MLS-C01 practice questions 141–150

1. Question 141 · 368 in bank

   A bank wants to launch a low-rate credit promotion campaign. The bank must identify which customers to target with the promotion and wants to make sure that each customer's full credit history is considered when an approval or denial decision is made. The bank's data science team used the XGBoost algorithm to train a classification model based on account transaction features. The data science team deployed the model by using the Amazon SageMaker model hosting service. The accuracy of the model is sufficient, but the data science team wants to be able to explain why the model denies the promotion to some customers. What should the data science team do to meet this requirement in the MOST operationally efficient manner?

   • ACreate a SageMaker notebook instance. Upload the model artifact to the notebook. Use the plot_importance() method in the Python XGBoost interface to create a feature importance chart for the individual predictions.
   • BRetrain the model by using SageMaker Debugger. Configure Debugger to calculate and collect Shapley values. Create a chart that shows features and SHapley. Additive explanations (SHAP) values to explain how the features affect the model outcomes.
   • CSet up and run an explainability job powered by SageMaker Clarify to analyze the individual customer data, using the training data as a baseline. Create a chart that shows features and SHapley Additive explanations (SHAP) values to explain how the features affect the model outcomes.
   • DUse SageMaker Model Monitor to create Shapley values that help explain model behavior. Store the Shapley values in Amazon S3. Create a chart that shows features and SHapley Additive explanations (SHAP) values to explain how the features affect the model outcomes.

   Show answer

   Answer: C

   Explanation

   "Explain individual model predictions Customers and internal stakeholders both want transparency into how models make their predictions. SageMaker Clarify integrates with SageMaker Experiments to show you the importance of each model input for a specific prediction. Results can be made available to customer-facing employees so that they have an understanding of the model’s behavior when making decisions based on model predictions." https://www.amazonaws.cn/en/sagemaker/clarify/

   Source: Adapted from community discussion — sevosevo

   Next question

2. Question 142 · 368 in bank

   Select 3

   A company has hired a data scientist to create a loan risk model. The dataset contains loan amounts and variables such as loan type, region, and other demographic variables. The data scientist wants to use Amazon SageMaker to test bias regarding the loan amount distribution with respect to some of these categorical variables. Which pretraining bias metrics should the data scientist use to check the bias distribution? (Choose three.)

   • AClass imbalance
   • BConditional demographic disparity
   • CDifference in proportions of labels
   • DJensen-Shannon divergence
   • EKullback-Leibler divergence
   • FTotal variation distance

   Show answer

   Answer: DEF

   Explanation

   Since it is indicated in the official web site that D, E and F are used to determine how different the distributions for loan application outcomes are for different demographic groups https://docs.aws.amazon.com/sagemaker/latest/dg/clarify-measure-data-bias.html

   Source: Adapted from community discussion — seifskl

   References

   • AWS documentation

   Next question

3. Question 143 · 368 in bank

   A retail company wants to use Amazon Forecast to predict daily stock levels of inventory. The cost of running out of items in stock is much higher for the company than the cost of having excess inventory. The company has millions of data samples for multiple years for thousands of items. The company’s purchasing department needs to predict demand for 30-day cycles for each item to ensure that restocking occurs. A machine learning (ML) specialist wants to use item-related features such as "category," "brand," and "safety stock count." The ML specialist also wants to use a binary time series feature that has "promotion applied?" as its name. Future promotion information is available only for the next 5 days. The ML specialist must choose an algorithm and an evaluation metric for a solution to produce prediction results that will maximize company profit. Which solution will meet these requirements?

   • ATrain a model by using the Autoregressive Integrated Moving Average (ARIMA) algorithm. Evaluate the model by using the Weighted Quantile Loss (wQL) metric at 0.75 (P75).
   • BTrain a model by using the Autoregressive Integrated Moving Average (ARIMA) algorithm. Evaluate the model by using the Weighted Absolute Percentage Error (WAPE) metric.
   • CTrain a model by using the Convolutional Neural Network - Quantile Regression (CNN-QR) algorithm. Evaluate the model by using the Weighted Quantile Loss (wQL) metric at 0.75 (P75).
   • DTrain a model by using the Convolutional Neural Network - Quantile Regression (CNN-QR) algorithm. Evaluate the model by using the Weighted Absolute Percentage Error (WAPE) metric.

