MLS-C01 Practice Exam 69 — 20 Questions

Question 1

A company deploys a machine learning model on Amazon SageMaker for real-time inference. The model receives requests with large payloads (up to 5 MB) and the inference latency is high. Which configuration change would MOST likely reduce latency?

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

Accepted Answer

Use a larger instance type with more memory and compute. Increasing the instance type to one with more memory and compute directly addresses the bottleneck caused by large payloads (up to 5 MB) and high inference latency. SageMaker real-time endpoints process requests synchronously, so a larger instance provides more CPU/GPU and memory bandwidth to serialize/deserialize and process the payload faster, reducing overall latency.

Answer

Pre-load multiple model containers on the same endpoint

Answer

Reduce the batch size for inference requests

Answer

Enable payload compression using SageMaker built-in compression

Question 2

A company has deployed a machine learning model on a SageMaker endpoint that serves predictions to a web application. The model uses a custom inference container that loads the model artifacts from an ECR repository. After updating the model with new training data, the data scientist creates a new model and updates the endpoint. However, some users report that they still get predictions from the old model. The data scientist confirms that the endpoint configuration points to the new model. What is the most likely cause?

Accepted Answer

DNS caching on the client side is resolving to the old endpoint IP address. DNS caching on the client side can cause the endpoint's DNS name to resolve to an old IP address, especially if the endpoint's underlying instances have not changed. This explains why some users still receive predictions from the old model even though the endpoint configuration points to the new model. Option A is incorrect because incorrect model artifacts would affect all users uniformly. Option B is incorrect because SageMaker endpoints do not have load balancers in the traditional sense; the endpoint itself routes traffic to the instances. Option C is incorrect because SageMaker automatically handles container image updates when a new model is deployed to an endpoint.

Answer

The new model artifacts are not correctly uploaded to S3

Answer

The endpoint is behind a load balancer that is not updated

Answer

The inference container is cached and not pulling the new image

Question 3

A company runs a real-time recommendation system that uses Amazon SageMaker endpoints for inference. The system ingests user activity data from a mobile app via Amazon API Gateway and AWS Lambda, which writes events to an Amazon Kinesis Data Stream. A second Lambda function consumes the stream, calls a SageMaker endpoint to generate recommendations, and stores the results in Amazon DynamoDB. The system has been working well, but recently the team noticed an increase in latency from the time a user action occurs to when the recommendation is stored. The SageMaker endpoint shows increased invocation latency but no throttling. CloudWatch metrics show that the Kinesis stream's IteratorAgeMilliseconds is increasing, indicating the consumer is falling behind. The Lambda consumer's duration is within limits, but the number of invocations is lower than expected. The team suspects the issue is with the event source mapping. Which course of action should the team take to reduce the latency?

Accepted Answer

Increase the batch size in the event source mapping to process more records per invocation.. The increasing IteratorAgeMilliseconds indicates the consumer is falling behind. The Lambda consumer's duration is within limits but the number of invocations is lower than expected, suggesting that the event source mapping is not invoking the function often enough. Increasing the batch size allows each invocation to process more records per invocation, effectively increasing throughput without requiring more invocations. This directly addresses the lag. Option B (increase shards) could help if the consumer had sufficient concurrency, but the root cause is low invocations per shard, not lack of shards. Option C (decrease reserved concurrency) would worsen the problem. Option D (Firehose) does not solve the consumer lag and changes the architecture unnecessarily.

Answer

Increase the number of shards in the Kinesis data stream to increase parallelism.

Answer

Decrease the Lambda function's reserved concurrency to force it to scale down.

Answer

Replace the Lambda consumer with an Amazon Kinesis Data Firehose delivery stream.

Question 4

A company runs a streaming data pipeline using Amazon Kinesis Data Streams with 10 shards. The pipeline ingests sensor data from thousands of devices. Each device sends a JSON payload every 5 seconds. The payload size is approximately 2 KB. The data is consumed by a fleet of EC2 instances running a custom Java application that uses the Kinesis Client Library (KCL). Over the past week, the company has observed that the consumer application is experiencing increased latency, and the Kinesis stream's 'GetRecords.IteratorAgeMilliseconds' CloudWatch metric is consistently above 10 seconds. The company has verified that the EC2 instances have sufficient CPU and memory resources. The KCL application is configured with 10 workers, one per shard. The application processes each record by performing a simple transformation and writing to Amazon DynamoDB. The DynamoDB table has sufficient write capacity and is not throttling. The company wants to reduce the iterator age to under 2 seconds. Which action should the company take?

