MLS-C01 Practice Exam 61 — 20 Questions

Question 1

A machine learning engineer is using SageMaker Pipelines to orchestrate a training workflow. The pipeline includes a processing step that outputs a dataset to S3, followed by a training step that consumes that dataset. The engineer wants to ensure that the training step uses the output from the processing step. How should the engineer configure the training step?

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

Accepted Answer

Use the ProcessingOutput object from the processing step as the input to the training step.. SageMaker Pipelines allows steps to pass data by referencing the output of a previous step. The ProcessingOutput object contains the S3 URI where the processing step saves its output. By using it as input to the training step, the pipeline automatically resolves the correct location and establishes a dependency. This is the intended and most reliable method.

Answer

Add a condition step between the processing and training steps to check if the processing output exists.

Answer

Configure the training step to use the same instance type as the processing step.

Answer

Manually specify the S3 bucket and prefix that the processing step will use, and hardcode it in the training step's input.

Question 2

A company is using SageMaker to train a TensorFlow model for image classification. The training is slow on a single GPU instance. Which TWO strategies can reduce training time? (Choose TWO.)

Accepted Answer

Use SageMaker Pipe Mode for data ingestion. Option B is correct because SageMaker Pipe Mode streams training data directly from Amazon S3 into the training container instead of downloading the full dataset to local disk first, which removes the I/O bottleneck and keeps the GPU fed with data faster, reducing overall training time. Option D is correct because distributed training across multiple GPUs (e.g., using Horovod or SageMaker's distributed data parallel libraries) splits the compute and gradient synchronization work across devices, so each epoch completes in a fraction of the time compared to a single GPU. Option A is incorrect because increasing image size raises the computational cost per sample, making training slower, not faster. Option C is incorrect because adding more epochs increases total training time rather than reducing it. Option E is incorrect because decreasing the batch size means more forward/backward passes per epoch, which typically slows training and underutilizes GPU parallelism.

Answer

Increase the image size

Answer

Increase the number of training epochs

Answer

Decrease the batch size

Question 3

A data scientist is performing EDA on a dataset with a timestamp column. They want to detect seasonality. Which visualization is most appropriate?

Accepted Answer

Line plot of value over time. A line plot of value over time is the most appropriate visualization to detect seasonality because it displays the continuous temporal trend and repeating patterns (e.g., daily, weekly, yearly cycles) directly. By plotting the timestamp on the x-axis and the value on the y-axis, seasonal peaks and troughs become visually apparent. Other plots aggregate or lose the temporal ordering needed to see cyclicality.

Answer

Box plot of value grouped by month

Answer

Bar chart of average value per month

Answer

Scatter plot of timestamp vs. value

Question 4

A data scientist is performing EDA on a dataset with 500,000 rows and 10 columns. The dataset is stored in an S3 bucket as CSV files. The scientist wants to generate summary statistics (mean, median, min, max) for all numeric columns. Which service allows the quickest ad-hoc analysis without provisioning any infrastructure?

Accepted Answer

Amazon Athena. Amazon Athena is a serverless query service that can directly query CSV files in S3 using standard SQL, making it ideal for quick ad-hoc analysis without provisioning infrastructure. It automatically scales and charges per query, so the data scientist can run summary statistics immediately.

Answer

AWS Glue ETL

Answer

Amazon SageMaker Data Wrangler

Answer

Amazon QuickSight

Question 5

A data engineer needs to automate the transformation of CSV files to Parquet format as soon as they are uploaded to an S3 bucket. The transformed files should be stored in another S3 bucket. Which solution is the most cost-effective and requires the least maintenance?

Accepted Answer

Configure an S3 event notification to invoke a Lambda function.. AWS Lambda, triggered by S3 event notifications, provides a serverless, event-driven architecture that automatically converts CSV to Parquet upon file upload. This approach is cost-effective because you pay only for compute time during execution, and it requires minimal maintenance as AWS manages the infrastructure, scaling, and fault tolerance.

Answer

Configure an S3 event notification to invoke an AWS Glue job.

Answer

Run an Amazon EMR cluster continuously to watch for new files.

Answer

Set up an EC2 instance with a cron job to poll the S3 bucket.

