Skip to content

AI-900 Labs

To efficiently prepare for the exam through hands-on learning, we can create small projects or practical exercises for each of the main areas covered by the AI-900 exam.

  1. Describe Artificial Intelligence Workloads and Considerations
    • Classify AI vs. Non-AI Solutions
  2. Describe Fundamental Principles of Machine Learning on Azure
    • Build a Basic Predictive Model using Azure Machine Learning
  3. Describe Features of Computer Vision Workloads on Azure
    • Image Classification using Azure Custom Vision
  4. Describe Features of Natural Language Processing (NLP) Workloads on Azure
    • Sentiment Analysis with Azure Text Analytics
  5. Describe Features of Conversational AI Workloads on Azure
    • Build a Simple FAQ Bot using Azure Bot Service and QnA Maker

1. Describe Artificial Intelligence Workloads and Considerations

Project: Classify AI vs. Non-AI Solutions

Objective: To understand the types of AI workloads and identify scenarios where AI can be used effectively.

Steps: 1. Research AI Use Cases: Find examples of AI applications in various industries (e.g., healthcare, finance, retail, etc.). Write a brief description of each use case. 2. Identify AI Components: For each use case, identify the AI components involved (e.g., machine learning, computer vision, natural language processing). 3. Categorize Workloads: Classify the workloads based on the AI solutions: - Knowledge mining - Anomaly detection - Prediction and forecasting - Natural language processing - Computer vision 4. Non-AI vs. AI Solution: Provide a real-world example of a problem that could be solved with a traditional non-AI approach versus an AI approach (e.g., rule-based vs. machine learning for spam detection). 5. Ethical Considerations: List potential ethical implications or considerations for each AI use case, such as bias, privacy, or security concerns.

Tools: Pen and paper, or a document editor like Microsoft Word or Google Docs.

Outcome: Develop a deeper understanding of when and how to apply AI technologies and the ethical considerations associated with AI solutions.

2. Describe Fundamental Principles of Machine Learning on Azure

Project: Build a Basic Predictive Model using Azure Machine Learning

Objective: To create a basic machine learning model to understand the process of data ingestion, model training, and evaluation on Azure.

Steps: 1. Set Up Azure Machine Learning Workspace: - Create an Azure Machine Learning workspace. - Set up the necessary environment, including compute resources.

  1. Data Ingestion:
  2. Use a sample dataset from Azure Open Datasets (e.g., NYC taxi fares dataset).

  3. Explore and clean the data using Azure Machine Learning Designer or Python SDK in Jupyter Notebook.

  4. Model Training:

  5. Use the Azure Machine Learning Designer (drag-and-drop interface) to build a simple regression model to predict taxi fares.
  6. Split the data into training and testing datasets.

  7. Train the model using a linear regression algorithm.

  8. Evaluate the Model:

  9. Evaluate model performance using metrics such as Mean Absolute Error (MAE) and R-squared.

  10. Tune the model by adjusting parameters to improve accuracy.

  11. Deploy the Model:

  12. Deploy the trained model as a web service on Azure.
  13. Test the web service by sending a sample input and receiving the predicted output.

Tools: Azure Machine Learning, Azure Machine Learning Designer, Jupyter Notebook.

Outcome: Gain practical experience in setting up an Azure ML environment, training a basic model, and deploying it as a web service.

3. Describe Features of Computer Vision Workloads on Azure

Project: Image Classification using Azure Custom Vision

Objective: To build, train, and test a custom image classification model using Azure's Custom Vision service.

Steps: 1. Set Up Custom Vision Service: - Create a Custom Vision project in the Azure portal. - Choose a classification project type (e.g., multi-class classification).

  1. Upload and Label Images:
  2. Upload a dataset of labeled images. You can use public datasets or your own set of images (e.g., pictures of different animals or objects).

  3. Ensure a balanced number of images per class.

  4. Train the Model:

  5. Train the image classification model using the labeled images.

  6. Adjust training iterations for optimal performance.

  7. Evaluate Model Performance:

  8. Use the platform’s tools to evaluate model accuracy.

  9. Test the model with new images to see how well it generalizes.

  10. Deploy and Test the Model:

  11. Deploy the model as an endpoint.
  12. Test the endpoint using REST API or the Custom Vision portal by sending a new image and receiving the classification result.

Tools: Azure Custom Vision, Azure Portal, REST API tools.

Outcome: Understand the process of creating, training, evaluating, and deploying a computer vision model using Azure's Custom Vision service.

4. Describe Features of Natural Language Processing (NLP) Workloads on Azure

Project: Sentiment Analysis with Azure Text Analytics

Objective: To use Azure Text Analytics to analyze the sentiment of customer reviews.

Steps: 1. Set Up Text Analytics Resource: - Create a Text Analytics resource in the Azure portal.

  1. Prepare Dataset:
  2. Use a sample dataset of customer reviews (e.g., product reviews from Amazon or Yelp).

  3. Ensure that the dataset is in a text format (CSV or JSON).

  4. Analyze Sentiment:

  5. Use the Azure Text Analytics API to analyze the sentiment of each review in the dataset.

  6. Write a script in Python or use a Jupyter Notebook to send requests to the API and parse the responses.

  7. Visualize Results:

  8. Aggregate the results to see the overall sentiment distribution (positive, neutral, negative).

  9. Use a visualization tool (e.g., Matplotlib or Power BI) to create a bar chart or pie chart of the sentiment analysis results.

  10. Analyze Key Phrases and Entities:

  11. Use the Text Analytics API to extract key phrases and named entities from the reviews.
  12. Identify common themes or trends in customer feedback.

Tools: Azure Text Analytics, Python, Jupyter Notebook, data visualization tools (Matplotlib, Power BI).

Outcome: Learn how to perform sentiment analysis and extract key insights from text data using Azure's NLP capabilities.

5. Describe Features of Conversational AI Workloads on Azure

Project: Build a Simple FAQ Bot using Azure Bot Service and QnA Maker

Objective: To create a basic chatbot that answers frequently asked questions (FAQ) using Azure Bot Service and QnA Maker.

Steps: 1. Set Up QnA Maker Resource: - Create a QnA Maker resource in the Azure portal.

  1. Create a Knowledge Base:
  2. Use a pre-existing FAQ document or create one with questions and answers relevant to a specific topic (e.g., company policies, product information).

  3. Import this document into QnA Maker to create a knowledge base.

  4. Train the QnA Maker:

  5. Train the QnA Maker with the uploaded questions and answers.

  6. Test the knowledge base by asking sample questions to ensure it retrieves correct responses.

  7. Deploy the Knowledge Base:

  8. Publish the knowledge base to make it available for integration with the bot.

  9. Create a Bot using Azure Bot Service:

  10. Use the Azure Bot Service to create a new bot and link it to your QnA Maker knowledge base.

  11. Test the bot using the Web Chat feature in the Azure portal.

  12. Enhance the Bot’s Capabilities:

  13. Add additional intents or use Language Understanding (LUIS) to enhance the bot’s ability to understand different queries.
  14. Integrate the bot with other channels like Microsoft Teams, Slack, or a web application.

Tools: Azure Bot Service, QnA Maker, Azure Portal.

Outcome: Learn how to create and deploy a simple FAQ bot using Azure's Conversational AI services, and understand the basics of bot development and integration.