器学习平台上提交一个作业了。这次,你需要对 ml_client 使用 create_or_update 功能。

ml_client.create_or_update(job)

八、查看任务结果并等待完成

你可以通过点击前一个代码单元的输出链接,在 Azure 机器学习工作室里查看任务的进展。

任务的各类输出,比如指标、结果等,都可以在 Azure 机器学习工作室里查看。当任务完成后,其训练出的模型会被注册到你的工作区。

九、部署模型为在线服务

是时候将你的机器学习模型作为一个 Web 服务,部署到 Azure 云上了。
为了部署这个服务,你应当使用已经注册过的机器学习模型。持有一个已经注册过的模型,接下来,你可以着手搭建一个在线端点。需要确保你为端点选择的名称在整个Azure地区是独一无二的。为了确保名字的唯一性,在这个教程里,我们建议采用UUID作为端点名称。

import uuid# Creating a unique name for the endpointonline_endpoint_name = "credit-endpoint-" + str(uuid.uuid4())[:8]```python```python# Expect the endpoint creation to take a few minutesfrom azure.ai.ml.entities import (ManagedOnlineEndpoint,ManagedOnlineDeployment,Model,Environment,)# create an online endpointendpoint = ManagedOnlineEndpoint(name=online_endpoint_name,description="this is an online endpoint",auth_mode="key",tags={"training_dataset": "credit_defaults","model_type": "sklearn.GradientBoostingClassifier",},)endpoint = ml_client.online_endpoints.begin_create_or_update(endpoint).result()print(f"Endpoint {endpoint.name} provisioning state: {endpoint.provisioning_state}")

十、模型部署到终结点

端点构建完毕后,你可以采用入口脚本将模型部署到此端点。值得注意的是,一个端点可以支持多个部署版本,并能够设定特定规则来分流到不同的部署版本。在此,我们会为你创建一个部署版本,它将处理所有的流入流量。对于部署的命名,我们提供了一些建议,如“蓝色”、“绿色”和“红色”,你可以根据自己的喜好选择。

你还可以浏览Azure机器学习工作室的“模型”页面,这有助于你识别已注册模型的最新版本。另外,你也可以利用下面的代码来获取最新的版本信息。

# Let's pick the latest version of the modellatest_model_version = max([int(m.version) for m in ml_client.models.list(name=registered_model_name)])print(f'Latest model is version "{latest_model_version}" ')
# picking the model to deploy. Here we use the latest version of our registered modelmodel = ml_client.models.get(name=registered_model_name, version=latest_model_version)# Expect this deployment to take approximately 6 to 8 minutes.# create an online deployment.# if you run into an out of quota error, change the instance_type to a comparable VM that is available.\# Learn more on https://azure.microsoft.com/en-us/pricing/details/machine-learning/.blue_deployment = ManagedOnlineDeployment(name="blue",endpoint_name=online_endpoint_name,model=model,instance_type="Standard_DS3_v2",instance_count=1,)blue_deployment = ml_client.begin_create_or_update(blue_deployment).result()

十一、实例推理测试

完成模型的部署之后,你现在可以对它进行推理测试了。

按照评分脚本中run函数的要求,制定一个示例请求文件。

deploy_dir = "./deploy"os.makedirs(deploy_dir, exist_ok=True)%%writefile {deploy_dir}/sample-request.json{"input_data": {"columns": [0,1,2,3,4,5,6,7,8,9,10,11,12,13,14,15,16,17,18,19,20,21,22],"index": [0, 1],"data": [[20000,2,2,1,24,2,2,-1,-1,-2,-2,3913,3102,689,0,0,0,0,689,0,0,0,0],[10, 9, 8, 7, 6, 5, 4, 3, 2, 1, 10, 9, 8, 7, 6, 5, 4, 3, 2, 1, 10, 9, 8]]}}# test the blue deployment with some sample dataml_client.online_endpoints.invoke(endpoint_name=online_endpoint_name,request_file="./deploy/sample-request.json",deployment_name="blue",)

十二、节点删除

如果你暂时不需要使用该端点,请记得删除,以避免不必要的费用。在进行删除之前,请确保没有其他部署正在使用这个端点。

ml_client.online_endpoints.begin_delete(name=online_endpoint_name)

十三、停止计算实例

如果你暂时不使用计算实例,建议暂停:
在工作室左侧导航栏,点击“计算”。
选择“计算实例”选项卡。
从列表中选择对应的计算实例。
点击顶部工具栏的“停止”按钮。

十四、资源清理

若你决定不再使用已创建的资源,为避免费用,请进行清理:
在Azure门户里,点击左侧的“资源组”。
从列表中找到并选择你所创建的资源组。
点击“删除资源组”,在弹出的确认框里输入资源组名称,并点击“删除”。

关注TechLead,分享AI全维度知识。作者拥有10+年互联网服务架构、AI产品研发经验、团队管理经验,同济本复旦硕,复旦机器人智能实验室成员,阿里云认证的资深架构师,项目管理专业人士,上亿营收AI产品研发负责人。