   Show answer

   Answer: C

   Explanation

   A. NO - lot of data is available, better to use CNN-QR B. NO - lot of data is available, better to use CNN-QR C. YES - wQL is particularly useful when there are different costs for underpredicting and overpredicting (https://docs.aws.amazon.com/forecast/latest/dg/metrics.html#metrics-wQL) D. NO - WAPE will measure deviation, but no over vs. under forecasting

   Source: Adapted from community discussion — loict

   References

   • AWS documentation
   • AWS documentation
   • AWS documentation

   Next question

4. Question 144 · 368 in bank

   Select 3

   An online retail company wants to develop a natural language processing (NLP) model to improve customer service. A machine learning (ML) specialist is setting up distributed training of a Bidirectional Encoder Representations from Transformers (BERT) model on Amazon SageMaker. SageMaker will use eight compute instances for the distributed training. The ML specialist wants to ensure the security of the data during the distributed training. The data is stored in an Amazon S3 bucket. Which combination of steps should the ML specialist take to protect the data during the distributed training? (Choose three.)

   • ARun distributed training jobs in a private VPC. Enable inter-container traffic encryption.
   • BRun distributed training jobs across multiple VPCs. Enable VPC peering.
   • CCreate an S3 VPC endpoint. Then configure network routes, endpoint policies, and S3 bucket policies.
   • DGrant read-only access to SageMaker resources by using an IAM role.
   • ECreate a NAT gateway. Assign an Elastic IP address for the NAT gateway.
   • FConfigure an inbound rule to allow traffic from a security group that is associated with the training instances.

   Show answer

   Answer: ACD

   Explanation

   A - Running the training jobs in a private VPC will ensure that the data is transmitted over an encrypted channel. Enabling inter-container traffic encryption will encrypt data that is transmitted between containers. This will help protect the data during the distributed training. C - Creating an S3 VPC endpoint will provide a secure and private connection between the VPC and the S3 bucket. Network routes, endpoint policies, and S3 bucket policies can be configured to further secure the data during the distributed training. D - Granting read-only access to SageMaker resources by using an IAM role will ensure that the data is only accessed by the necessary resources during the distributed training. This will help prevent unauthorized access to the data.

   Source: Adapted from community discussion — oso0348

   References

   • AWS documentation
   • AWS documentation

   Next question

5. Question 145 · 368 in bank

   A Machine Learning Specialist is building a convolutional neural network (CNN) that will classify 10 types of animals. The Specialist has built a series of layers in a neural network that will take an input image of an animal, pass it through a series of convolutional and pooling layers, and then finally pass it through a dense and fully connected layer with 10 nodes. The Specialist would like to get an output from the neural network that is a probability distribution of how likely it is that the input image belongs to each of the 10 classes. Which function will produce the desired output?

   • ADropout
   • BSmooth L1 loss
   • CSoftmax
   • DRectified linear units (ReLU)

   Show answer

   Answer: C

   Explanation

   C --> Softmax. Let's go over the alternatives: A. Dropout --> Not really a function, but rather a method to avoid overfitting. It consists of dropping some neurons during the training process, so that the performance of our algorithm does not become very dependent on any single neuron. B. Smooth L1 loss --> It's a loss function, thus a function to be minimized by the entire neural network. It's not an activation function. C. Softmax --> This is the traditional function used for multi-class classification problems (such as classifying an animal into one of 10 categories) D. Rectified linear units (ReLU) --> This activation function is often used on the first and intermediate (hidden) layers, not on the final layer. In any case, it wouldn't make sense to use it for classification because its values can exceed 1 (and probabilities can't)

   Source: Adapted from community discussion — felbuch

   Next question

6. Question 146 · 368 in bank

   An analytics company has an Amazon SageMaker hosted endpoint for an image classification model. The model is a custom-built convolutional neural network (CNN) and uses the PyTorch deep learning framework. The company wants to increase throughput and decrease latency for customers that use the model. Which solution will meet these requirements MOST cost-effectively?

   • AUse Amazon Elastic Inference on the SageMaker hosted endpoint.
   • BRetrain the CNN with more layers and a larger dataset.
   • CRetrain the CNN with more layers and a smaller dataset.
   • DChoose a SageMaker instance type that has multiple GPUs.

   Show answer

   Answer: A

   Explanation

   Use Amazon Elastic Inference on the SageMaker hosted endpoint would be the most cost-effective solution for increasing throughput and decreasing latency. Amazon Elastic Inference is a service that allows you to attach GPU-powered inference acceleration to Amazon SageMaker hosted endpoints and EC2 instances. By attaching an Elastic Inference accelerator to the SageMaker endpoint, you can achieve better performance with lower costs than using a larger, more expensive instance type.