Accepted Answer

Increase the number of shards in the Kinesis stream to 20. The observed high iterator age indicates that the consumer is falling behind. Since the EC2 instances have sufficient resources and DynamoDB is not throttling, the bottleneck is the number of shards. Each shard provides a fixed amount of read throughput (up to 2 MB/s and 5 reads/s). With 10 shards, the consumer's aggregate read throughput is limited. Increasing the number of shards to 20 doubles the read capacity, allowing the KCL workers to fetch records faster and reduce the iterator age to under 2 seconds. Option A is incorrect because Kinesis Data Firehose is a delivery service that buffers and writes to destinations like S3 or Redshift; it does not support custom transformations with a Java application. Option B is incorrect because the DynamoDB table has sufficient write capacity and is not throttling, so increasing it would not address the read-side bottleneck. Option D is incorrect because KCL enforces one worker per shard; adding more workers than shards does not increase parallelism. The workers would sit idle or conflict, since each shard can only be processed by a single worker at a time.

Answer

Replace Kinesis Data Streams with Amazon Kinesis Data Firehose

Answer

Increase the write capacity of the DynamoDB table

Answer

Increase the number of KCL workers to 20

Question 5

A company runs a real-time analytics platform that ingests IoT sensor data from millions of devices. The data is sent to Amazon Kinesis Data Streams with 16 shards. A custom Java application using the Kinesis Client Library (KCL) processes the data and writes aggregated results to Amazon DynamoDB. The application runs on a fleet of EC2 instances in an Auto Scaling group. Recently, the team noticed that some records are being processed multiple times, resulting in duplicate entries in DynamoDB. The application uses the DynamoDB PutItem API to write records. The team needs to eliminate duplicates without significantly increasing latency. Which solution should the team implement?

Accepted Answer

Use DynamoDB TransactWriteItems with a condition check that the record's Kinesis sequence number does not already exist in the table.. Using a DynamoDB transaction with a condition check on the Kinesis sequence number ensures that each record is written only once. Option A is wrong because increasing write capacity does not address duplicate processing; duplicates arise from the KCL consumer processing records multiple times, not from throttling. Option C is wrong because while SQS FIFO provides deduplication at the queue level, it does not guarantee exactly-once processing downstream in DynamoDB; the consumer could still write duplicates if it fails after writing but before deleting the message. Additionally, adding an extra queue increases latency and complexity. Option D is wrong because BatchWriteItem does not decrease duplicates; it only batches multiple put requests into one API call and still requires idempotency measures.

Answer

Enable DynamoDB auto scaling to increase write capacity and reduce throttling, which causes retries and duplicates.

Answer

Place an Amazon SQS FIFO queue between the KCL application and DynamoDB to deduplicate messages.

Answer

Modify the application to use DynamoDB BatchWriteItem instead of PutItem to reduce the number of write requests.

Question 6

A data engineer is responsible for managing a data lake on Amazon S3. The data lake contains CSV files from various sources, totaling 10 TB. The engineer needs to make this data queryable using Amazon Athena. However, Athena queries are currently taking a long time and scanning large amounts of data. The engineer has noticed that the CSV files are not partitioned, and there are no indexes. The engineer wants to improve query performance and reduce costs. The data is accessed frequently for the last 30 days, but older data is rarely queried. The engineer also wants to minimize the amount of data scanned by Athena. What should the engineer do?

Accepted Answer

Convert the CSV files to Parquet format and partition the data by date.. The best choice. Converting CSV to Parquet reduces data scanned due to columnar storage and compression. Partitioning by date allows Athena to skip older data that is rarely queried, further minimizing scan size and cost. Option A (JSON) does not improve performance significantly and still lacks partitioning. Option C is invalid because Athena does not support indexes. Option D (ORC) is columnar but without partitioning it performs worse than Parquet with partitioning, and views do not reduce scan size.

