Complete MLS-C01 Amazon Web Services Materials

The combination of steps that the data scientist must take to perform the A/B testing are to create a new endpoint configuration that includes a production variant for each of the two models, and update the existing endpoint to use the new endpoint configuration. This approach will allow the data scientist to deploy both models on the same endpoint and split the inference traffic between them based on a specified distribution.

Amazon SageMaker is a fully managed service that provides developers and data scientists the ability to quickly build, train, and deploy machine learning models. Amazon SageMaker supports A/B testing on machine learning models by allowing the data scientist to run multiple production variants on an endpoint. A production variant is a version of a model that is deployed on an endpoint. Each production variant has a name, a machine learning model, an instance type, an initial instance count, and an initial weight. The initial weight determines the percentage of inference requests that the variant will handle. For example, if there are two variants with weights of 0.5 and 0.5, each variant will handle 50% of the requests. The data scientist can use production variants to test models that have been trained using different training datasets, algorithms, and machine learning frameworks; test how they perform on different instance types; or a combination of all of the above1.

To perform A/B testing on machine learning models, the data scientist needs to create a new endpoint configuration that includes a production variant for each of the two models. An endpoint configuration is a collection of settings that define the properties of an endpoint, such as the name, the production variants, and the data capture configuration. The data scientist can use the Amazon SageMaker console, the AWS CLI, or the AWS SDKs to create a new endpoint configuration. The data scientist needs to specify the name, model name, instance type, initial instance count, and initial variant weight for each production variant in the endpoint configuration2.

After creating the new endpoint configuration, the data scientist needs to update the existing endpoint to use the new endpoint configuration. Updating an endpoint is the process of deploying a new endpoint configuration to an existing endpoint. Updating an endpoint does not affect the availability or scalability of the endpoint, as Amazon SageMaker creates a new endpoint instance with the new configuration and switches the DNS record to point to the new instance when it is ready. The data scientist can use the Amazon SageMaker console, the AWS CLI, or the AWS SDKs to update an endpoint. The data scientist needs to specify the name of the endpoint and the name of the new endpoint configuration to update the endpoint3.

The other options are either incorrect or unnecessary. Creating a new endpoint configuration that includes two target variants that point to different endpoints is not possible, as target variants are only used to invoke a specific variant on an endpoint, not to define an endpoint configuration. Deploying the new model to the existing endpoint would replace the existing model, not run it side-by-side with the new model. Updating the existing endpoint to activate the new model is not a valid operation, as there is no activation parameter for an endpoint.

References:

• 1: A/B Testing ML models in production using Amazon SageMaker | AWS Machine Learning Blog
• 2: Create an Endpoint Configuration - Amazon SageMaker
• 3: Update an Endpoint - Amazon SageMaker

The solution A will meet the requirements with the least development effort because it uses Amazon Kendra, which is a highly accurate and easy to use intelligent search service powered by machine learning. Amazon Kendra can index company documents from various sources and formats, such as PDF, HTML, Word, and more. Amazon Kendra can also integrate with chatbots by using the Amazon Kendra Query API operation, which can understand natural language questions and provide relevant answers from the indexed documents. Amazon Kendra can also provide additional information, such as document excerpts, links, and FAQs, to enhance the chatbot experience1.

The other options are not suitable because:

• Option B: Training a Bidirectional Attention Flow (BiDAF) network based on past customer questions and company documents, deploying the model as a real-time Amazon SageMaker endpoint, and integrating the model with the chatbot by using the SageMaker Runtime InvokeEndpoint API operation will incur more development effort than using Amazon Kendra. The company will have to write the code for the BiDAF network, which is a complex deep learning model for question answering. The company will also have to manage the SageMaker endpoint, the model artifact, and the inference logic2.
• Option C: Training an Amazon SageMaker BlazingText model based on past customer questions and company documents, deploying the model as a real-time SageMaker endpoint, and integrating the model with the chatbot by using the SageMaker Runtime InvokeEndpoint API operation will incur more development effort than using Amazon Kendra. The company will have to write the code for the BlazingText model, which is a fast and scalable text classification and word embedding algorithm. The company will also have to manage the SageMaker endpoint, the model artifact, and the inference logic3.
• Option D: Indexing company documents by using Amazon OpenSearch Service and integrating the chatbot with OpenSearch Service by using the OpenSearch Service k-nearest neighbors (k-NN) Query API operation will not meet the requirements effectively. Amazon OpenSearch Service is a fully managed service that provides fast and scalable search and analytics capabilities. However, it is not designed for natural language question answering, and it may not provide accurate or relevant answers for the chatbot. Moreover, the k-NN Query API operation is used to find the most similar documents or vectors based on a distance function, not to find the best answers based on a natural language query4.

