MLS-C01 Modeling Practice Test 17 — 15 Questions

Question 1

A machine learning engineer is training a deep learning model on Amazon SageMaker using the PyTorch framework. The training data is stored in an Amazon S3 bucket in TFRecord format. The engineer wants to maximize GPU utilization and minimize training time. The dataset is large (500 GB) and consists of many small files. Which data input mode and configuration should the engineer use to achieve the best performance?

Choose an answer to begin — your selection is scored in the full session.

Accepted Answer

Use FastFile mode with the PyTorch framework and configure the training script to read files as if they were local.. FastFile mode is designed for high-performance, on-demand access to S3 data without pre-downloading. It allows the training script to read files as if they were local, which works seamlessly with PyTorch's data loading. For large datasets with many small files, FastFile mode minimizes startup latency and maximizes GPU utilization by streaming data efficiently.

Answer

Use File mode but first combine all small files into a single large TFRecord file using a SageMaker Processing job.

Answer

Use Pipe mode with the PyTorch framework and configure the training script to read from the pipe.

Answer

Use File mode and rely on the default SageMaker data download to the training instance's EBS volume.

Question 2

A data scientist is using Amazon SageMaker to train a deep learning model for natural language processing. The training job is taking too long, and the data scientist wants to reduce the training time without sacrificing model accuracy. The model uses a large embedding layer and many LSTM layers. Which of the following changes is most likely to achieve this goal?

Accepted Answer

Increase the batch size and adjust the learning rate accordingly.. Increasing batch size is a straightforward way to improve training throughput by leveraging parallelism in GPUs. When done with a corresponding increase in learning rate, it can maintain or even improve convergence speed and accuracy. This is a common optimization in deep learning training.

Answer

Use a smaller embedding dimension.

Answer

Switch from LSTM to GRU layers.

Answer

Enable distributed training with multiple instances.

Question 3

A machine learning engineer is training a deep learning model on Amazon SageMaker using the PyTorch framework. The training data is stored in an Amazon S3 bucket in TFRecord format. The engineer wants to maximize GPU utilization and minimize training time. The dataset is large (500 GB) and consists of many small files. Which data input mode and configuration should the engineer use to achieve the best performance?

Choose an answer to begin — your selection is scored in the full session.

Accepted Answer

Enable early stopping in the Linear Learner hyperparameters and set the predictor type to 'regressor'.. Linear Learner's early stopping feature monitors validation loss and stops training when it stops improving, which directly prevents overfitting. The predictor type must be set to 'regressor' for regression. This is the correct configuration because it automatically halts training at the optimal point and retains the best model, unlike manual intervention or hyperparameter tuning.

Answer

Use SageMaker Automatic Model Tuning with a large hyperparameter range and set the objective metric to validation:loss.

Answer

Configure the training job to use a smaller instance type and reduce the number of epochs manually.

Answer

Enable checkpointing in the training job and set the checkpoint frequency to 1 epoch.

Question 4

A data scientist is training a linear regression model using Amazon SageMaker's built-in Linear Learner algorithm. The dataset contains 20 features, and the target variable is a continuous value. The data scientist wants to ensure the model does not overfit. Which hyperparameter should be tuned to control the strength of L2 regularization?

Accepted Answer

`wd` (weight decay). The SageMaker Linear Learner algorithm provides hyperparameters for both L1 and L2 regularization. The `wd` hyperparameter specifically controls the L2 regularization strength. By increasing `wd`, the model penalizes large weights, reducing overfitting. This is the appropriate parameter to tune when the goal is to apply L2 regularization.

Answer

`learning_rate`

Answer

`l1`

Answer

`mini_batch_size`

Question 5

A machine learning engineer is using Amazon SageMaker to train a model with the built-in K-Means algorithm for customer segmentation. The engineer notices that the clusters formed are not meaningful and wants to improve the clustering quality. Which two actions should the engineer take to improve the results? (Choose two.)

Accepted Answer

Use the elbow method or silhouette analysis to determine an appropriate number of clusters.. For K-Means clustering, feature scaling is critical because the algorithm relies on distance metrics. Standardizing features ensures equal contribution. Additionally, selecting the optimal number of clusters using methods like the elbow method or silhouette analysis prevents poor clustering. These two actions together significantly improve the meaningfulness of the segments.

Answer

Remove outliers from the dataset to prevent them from skewing cluster centroids.

Answer

Switch to a different algorithm like PCA for dimensionality reduction before clustering.