Answer

Convert the CSV files to JSON format and use Athena to query them.

Answer

Create indexes on the S3 objects using AWS Glue.

Answer

Convert the CSV files to ORC format and create a view in Athena.

Question 7

A machine learning engineer is analyzing a dataset for a regression problem. The target variable has a long-tail distribution with extreme outliers. The engineer wants to reduce the influence of outliers while preserving the relative order of values. Which data transformation should the engineer apply to the target variable?

Accepted Answer

Rank transformation. The rank transformation is correct because it maps each value to its rank in the dataset, preserving the relative order of values while eliminating the impact of magnitude differences. This effectively reduces the influence of extreme outliers without distorting the ordinal relationships. Option A (min-max normalization) is incorrect because it linearly scales values to a fixed range, and outliers can still dominate the scaling. Option B (Box-Cox transformation) can reduce skew but requires positive values and does not fully remove outlier influence; it transforms the distribution shape but still allows extreme values to affect the transformation parameters. Option D (log transformation) reduces right skew but remains monotonic, so extreme high values still have a disproportionate effect on the model compared to rank transformation.

Answer

Min-max normalization

Answer

Box-Cox transformation

Answer

Log transformation

Question 8

A data scientist is performing exploratory data analysis on a large dataset stored in Amazon S3 (100 GB, CSV format, 500 columns). The dataset contains customer transaction records with features such as transaction amount, timestamp, customer ID, and numerous categorical variables (e.g., product category, payment method, location). The scientist wants to understand the distribution of transaction amounts across different product categories and identify any outliers. They have an Amazon SageMaker notebook instance with a ml.t3.medium instance and are using pandas. However, when trying to load the entire dataset into a DataFrame using pd.read_csv('s3://bucket/data.csv'), the notebook crashes with a memory error. Additionally, the scientist suspects that some categorical columns have high cardinality (e.g., product category has thousands of unique values), and there are missing values in several columns. What is the MOST efficient approach to perform the EDA without modifying the original dataset or using additional AWS services? Options: A) Use the SageMaker SDK to launch a parallel processing job with PySpark and read the data into a Spark DataFrame, then compute statistics and visualize with matplotlib. B) Use pandas with chunksize parameter to iterate through the dataset in chunks, compute per-chunk statistics, and aggregate results; for high-cardinality columns, use value_counts() with dropna=False and then plot the top 20 categories. C) Use the S3 Select API to filter rows and columns before loading into pandas, reducing the data size; then use pandas for EDA. D) Use SageMaker Data Wrangler to import the dataset, create a flow to handle missing values and reduce cardinality, and export a sample to the notebook for analysis.

Accepted Answer

Use pandas with chunksize parameter to iterate through the dataset in chunks, compute per-chunk statistics, and aggregate results; for high-cardinality columns, use value_counts() with dropna=False and then plot the top 20 categories.. It addresses the memory issue by reading the data in chunks using the chunksize parameter, allowing processing without loading the entire dataset into memory. It computes per-chunk statistics and aggregates them, which is efficient for EDA. For high-cardinality categorical columns, it uses value_counts() with dropna=False to capture missing values, and then plots the top 20 categories, which is manageable and insightful. This approach stays within pandas and the existing SageMaker notebook without requiring additional services or changing the dataset. Option A is incorrect because launching a separate PySpark job adds complexity and extra cost, and is not the most efficient for an ad-hoc EDA. Option B (S3 Select) can reduce the data volume but cannot natively perform complex aggregations like distribution across categories without pulling all rows; it is more suited for simple filtering. Option C (SageMaker Data Wrangler) is a separate service that requires additional setup and is overkill for this simple EDA task; it also modifies the workflow and is not the most efficient for immediate analysis.

Answer

Use the SageMaker SDK to launch a parallel processing job with PySpark and read the data into a Spark DataFrame, then compute statistics and visualize with matplotlib.

Answer

Use the S3 Select API to filter rows and columns before loading into pandas, reducing the data size; then use pandas for EDA.

Answer

Use SageMaker Data Wrangler to import the dataset, create a flow to handle missing values and reduce cardinality, and export a sample to the notebook for analysis.