References:

• 1: Amazon Kendra
• 2: Bidirectional Attention Flow for Machine Comprehension
• 3: Amazon SageMaker BlazingText
• 4: Amazon OpenSearch Service

The solution C is the best option to identify and address training issues with the least development effort. The solution C involves the following steps:

• Use the SageMaker Debugger vanishing_gradient and LowGPUUtilization built-in rules to detect issues. SageMaker Debugger is a feature of Amazon SageMaker that allows data scientists to monitor, analyze, and debug machine learning models during training. SageMaker Debugger provides a set of built-in rules that can automatically detect common issues and anomalies in model training, such as vanishing or exploding gradients, overfitting, underfitting, low GPU utilization, and more1. The data scientist can use the vanishing_gradient rule to check if the gradients are becoming too small and causing the training to not converge. The data scientist can also use the LowGPUUtilization rule to check if the GPU resources are underutilized and causing the training to be inefficient2.
• Launch the StopTrainingJob action if issues are detected. SageMaker Debugger can also take actions based on the status of the rules. One of the actions is StopTrainingJob, which can terminate the training job if a rule is in an error state. This can help the data scientist to save time and money by stopping the training early if issues are detected3.

The other options are not suitable because:

• Option A: Using CPU utilization metrics that are captured in Amazon CloudWatch and configuring a CloudWatch alarm to stop the training job early if low CPU utilization occurs will not identify and address training issues effectively. CPU utilization is not a good indicator of model training performance, especially for GPU instances. Moreover, CloudWatch alarms can only trigger actions based on simple thresholds, not complex rules or conditions4.
• Option B: Using high-resolution custom metrics that are captured in Amazon CloudWatch and configuring an AWS Lambda function to analyze the metrics and to stop the training job early if issues are detected will incur more development effort than using SageMaker Debugger. The data scientist will have to write the code for capturing, sending, and analyzing the custom metrics, as well as for invoking the Lambda function and stopping the training job. Moreover, this solution may not be able to detect all the issues that SageMaker Debugger can5.
• Option D: Using the SageMaker Debugger confusion and feature_importance_overweight built-in rules and launching the StopTrainingJob action if issues are detected will not identify and address training issues effectively. The confusion rule is used to monitor the confusion matrix of a classification model, which is not relevant for a regression model that predicts prices. The feature_importance_overweight rule is used to check if some features have too much weight in the model, which may not be related to the convergence or resource utilization issues2.

References:

• 1: Amazon SageMaker Debugger
• 2: Built-in Rules for Amazon SageMaker Debugger
• 3: Actions for Amazon SageMaker Debugger
• 4: Amazon CloudWatch Alarms
• 5: Amazon CloudWatch Custom Metrics

 The problem described in the question is a case of overfitting, where the neural network model performs well on the training data but poorly on the test data. This means that the model has learned the noise and specific patterns of the training data, but cannot generalize to new and unseen data. To solve this issue, the Specialist should consider the following changes:

• Choose a lower number of layers: Reducing the number of layers can reduce the complexity and capacity of the neural network model, making it less prone to overfitting. A model with 50 hidden layers is likely too deep for the given data size and task. A simpler model with fewer layers can learn the essential features of the data without memorizing the noise.
• Enable dropout: Dropout is a regularization technique that randomly drops out some units in the neural network during training. This prevents the units from co-adapting too much and forces the model to learn more robust features. Dropout can improve the generalization and test performance of the model by reducing overfitting.
• Enable early stopping: Early stopping is another regularization technique that monitors the validation error during training and stops the training process when the validation error stops decreasing or starts increasing. This prevents the model from overtraining on the training data and reduces overfitting.

References:

• Deep Learning - Machine Learning Lens
• How to Avoid Overfitting in Deep Learning Neural Networks
• How to Identify Overfitting Machine Learning Models in Scikit-Learn