Answer

Increase the number of clusters (k) to a very large value to capture more granular segments.

Question 6

A data scientist is training a k-means clustering model using the SageMaker built-in KMeans algorithm on a dataset with 10 million records and 100 features. The training is taking a long time. Which action will most effectively reduce training time while still producing a reasonable clustering?

Accepted Answer

Decrease the value of the k (number of clusters) parameter.. The k parameter controls the number of clusters, and its value directly affects the computational complexity of each iteration. Fewer clusters mean fewer distance computations, which reduces training time significantly on large datasets. While other parameters like mini_batch_size and epochs also influence speed, decreasing k offers a straightforward trade-off between granularity and performance. If a coarser clustering is acceptable, this is the most effective action.

Answer

Increase the value of the mini_batch_size parameter.

Answer

Increase the value of the epochs parameter.

Answer

Set the init_method parameter to 'random' instead of 'k-means++'.

Question 7

A data scientist is training a gradient boosting model using Amazon SageMaker's built-in XGBoost algorithm on a dataset with 10,000 rows and 200 features. The model performs well on training data but poorly on validation data. The scientist suspects overfitting. Which hyperparameter tuning strategy is most likely to reduce overfitting?

Accepted Answer

Decrease the maximum depth of each tree and increase the minimum child weight.. Reducing maximum depth and increasing minimum child weight are direct regularization techniques that constrain the complexity of the boosted trees. They prevent the model from fitting noise in the training data, thereby improving validation performance. Other options either do not address overfitting or could increase model complexity.

Answer

Increase the learning rate and decrease the number of boosting rounds.

Answer

Increase the number of parallel trees and use a higher subsample ratio.

Answer

Increase the L2 regularization parameter and decrease the feature dimension.

Question 8

A data scientist is using Amazon SageMaker to train a model with the built-in K-Means algorithm for customer segmentation. The data scientist wants to ensure the algorithm converges to a good solution and produces meaningful clusters. Which two actions should the data scientist take to improve the quality of the clustering results? (Choose two.)

Accepted Answer

Standardize the features to have zero mean and unit variance before training.. Standardizing features ensures that all features contribute equally to distance calculations, which is crucial for K-Means. Using k-means++ initialization helps avoid poor local minima by choosing better starting centroids. These two actions directly address common pitfalls in K-Means clustering and improve the likelihood of meaningful clusters.

Answer

Increase the number of clusters (k) to a very large value to capture all nuances.

Answer

Set the 'feature_dim' parameter to match the number of features in the dataset.

Answer

Reduce the number of features by applying PCA before training.

Question 9

A machine learning engineer is using Amazon SageMaker to train a model with the built-in K-Means algorithm. The engineer wants to determine the optimal number of clusters for a dataset of customer behaviors. Which method should the engineer use?

Accepted Answer

Plot the within-cluster sum of squares (WCSS) against the number of clusters and look for an elbow.. The elbow method is the correct approach because it visualizes the trade-off between the number of clusters and the compactness of clusters, helping identify the point of diminishing returns. This is a widely accepted technique for K-Means. Other options are either not applicable or less standard for this specific task.

Answer

Use the Akaike Information Criterion (AIC) to compare models with different numbers of clusters.

Answer

Set the number of clusters equal to the square root of the number of data points.

Answer

Perform a grid search over the number of clusters and select the one with the highest silhouette score.

Question 10

A data scientist is using Amazon SageMaker to train a linear learner model for a regression task. The dataset contains 20 features, some of which are highly correlated. After training, the model's performance on the validation set is poor, and the training loss is much lower than the validation loss. The data scientist suspects overfitting. Which action should the data scientist take to reduce overfitting while maintaining model interpretability?

Accepted Answer

Set the L2 regularization parameter (wd) to a non-zero value in the linear learner hyperparameters.. L2 regularization (weight decay) is the most appropriate technique here because it reduces overfitting by shrinking coefficients, and it handles correlated features gracefully by distributing weight among them. It preserves interpretability since the original features remain in the model. The linear learner's 'wd' hyperparameter controls this regularization strength.

Answer

Reduce the number of features by applying principal component analysis (PCA) before training.

Answer

Increase the number of epochs and use early stopping with a higher patience value.

Answer

Set the L1 regularization parameter (l1) to a non-zero value in the linear learner hyperparameters.

Question 11

A machine learning team is using Amazon SageMaker to train a model with the built-in k-means algorithm for customer segmentation. They have a dataset with 10,000 records and 20 features. After training, they notice that the clusters are not meaningful and the within-cluster sum of squares (WCSS) is high. They want to improve the clustering results. Which of the following steps should they take? (Choose two.)