Question 9

A data scientist is analyzing a dataset with a binary target variable. They compute the correlation matrix and find that all features have correlations between -0.1 and 0.1 with the target. They suspect that the relationship might be non-linear. Which of the following techniques should they use to detect non-linear relationships?

Accepted Answer

Mutual information. Mutual information is the correct technique because it measures the dependency between variables and can capture any type of relationship, including non-linear and non-monotonic. Options A and C are incorrect: ANOVA is used for comparing means of categorical vs continuous variables, and Pearson correlation only measures linear relationships. Option B is also incorrect: Spearman's rank correlation captures monotonic relationships but may miss other non-linear patterns.

Answer

ANOVA test

Answer

Spearman's rank correlation

Answer

Pearson correlation coefficient

Question 10

A company is building a data pipeline that uses Amazon Kinesis Data Streams to ingest real-time events. The pipeline then uses AWS Lambda to process the events and store results in Amazon DynamoDB. The company wants to ensure that the Lambda function can process all events without data loss and without duplicating processing. Which TWO configuration steps should the company take?

Accepted Answer

Increase the data retention period of the Kinesis stream to 7 days to allow reprocessing. To ensure all events are processed without data loss and without duplicating processing, the company should increase the Kinesis stream retention to 7 days (option A). This allows reprocessing of events that fail initial processing, thus preventing data loss. The company should also configure the Lambda function to disable retries on failure (option E). Automatic retries can cause the same batch of events to be processed multiple times, leading to duplication unless the Lambda function is idempotent. By disabling retries, the company can handle failures manually, ensuring exactly-once processing when combined with idempotent reprocessing logic. Option B (reducing batch window) does not prevent duplicates and may increase invocations. Option C (enabling iterator age metric) is for monitoring, not preventing loss/duplication. Option D (single shard) does not prevent duplication and can cause throughput limitations.

Answer

Set the Lambda function's batch window to a small value (e.g., 1 second) to reduce processing latency

Answer

Enable the 'iterator age' metric in Amazon CloudWatch to monitor consumer lag

Answer

Use a single shard for the Kinesis stream to ensure order and avoid parallel processing issues

Question 11

A financial services company is building a fraud detection model using a large dataset of credit card transactions. The dataset contains 10 million rows with 50 features, including transaction amount, merchant category, time of day, and customer historical features. The label is binary: fraudulent (1% of data) or legitimate. The company wants to deploy a real-time inference endpoint using Amazon SageMaker that can score transactions with sub-100ms latency. The current model is a gradient boosting model (XGBoost) trained on a sample of 1 million rows due to memory constraints. The model achieves 0.95 AUC on a held-out test set but the fraud recall (sensitivity) is only 0.4, which is unacceptable because the cost of missing a fraud is high. The data science team has access to a larger compute instance (ml.m5.24xlarge) for training. Which course of action is most likely to improve fraud recall while maintaining latency requirements?

Accepted Answer

Train the XGBoost model on the full 10 million rows using an ml.p3.2xlarge instance with GPU support, and apply SMOTE oversampling to the minority class before training.. Training on the full 10 million rows with a GPU-accelerated instance (ml.p3.2xlarge) allows the XGBoost model to learn from the complete data distribution, addressing the bias introduced by the 1 million row sample. Applying SMOTE oversampling to the minority class (fraud) directly tackles the class imbalance (1% fraud), which is the root cause of the low recall (0.4). SMOTE generates synthetic fraudulent examples, improving the model's ability to detect fraud without significantly increasing inference latency, as the model architecture and deployment remain unchanged.

Answer

Engineer additional features from transaction time and merchant category, then retrain the XGBoost model on the same 1 million row sample.

Answer

Downsample the majority class to 1% of the original size to create a balanced dataset of 200,000 rows, then retrain the XGBoost model on this balanced sample.

Answer

Replace XGBoost with a logistic regression model trained on the full dataset, as linear models are faster to train and may generalize better on large data.

Question 12

A data scientist wants to identify outliers in a dataset with 1,000 samples and 5 numerical features. Which technique is most appropriate for univariate outlier detection?

