This article explores how to train and test a knowledge base in light of the AI-102 Designing and Implementing a Microsoft Azure AI Solution.
Preparing the Knowledge Base
First things first, you need to create the knowledge base for your AI model. Azure uses the QnA Maker to create a cloud-based service implementing a conversational layer. Pre-existing data from social media, documents, websites, and other sources can aid in building the knowledge base.
To create a knowledge base using QnA Maker, follow these steps:
- Sign in to the QnA Maker portal
- Click ‘Create a knowledge base’
- If you don’t have an existing Azure resource, click on ‘Create a QnA resource,’ which will redirect you to Azure portal.
- After creating a QnA resource in the Azure portal, move back to the QnA Maker portal, and under Step 2, click ‘Refresh’ to use that resource.
- Concatenate or edit the name of your knowledge base under Step 3.
- In Step 4, populate your knowledge base.
- Save and train your knowledge base model using the ‘Save and train’ button.
Training and Testing the Knowledge Base
Training incorporates teaching the AI model essential information while testing validates the model’s performance. In Azure, we use a LUIS app for training and testing. LUIS (Language Understanding) trains the application to understand language intents based on given data.
To train and test the knowledge base using the LUIS app, follow these steps:
- Log on to the LUIS portal.
- Import your knowledge base (JSON file). Select ‘My Apps,’ then ‘Import existing resource.’
- Train the model using the ‘Train’ button located on the top right.
- Next, test the model. Click on ‘Test’ on the panel’s top right. Type a query and press Enter.
The model is now ready for publishing.
Understand LUIS App Results
Understanding LUIS app results is crucial for improving the model.
- Utterance: This is the user’s input or query. For example, “What are the store hours?”
- Intent: The user’s underlying intention. For example, “FindStoreHours.”
- Entity: Important information contained in the utterance. For example, “store hours.”
The format of the result will resemble the following JSON object:
{
"query": "What are the store hours?",
"topScoringIntent": {
"intent": "FindStoreHours",
"score": 0.9612575
},
"entities": []
}
Remember, every AI venture needs a clear understanding of its knowledge base. You should streamline your model to cater to your unique requirements. Training and testing present opportunities for improving the efficiency of the model while evaluating its performance. Keep track of your LUIS app results to enhance the AI model continually. In AI-102 Designing and Implementing a Microsoft Azure AI Solution, these concepts of preparing the knowledge base, training, and testing are significant and indispensable.
Practice Test
True/False: The training data is used to teach the model about the problem it needs to solve.
- True
Answer: True
Explanation: Training data is necessary for the learning process, allowing the model to gradually improve its performance by learning from mistakes.
Multiple select: Which of the following can be done to test the performance of an AI model?
- A. Implement A/B Testing
- B. Use Cross-Validation
- C. Utilize a Validation Set
Answer: A, B, and C
Explanation: All methods – A/B testing, Cross-Validation, and using a validation set are used to evaluate the performance of an AI model.
Single select: What is the final step after training and testing a knowledge base?
- A. Validation
- B. Deployment
- C. Plenty of rest
Answer: B. Deployment
Explanation: After the model has been trained and tested, it can now be deployed and integrated into the system for actual use.
True/False: The test data should be a subset of the training data.
- False
Answer: False
Explanation: The test data should be separate from the training data to avoid overfitting and to be able to correctly evaluate the performance of the model.
Multiple select: What are some common methods to optimize a trained model?
- A. Hyperparameter Tuning
- B. Feature Engineering
- C. Random Sampling
Answer: A and B
Explanation: Hyperparameter Tuning and Feature Engineering are methods used to improve the performance of a trained AI model, Random Sampling is not.
Single select: What is the purpose of using a validation set?
- A. To fine-tune the model’s hyperparameters
- B. To train the model
- C. To collect more data
Answer: A. To fine-tune the model’s hyperparameters
Explanation: A validation set is used to avoid overfitting of the model during the hyperparameter tuning process.
True/False: It is necessary to retrain a model from time to time.
- True
Answer: True
Explanation: Retraining a model is essential to keep it updated and maintain its performance, especially when there are significant changes in the data.
Single select: Which of the following is not a good practice when training an AI model?
- A. Balancing the data
- B. Using all available data for training
- C. Regularly checking for overfitting
Answer: B. Using all available data for training
Explanation: It’s crucial to divide the dataset into a training set, a validation set, and a test set, not to use all data for training.
Multiple select: What are the key steps in building a knowledge base in AI?
- A. Gathering Data
- B. Preprocessing Data
- C. Data Visualization
Answer: A and B
Explanation: Building a knowledge base involves gathering and preprocessing data. Data visualization, while useful, is not a key step in building the knowledge base.
True/False: The testing phase is crucial to evaluate the performance of a model and identify where improvements can be made.
- True
Answer: True
Explanation: The testing phase is when the model is evaluated on the unseen data, making it crucial for gauging its true performance.
Single select: Which algorithm is used to reduce high dimensionality in your dataset when training the AI model?
- A. Linear Regression
- B. Dimension Reduction
- C. Random Forest
Answer: B. Dimension Reduction
Explanation: For high dimensional data, the dimension reduction algorithm such as PCA (Principal Component Analysis) is used. It helps to reduce the number of input variables.