Question 6

A company has a SageMaker training job that reads a large dataset from Amazon S3. The data is stored in a single S3 bucket in the same AWS Region. The training job is taking a very long time to start because SageMaker is downloading the entire dataset to the training instance's local disk before training begins. The data is in CSV format and is used in every epoch. The company wants to reduce the startup time and allow the training job to start training faster. What should the company do?

Accepted Answer

Use Pipe mode for the training input instead of File mode.. Using Pipe mode streams data directly from S3 to the training algorithm, avoiding the initial full download. This reduces startup time and allows training to begin faster. File mode downloads the entire dataset to local disk before training, which is slow for large datasets. Other options either do not address the download bottleneck or add overhead without solving the startup delay.

Answer

Enable Amazon S3 Transfer Acceleration on the bucket.

Answer

Convert the dataset to RecordIO format and use File mode.

Answer

Increase the number of training instances to distribute the data loading.

Question 7

A data engineering team is designing a data pipeline to process large CSV files (10-50 GB each) stored in Amazon S3. The pipeline must transform the data using AWS Glue and load it into Amazon Redshift for analytics. The team wants to minimize costs while ensuring the pipeline can handle peak loads. Which approach is the most cost-effective?

Accepted Answer

Use AWS Glue with Spark and dynamic frames, scaling the number of workers based on file size.. AWS Glue with Spark and dynamic frames is the most cost-effective approach because it is serverless, automatically scales workers based on file size, and is optimized for ETL on large CSV files (10-50 GB) in S3. Dynamic frames provide built-in transformations and schema inference, reducing development effort, while the ability to adjust the number of workers allows handling peak loads without over-provisioning. This minimizes idle compute costs compared to always-on clusters like EMR.

Answer

Use AWS Lambda to process each file and load into Redshift.

Answer

Use Amazon EMR with Hive to transform the data and load into Redshift.

Answer

Use an AWS Glue Python shell job with a single r5.xlarge worker.

Question 8

A company uses SageMaker to train a large language model. The training job is taking too long. The data scientist wants to use distributed training with data parallelism. Which SageMaker feature should be used?

Accepted Answer

SageMaker distributed training libraries. SageMaker distributed training libraries provide data parallelism and model parallelism specifically optimized for AWS infrastructure, enabling efficient multi-GPU/multi-node training of large models. For a large language model with data parallelism, these libraries handle gradient synchronization and communication overhead, directly addressing the slow training job. This is the purpose-built feature for distributed training in SageMaker.

Answer

SageMaker Neo

Answer

SageMaker Processing

Answer

SageMaker Debugger

Question 9

A machine learning engineer is training a deep neural network for a regression task using Amazon SageMaker's built-in Linear Learner algorithm. The model's predictions are consistently biased, and after inspecting the training data, the engineer discovers that one input feature has a mean of 500 and a standard deviation of 200, while all other features have means near 0 and standard deviations near 1. To improve model convergence and accuracy without changing the algorithm, what should the engineer do?

Accepted Answer

Apply feature scaling to the dataset before training, such as standardization or normalization.. Feature scaling is essential for algorithms like Linear Learner that rely on gradient descent. When one feature has a much larger scale, it can dominate the learning process, leading to slow convergence and biased models. Standardization or normalization brings all features to a comparable range, improving optimization and model accuracy without altering the algorithm.

Answer

Increase the number of epochs in the training job to allow the model more time to learn the feature's scale.

Answer

Switch to the XGBoost algorithm, which is inherently robust to unscaled features.

Answer

Remove the large-scale feature from the dataset to eliminate its influence on the model.

Question 10

A data scientist is using Amazon SageMaker to train a model with the built-in XGBoost algorithm. The training job is taking a long time, and the scientist wants to reduce the training time without significantly impacting model accuracy. The dataset is large, and the scientist has already tried increasing the instance count. Which hyperparameter should the scientist tune to speed up training by reducing the number of boosting rounds?

Accepted Answer

num_round. The number of boosting rounds (num_round) is directly proportional to training time. Reducing it will shorten training, though it may affect accuracy. Other hyperparameters like max_depth, eta, and subsample can influence training time per tree or the number of rounds needed, but num_round is the most direct control over the total number of trees built.