Accepted Answer

Use the elbow method to determine the optimal number of clusters.. Standardizing features ensures that all features contribute equally to distance calculations, which is crucial for k-means. Using the elbow method helps select an optimal k that balances WCSS and cluster meaningfulness. Together, these steps directly address the high WCSS and poor cluster quality.

Answer

Increase the number of clusters (k).

Answer

Reduce the number of features using PCA before training.

Answer

Change the distance metric to Manhattan distance.

Question 12

A machine learning engineer is using Amazon SageMaker to train a deep learning model for image classification. The training job uses a large dataset stored in Amazon S3 and takes several hours. The engineer wants to reduce the training time without sacrificing model accuracy. The model uses a batch size of 32 and a learning rate of 0.001. Which change is MOST likely to reduce training time while maintaining accuracy?

Accepted Answer

Increase the batch size to 256 and proportionally increase the learning rate to 0.008.. Increasing batch size and proportionally scaling the learning rate is a standard technique to accelerate deep learning training. It improves hardware utilization and reduces the number of updates per epoch, while the learning rate adjustment maintains convergence. Other options either increase training time or do not directly address per-epoch efficiency. Thus, this change is most likely to reduce training time while preserving accuracy.

Answer

Use a larger instance type with more GPUs but keep the batch size and learning rate unchanged.

Answer

Enable early stopping with a patience of 5 epochs to halt training when validation loss stops improving.

Answer

Decrease the batch size to 16 and decrease the learning rate to 0.0005.

Question 13

A machine learning engineer is using Amazon SageMaker to train a model with the built-in K-Means algorithm on a dataset with 10 million records and 50 features. The training job is taking too long, and the engineer suspects that the algorithm is not converging efficiently. The engineer wants to reduce training time while maintaining reasonable cluster quality. Which action should the engineer take?

Accepted Answer

Increase the mini_batch_size hyperparameter.. The K-Means algorithm in SageMaker uses mini-batch updates to handle large datasets. Increasing mini_batch_size allows more samples per update, which can accelerate convergence and reduce training time. While it may slightly affect cluster quality, it is the most direct action to address the slow training without changing the problem definition.

Answer

Set the init_method hyperparameter to 'random' instead of 'k-means++'.

Answer

Enable early stopping by setting the early_stopping hyperparameter to 'true'.

Answer

Decrease the number of clusters (k).

Question 14

A machine learning engineer is using Amazon SageMaker to train a deep learning model with the built-in Object Detection algorithm. The training job is using a p3.8xlarge instance and a large dataset stored in Amazon S3. The engineer notices that the GPU utilization is consistently below 30%, and the training job is taking much longer than expected. Which action should the engineer take to improve GPU utilization and reduce training time?

Accepted Answer

Increase the value of the mini_batch_size hyperparameter.. Low GPU utilization typically means the GPU is starved for data or the batch size is too small to keep it busy. Increasing mini_batch_size allows more samples to be processed in parallel, which can improve GPU utilization and reduce training time, provided memory constraints are respected. Other options like changing learning rate or instance type do not directly address the bottleneck. Pipe mode can help with data loading but is not the primary fix for small batch size.

Answer

Use a larger instance type with more GPUs, such as p3.16xlarge.

Answer

Decrease the value of the learning_rate hyperparameter.

Answer

Enable SageMaker Pipe mode for the input data.

Question 15

A data scientist must train a model on a tabular dataset where the target is the count of insurance claims per policyholder, ranging from 0 to about 15, with many zeros. The team wants a SageMaker built-in algorithm whose objective matches count data. Which approach should the data scientist take?

Accepted Answer

Use the SageMaker built-in XGBoost algorithm with the count:poisson objective and set the number of classes appropriately for the target range.. Claim counts are non-negative integers with many zeros, which is the canonical setting for a Poisson objective. The SageMaker built-in XGBoost algorithm exposes count:poisson, which models the log of the expected count and handles the variance structure of count data far better than squared error or a binarized logistic objective.

Answer

Use the SageMaker built-in Linear Learner algorithm with the regressor predictor type and default squared loss on the raw counts.

Answer

Use the SageMaker built-in K-Means algorithm to cluster policyholders by claim count and predict the cluster centroid.

Answer

Use the SageMaker built-in XGBoost algorithm with the binary:logistic objective and binarize the target at a threshold of one claim.