True/False: Overfitting is the case when the model learns too well from the training data, and as a result, doesn’t perform well on unseen data.
- True
Answer: True
Explanation: Overfitting occurs when the model performs exceptionally well on the training data but fails to generalize to new, unseen data.
Single select: Why is it crucial to split your dataset into training and testing sets while training an AI model?
- A. To avoid overfitting
- B. To gather more data
- C. To incorporate as much data as possible into the model
Answer: A. To avoid overfitting
Explanation: Dividing the dataset into training and testing sets, helps to avoid overfitting. It ensures that the model tests on unseen data.
Multiple select: What are the key components of an AI solution in Azure?
- A. Azure Machine Learning
- B. Azure Bot Service
- C. Azure Storage
Answer: A, B and C
Explanation: All the mentioned components – Azure Machine Learning, Azure Bot Service, and Azure Storage are key to an AI solution in Azure.
True/False: You can use Azure ML studio to design and implement a Microsoft Azure AI Solution.
- True
Answer: True
Explanation: Azure ML studio is a Microsoft Azure tool that provides an interactive, visual workspace to build and test AI models.
Interview Questions
What is a Knowledge Base in the context of AI?
A Knowledge Base in AI is a technology used to store complex structured and unstructured information used by a computer system. It provides a means for information to be collected, organized, shared, searched and utilized.
What is the purpose of testing and training Knowledge Base in Azure AI?
The purpose of training and testing a Knowledge Base in Azure AI is to improve its accuracy, efficiency, and performance in answering queries. Training allows the model to learn and understand from the provided data, while testing helps in validating the model’s accuracy in answering unknown queries.
How do you create a Knowledge Base in QnA Maker in Azure AI?
A Knowledge Base in QnA Maker can be created by following these steps:
– Sign in to the QnA Maker portal,
– Click on ‘Create a knowledge base’,
– If you don’t have a QnA service under ‘Resource’, create a new one,
– Fill in the required fields,
– Click on ‘Create your KB’.
What types of data can be added to a Knowledge Base in Azure AI?
In Azure AI, a Knowledge Base can have information from various sources such as structured data like FAQs, unstructured data like word documents, manuals, PDFs, URLs, and product manuals, and even semi-structured data.
What is the process of training a Knowledge Base in Azure AI?
Training a Knowledge Base in Azure AI involves feeding it with Q&A pairs or relevant documents. The model learns from this data, identifying key phrases and patterns. The training enables the AI model to generate answers to user queries accurately.
How can you update a Knowledge Base in Azure AI?
To update a Knowledge Base in Azure AI, you have to:
– Open the QnA Maker portal,
– Open the Knowledge Base,
– Add new Q&A pairs or delete/edit existing ones,
– Save and Train the Knowledge Base to implement the changes.
How do you test a Knowledge Base in Azure AI?
To test a Knowledge Base in Azure AI, you can use the ‘Test’ pane of the Knowledge Base. It allows you to enter a query and see the Knowledge Base’s response. This ensures that the Knowledge Base accurately answers queries.
What is the ‘Publish’ function in Azure AI’s Knowledge Base?
The ‘Publish’ function in Azure AI’s Knowledge Base is used to make the trained Knowledge Base available via HTTPs endpoint. Only after publishing, the Knowledge Base can be integrated into apps, bots, and other services.
What is an Azure Cognitive Service ‘QnA Maker’?
Azure Cognitive Service ‘QnA Maker’ is an API service that enables the creation of a question and answer system from your data. It’s used for building conversational question-answer layers over your data.
What is the role of ‘Active learning’ in training Knowledge Base in Azure AI?
In Azure AI, ‘Active learning’ facilitates the improvement of the Knowledge Base by suggesting more optimal questions or alternative phrasing based on the real-world inquiries received from the users. The suggested alternative phrase can be added to the Knowledge Base, improving the AI model.
How can you add metadata to a Knowledge Base in Azure AI?
Metadata can be added to a Knowledge Base in Azure AI through the Edit mode. Click on ‘Add Metadata’, enter the Name and Value for the metadata and then click ‘Save and Train’.
Can you delete a Knowledge Base in Azure AI?
Yes, a Knowledge Base in Azure AI can be deleted. You need to go to ‘My knowledge bases’ page, select the Knowledge Base you want to delete, and click on ‘Delete’. After confirming, your Knowledge Base will be deleted.
How does a Knowledge Base use ‘Ranker’ in Azure AI?
The Knowledge Base in Azure AI uses the ‘Ranker’ to determine the best answers to user queries. It does this by ranking the answers with a score based on the user’s query and the matching intent/answered determined by the model’s training.
What is the function of ‘Chit-chat’ dataset in Azure AI?
The ‘Chit-chat’ dataset is pre-built small talk datasets provided by Azure AI for casual and friendly conversation. They can be added to the Knowledge Base to enhance the user experience by providing responses to common conversational phrases.
What are ‘Follow-up Prompts’ in Azure AI’s Knowledge Base?
Follow-up Prompts in Azure AI’s Knowledge Base are used to guide the user’s dialogue. They suggest the next question for the user based on the current question. This can help have a more interactive and engaging conversation with the user.
extutorials.com