Answer

max_depth

Answer

eta

Answer

subsample

Question 11

A machine learning engineer is training a gradient boosting model on a tabular dataset with 500,000 rows and 120 features using Amazon SageMaker. The training job is taking excessively long, and the engineer suspects that the feature set contains many irrelevant or highly correlated features. The engineer wants to reduce the training time while preserving model accuracy. Which approach is MOST appropriate to address this issue?

Accepted Answer

Apply feature selection using techniques like variance threshold or recursive feature elimination before training, then retrain the model with the reduced feature set.. The excessive training time is likely due to a large number of features, many of which are irrelevant or correlated. Feature selection reduces the feature space, leading to faster training and potentially better model performance. Other options either do not address the dimensionality issue or may compromise accuracy. Thus, applying feature selection before retraining is the most effective and appropriate solution.

Answer

Switch to a linear learner algorithm, which is inherently faster for high-dimensional data.

Answer

Increase the instance count of the training job to distribute the workload across more machines.

Answer

Use SageMaker Automatic Model Tuning (AMT) with a large search space to find the best hyperparameters.

Question 12

A data scientist is using Amazon SageMaker to train a linear regression model. The dataset has 500 features and 50,000 observations. The model converges but has high bias. Which technique should the data scientist use to reduce bias?

Accepted Answer

Add polynomial features or interaction terms to the feature set.. High bias means the model is underfitting — it is too simple to capture the underlying patterns in the data. Adding polynomial features or interaction terms increases the model's capacity to represent complex relationships, which directly reduces bias. With 500 features and 50,000 observations, the dataset is large enough to support additional feature engineering without immediately overfitting.

Answer

Apply L2 regularization (Ridge) to penalize large coefficients.

Answer

Decrease the learning rate.

Answer

Use feature selection to remove irrelevant features.

Answer

Increase the number of training epochs.

Question 13

Which TWO services can be used to transform data in transit within a Kinesis Data Firehose delivery stream? (Choose 2)

Accepted Answer

AWS Lambda. AWS Lambda (A) is correct because Kinesis Data Firehose supports Lambda-based record transformation, where Firehose invokes a Lambda function synchronously on each incoming record to transform, parse, or enrich the data before delivery. AWS Glue (E) is correct because Firehose offers built-in integration with AWS Glue for record format conversion, such as converting JSON to Parquet or ORC using a Glue Data Catalog table schema during transit. Amazon Kinesis Data Analytics (B) is not a Firehose transform mechanism; it is a separate service for running SQL/Flink analytics over streaming data. Amazon Athena (C) is an interactive query service for data in S3, not an in-stream transformation service. Amazon S3 (D) is a destination for Firehose delivery, not a transformation service.

Answer

Amazon Kinesis Data Analytics

Answer

Amazon Athena

Answer

Amazon S3

Question 14

A team trains a binary classifier with Amazon SageMaker and the model outputs a probability score. The business requires that at most 2% of predicted positives be false positives, and within that constraint the team wants to maximize recall. Which evaluation and threshold strategy should the team apply?

Accepted Answer

Plot the precision-recall curve, select the operating point whose precision is at least 0.98, and among those points pick the one with the highest recall.. The constraint is expressed as a maximum false positive rate among predicted positives, which is precisely precision. A precision-recall curve exposes the achievable recall at each precision level, so selecting the highest-recall point with precision at least 0.98 enforces the business rule while maximizing recall.

Answer

Plot the ROC curve, choose the point closest to the top-left corner, and use its threshold for deployment.

Answer

Train the model with class weights inversely proportional to class frequency and deploy with the default 0.5 threshold.

Answer

Choose the decision threshold that maximizes accuracy on the validation set, then deploy the model with that fixed threshold.

Question 15

A company uses Amazon SageMaker to deploy a model for real-time predictions. The model is updated weekly. The company wants to ensure that the new model version is gradually rolled out to a small percentage of traffic before full deployment, and that it can be rolled back quickly if issues are detected. Which deployment strategy should be used?

Accepted Answer

Canary deployment using SageMaker endpoint variants. Amazon SageMaker endpoint variants support canary deployments, where you can shift a small percentage of traffic to a new model version (e.g., 5%) while the majority remains on the old version. This allows gradual rollout and immediate rollback by simply adjusting the traffic distribution weights or deleting the new variant, meeting the requirement for quick rollback without redeploying.