   Source: Adapted from community discussion — oso0348

   Next question

7. Question 147 · 368 in bank

   Select 3

   An ecommerce company is collecting structured data and unstructured data from its website, mobile apps, and IoT devices. The data is stored in several databases and Amazon S3 buckets. The company is implementing a scalable repository to store structured data and unstructured data. The company must implement a solution that provides a central data catalog, self-service access to the data, and granular data access policies and encryption to protect the data. Which combination of actions will meet these requirements with the LEAST amount of setup? (Choose three.)

   • AIdentify the existing data in the databases and S3 buckets. Link the data to AWS Lake Formation.
   • BIdentify the existing data in the databases and S3 buckets. Link the data to AWS Glue.
   • CRun AWS Glue crawlers on the linked data sources to create a central data catalog.
   • DApply granular access policies by using AWS Identity and Access Management (1AM). Configure server-side encryption on each data source.
   • EApply granular access policies and encryption by using AWS Lake Formation.
   • FApply granular access policies and encryption by using AWS Glue.

   Show answer

   Answer: ACE

   References

   • AWS documentation
   • AWS documentation
   • AWS documentation
   • AWS documentation

   Next question

8. Question 148 · 368 in bank

   Select 3

   A machine learning (ML) specialist is developing a deep learning sentiment analysis model that is based on data from movie reviews. After the ML specialist trains the model and reviews the model results on the validation set, the ML specialist discovers that the model is overfitting. Which solutions will MOST improve the model generalization and reduce overfitting? (Choose three.)

   • AShuffle the dataset with a different seed.
   • BDecrease the learning rate.
   • CIncrease the number of layers in the network.
   • DAdd L1 regularization and L2 regularization.
   • EAdd dropout.
   • FDecrease the number of layers in the network.

   Show answer

   Answer: DEF

   Explanation

   To improve the generalization of the deep learning sentiment analysis model and reduce overfitting, the following three solutions can be implemented: Add Dropout: Dropout is a regularization technique that randomly drops out (sets to zero) a certain percentage of nodes in the neural network during each training epoch. This helps to prevent overfitting and improve generalization. Add L1 and L2 Regularization: L1 and L2 regularization are techniques used to add a penalty to the loss function of the neural network, which helps to prevent overfitting. L1 regularization adds a penalty based on the absolute value of the weights, while L2 regularization adds a penalty based on the squared value of the weights. Decrease the number of layers in the network: Deep neural networks with too many layers can be prone to overfitting. Reducing the number of layers in the network can help to prevent overfitting and improve generalization.

   Source: Adapted from community discussion — oso0348

   Next question

9. Question 149 · 368 in bank

   An online advertising company is developing a linear model to predict the bid price of advertisements in real time with low-latency predictions. A data scientist has trained the linear model by using many features, but the model is overfitting the training dataset. The data scientist needs to prevent overfitting and must reduce the number of features. Which solution will meet these requirements?

   • ARetrain the model with L1 regularization applied.
   • BRetrain the model with L2 regularization applied.
   • CRetrain the model with dropout regularization applied.
   • DRetrain the model by using more data.

   Show answer

   Answer: A

   Explanation

   A. YES - best to reduce feature count B. NO - L2 will reduce large weights and smooth features, to get rid of them C. NO - dropout is for NN D. NO - we are already converging, no need for more data

   Source: Adapted from community discussion — loict

   Next question

10. Question 150 · 368 in bank

    A credit card company wants to identify fraudulent transactions in real time. A data scientist builds a machine learning model for this purpose. The transactional data is captured and stored in Amazon S3. The historic data is already labeled with two classes: fraud (positive) and fair transactions (negative). The data scientist removes all the missing data and builds a classifier by using the XGBoost algorithm in Amazon SageMaker. The model produces the following results: • True positive rate (TPR): 0.700 • False negative rate (FNR): 0.300 • True negative rate (TNR): 0.977 • False positive rate (FPR): 0.023 • Overall accuracy: 0.949 Which solution should the data scientist use to improve the performance of the model?

    • AApply the Synthetic Minority Oversampling Technique (SMOTE) on the minority class in the training dataset. Retrain the model with the updated training data.
    • BApply the Synthetic Minority Oversampling Technique (SMOTE) on the majority class in the training dataset. Retrain the model with the updated training data.
    • CUndersample the minority class.
    • DOversample the majority class.

    Show answer

    Answer: A

    Explanation

    A. YES - we want to oversample the minority class = Fraud B. NO - we want more fraudulent cases C. NO - we want more fraudulent cases D. NO - we want more fraudulent cases

    Source: Adapted from community discussion — loict

    Next question

Keyboard:J Kmove between questionsA–Dselect an answerRshow or hide the answer← →change page

MLS-C01 practice, question 141. 368 questions available.