Predicting Stock Price using Azure Automated Machine Learning (AutoML) in few clicks (2023)

In a typical machine learning process, we first go through data featurization activities, including missing and imbalanced data handling exercises. We then select an appropriate algorithm to handle the business problems we are trying to solve through the machine learning model, followed by a hyper parameter tuning exercise to choose an optimal set of parameters for our model. The process is complex, time consuming and would often require data scientist and data analytics expertise, even before we can decide if the exercise is reasonable business project to take it. This is where Automated Machine Learning can help a big way.

Automated machine learning, also referred to as automated ML or AutoML, is the process of automating the time-consuming, iterative tasks of machine learning model development. It allows data scientists, analysts, and developers to build ML models with high scale, efficiency, and productivity, all while sustaining model quality.

In this walkthrough, we will explore how easy it is to take the historical stock price data and make predictions on the stock price through Azure Automated Machine Learning (AutoML), following low code, no-code approach, with few clicks and without much data scientist knowledge to spare.

Step 1: Create Data Asset

To simplify the discussion, we will assume the stock data from yahoo finance is already downloaded to our local machine. The details of data ingestion are discussedLearnAI/data-generator.md at main · msdpalam/LearnAI (github.com).

We will also assume that Azure ML workspace is already created.

1. Sign in to Azure Machine Learning studio.
2. Select your subscription and workspace.
3. From the Assets section, select Data

4. Provide a name for your dataset in the appeared dialog box, keep the Dataset type as Tabularsince our stock is in tabular format

5. Click Browsein the dialog box appeared and select Browse files

6. Navigate to the folder where you downloaded stock data, select the csv file:

7. Here is how it would look after you select the data file:

Note: The file will be uploaded to the selected data store.

Click Next

8. The dialog box appeared shows the settings and the preview of the data asset.

Click Next

9. The next screen shows the schema for the data asset

10. In the next screen, confirm the details of the data asset.

Click Create.

Please note the data asset name and the location (blob store where it is uploaded)

Clicking on the workspaceblobstore, you can find the stock_data.csv file, in the following location of your default blob data store

Step 2: Create New Automated ML Job

Now that we have the data asset, let us create an Automated ML job in Azure ML Studio

1. Sign in to Azure Machine Learning studio.
2. Select your subscription and the workspace you created.
3. In the left pane, select Automated MLunder the Authorsection.
4. Select +New automated ML job.

5. From the appeared dialog box, select the data asset we created in Step 1, as shown below:

If you click on the data asset link, it will show you preview.

Click Next

6. In the Configure Jobdialog box,
   1. Note that the data asset is already selected
   2. Create Newor use Select existingexperiment name for the job.
   3. Select the Target Columnfrom the drop-down list of columns
   4. Select Compute Type(default is compute cluster), for the job
   5. Select an existing Compute Cluster (if you do not already have one, create a cluster, following the articleCreate compute clusters - Azure Machine Learning | Microsoft Learn.

Click Next

8. Under Select task and settingsdialog
   1. SelectRegression, as it fits the characteristics of an algorithm for the stock price prediction

b. Under View additional configuration setting(click the link to open the dialog), select Primary metric (default is set to Normalized root mean squared error), make sure explain model check box selected and if you want to explore all the models, select Use all supported modelscheck box. You can then block the models you do not want to try out, from the drop-down menu.

c. You can also click on the link view featurization settings, to explore the current featurization settings and adjust as needed

Click Next

d. For the Hyperparameter option keep the default

Click Finish

10. A new AutoML job will be created with Not Startedstate, which soon will be in the Runningstate.

11. Since we configured the Training job timeto 0.5 hours, it would finish within the next half an hour
12. Once the job completes, we do see that the job indeed completed in about 27 minutes

Step 3: Explore the best model

Upon completion of the AutoML job, we can explore the models trained and look for the best model to deploy as an endpoint to provide that model’s inference capabilities. Let us review the models.

1. Click on the Modelstab from the completed job dialog

2. We can see the VotingEnsemblealgorithm was the best model, based on the Normalized Root Means Squared Error metric. It is important to note that AutoML, automatically tried algorithms, based on the configuration, without us writing a single line of code, within a very short period of time (~27minutes)

Step 4: Deploy the best model for inferencing to an endpoint.

An endpoint is an HTTPS endpoint that clients can call to receive the inferencing (scoring) output of a trained model. A single endpoint can contain multiple deployments and deployment is a set of resources required for hosting the model that does the actual inferencing.

1. Click on the VotingEnsembleAlgorithm link from the Modelsdialog, to investigate the details of the model

While we can explore the model Metrics, Experiment details, Duration for training, Inputs etc. to complete our exercise, we will deploy the model and score against the model with new set of data.

2. Click on Deployand select Deploy to a web service. How to deploy an AutoML model to an online endpoint, has another option of deployment. For details on online endpoints, refer to What are Azure Machine Learning endpoints?

3. We will choose Container Instance as the web service deployment option for this example; however, Azure Kubernetes Service is a deployment option we often find.

4. Click Deployto start the deployment. The model deployment will be triggered

5. Once the deployment completes, the deployment state will show Healthy, a REST endpoint will be created and a Swagger URI will be created, that will show how we can interact with the endpoint. We will skip exploring the Swagger URI for thiswalkthrough.

Step 5: Scoring (inferencing) against the deployed model.

Scoring (or inferencing) is the process of running live data points into a machine learning algorithm (or ML model) to calculate an output. Now that the ML model is deployed, there are multiple ways you can inference an output against the deployed ML model. The simplest option is to use the Test feature in the ML Studio UI for the endpoint. Simply click on the Test Tab and provide the test data set to see the results (predicted close of the MSFT Stock)

Please note that when you first click on the Test tab, it will only show the sample data format, as below:

{ "Inputs": { "data": [ { "Datetime": "2000-01-01T00:00:00.000Z", "Open": 0.0, "High": 0.0, "Low": 0.0, "Adj Close": 0.0, "Volume": 0 } ] }, "GlobalParameters": 0.0}

In this test, we provided one data point for the MSFT stock price as below, picking up the first row from our stock_data.sv file, however we can provide multiple data points, following the JSON format of the data.

{ "Inputs": { "data": [ { "Datetime": "2022-08-01 09:30:00-04:00", "Open": 277.82000732421875, "High": 277.9399108886719, "Low": 277.07000732421875, "Adj Close": 277.67999267578125, "Volume": 731044 } ] }, "GlobalParameters": 0.0}

As we observe, using AutoML, it is very easy to take a prepared dataset and apply the algorithm of your choice to create and deploy a machine learning model that predicts the stock price, without writing a single line of code.

In part 2, we will walk through the stock price prediction exercise through Azure Machine Learning designer, yet another low code, no-code approach to perform machine learning tasks, leveraging Azure Machine Learning.