Answer

Blue/green deployment

Answer

A/B testing with a holdout group

Answer

Rolling deployment across multiple endpoints

Question 16

A data scientist is training a linear regression model using Amazon SageMaker's built-in Linear Learner algorithm. The model's training loss is decreasing but the validation loss starts to increase after a certain number of epochs. The data scientist wants to prevent overfitting and improve generalization. What should the data scientist do?

Accepted Answer

Enable early stopping in the training job to halt training when validation loss stops improving.. Overfitting occurs when a model learns noise in the training data, causing validation loss to increase. Early stopping is an effective regularization technique that halts training at the point of best validation performance. In SageMaker, Linear Learner supports early stopping through hyperparameters like early_stopping_patience and early_stopping_tolerance, which monitor validation loss and stop when no improvement is seen.

Answer

Reduce the size of the training dataset to make the model generalize better.

Answer

Increase the learning rate to help the model converge faster and avoid overfitting.

Answer

Add more features to the dataset to provide additional information to the model.

Question 17

A machine learning specialist is using Amazon SageMaker to train a model on a dataset with a highly skewed target distribution. The specialist wants to evaluate the model's performance using metrics that are robust to class imbalance. Which metric should be used?

Accepted Answer

Area Under the Precision-Recall Curve (AUC-PR). AUC-PR is specifically designed to evaluate performance on the positive class in imbalanced datasets. It summarizes the trade-off between precision and recall, making it sensitive to improvements in minority class prediction. Other metrics like accuracy and AUC-ROC can be overly optimistic when the majority class dominates.

Answer

Mean Squared Error (MSE)

Answer

Accuracy

Answer

Area Under the Receiver Operating Characteristic Curve (AUC-ROC)

Question 18

A machine learning engineer is using SageMaker Pipelines to orchestrate a training workflow. The pipeline includes a processing step that outputs a dataset to S3, followed by a training step that consumes that dataset. The engineer wants to ensure that the training step uses the output from the processing step. How should the engineer configure the training step?

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

Accepted Answer

eval_metrics. The SageMaker built-in K-Means algorithm provides the 'eval_metrics' hyperparameter, which accepts a list of metrics to compute on the test dataset. To use the elbow method, the engineer must enable 'msd' (mean squared distance to centroid) so that the algorithm logs this value for each trained model. By comparing msd across different k values, the engineer can identify the elbow point. Other hyperparameters like k, mini_batch_size, and feature_dim do not control metric computation.

Answer

feature_dim

Answer

k

Answer

mini_batch_size

Question 19

A company is using Amazon SageMaker to train a model and wants to automatically retrain the model every week using new data. Which AWS service should be used to orchestrate the retraining pipeline?

Accepted Answer

AWS Step Functions. AWS Step Functions is the correct choice because it provides a serverless workflow orchestration service that can coordinate multiple AWS services (e.g., SageMaker training jobs, Lambda functions, and data processing) into a state machine. It supports scheduling via Amazon EventBridge (formerly CloudWatch Events) to trigger the pipeline weekly, and it can handle retries, error handling, and parallel execution, making it ideal for automating a retraining pipeline.

Answer

Amazon CloudWatch Events

Answer

AWS Lambda

Answer

AWS Data Pipeline

Question 20

A machine learning engineer is training a gradient boosting model on Amazon SageMaker using the built-in XGBoost algorithm. The dataset contains a highly imbalanced binary target (2% positive). The engineer wants to improve recall without collecting more data. Which SageMaker XGBoost hyperparameter should be adjusted to directly address class imbalance during training?

Accepted Answer

scale_pos_weight. The SageMaker built-in XGBoost algorithm exposes scale_pos_weight specifically to handle skewed binary classification by weighting positive-class gradients. Setting it near the negative-to-positive ratio makes the boosting process pay proportionally more attention to the minority class, which raises recall. Other hyperparameters such as tree count, learning rate, and depth affect capacity and optimization speed but leave the class-wise loss weighting unchanged, so they cannot directly correct the imbalance.

Answer

max_depth

Answer

eta

Answer

num_round