Accepted Answer

Interquartile range (IQR) method. The IQR method, where outliers are defined as points below Q1 - 1.5*IQR or above Q3 + 1.5*IQR, is appropriate for univariate outlier detection as it does not assume a specific distribution and is robust to extreme values. PCA (A) is a dimensionality reduction technique, not for outlier detection. Mahalanobis distance (C) is for multivariate outliers. Z-score with threshold 3 (D) assumes normality and is sensitive to extreme outliers.

Answer

Principal component analysis (PCA)

Answer

Mahalanobis distance

Answer

Z-score with a threshold of 3

Question 13

A company is using Amazon SageMaker to train a model on a dataset with many categorical features. They want to use SageMaker's built-in Linear Learner algorithm. What preprocessing step is required for the categorical features?

Accepted Answer

Apply one-hot encoding to convert them to numerical vectors.. The SageMaker Linear Learner algorithm requires numerical input features. Categorical features must be converted to numerical vectors, typically via one-hot encoding, because the algorithm performs linear regression or classification on numerical data. Without this preprocessing, the algorithm cannot interpret categorical values directly.

Answer

Use label encoding to assign integers to categories.

Answer

Normalize the categorical features using min-max scaling.

Answer

Remove categorical features with high cardinality.

Question 14

A data scientist is building a binary classification model to predict loan default. The dataset is highly imbalanced (5% default, 95% non-default). Which TWO techniques should the data scientist use to address the class imbalance?

Accepted Answer

Oversample the minority class using SMOTE. Oversampling the minority class using SMOTE (Synthetic Minority Oversampling Technique) is correct because it generates synthetic samples for the minority class by interpolating between existing minority instances, rather than simply duplicating them. This helps balance the dataset without introducing exact copies, which can reduce overfitting and improve the model's ability to generalize to the minority class.

Answer

Undersample the majority class

Answer

Use RMSE as the evaluation metric

Answer

Use accuracy as the evaluation metric

Question 15

A data engineer has attached the above IAM policy to an IAM role used by an AWS Glue ETL job. The job reads from and writes to 'my-data-bucket'. The job is failing with an Access Denied error. What is the most likely cause?

Accepted Answer

The condition restricts access to a specific IP range that does not include the AWS Glue service IPs.. The IAM policy includes a condition that restricts access to requests originating from a specific IP address range. AWS Glue ETL jobs run on ephemeral compute resources that use a dynamic pool of IP addresses, which are not guaranteed to fall within any fixed customer-managed IP range. Therefore, the condition causes the Access Denied error because the Glue service IPs are not within the allowed range.

Answer

The IAM role needs to have s3:ListBucket permission.

Answer

The IAM role does not have permission to list the bucket.

Answer

The resource ARN should include the bucket itself, not just the objects.

Question 16

A company uses Amazon Kinesis Data Firehose to deliver streaming data to Amazon S3. They notice that the data is delivered in 5-minute intervals even though they set the buffer interval to 60 seconds. What could be the cause?

Accepted Answer

The buffer size is set to a value larger than the incoming data rate.. B is correct because Kinesis Data Firehose delivers data based on whichever condition is met first: the buffer interval (60 seconds) or the buffer size (e.g., 5 MB). If the incoming data rate is very low, the buffer size threshold may never be reached within 60 seconds, causing Firehose to wait longer—up to the maximum buffer interval of 900 seconds—before delivering. In this case, the data rate is so low that it takes 5 minutes to fill the buffer, overriding the 60-second interval setting.

Answer

The source Kinesis stream has insufficient shards.

Answer

The S3 bucket is in a different region.

Answer

The IAM role does not have permission to write to S3.

Question 17

A data science team is building a real-time fraud detection system. Transactions are streamed via Amazon Kinesis Data Streams, and a Lambda function performs feature engineering and invokes an Amazon SageMaker endpoint for predictions. The team notices that the Lambda function is timing out and causing data loss. Which solution should the team implement to process the stream reliably and at low latency?

Accepted Answer

Use Amazon Kinesis Data Analytics for Apache Flink to consume the stream, perform feature engineering, and invoke the SageMaker endpoint with exactly-once processing.. Amazon Kinesis Data Analytics for Apache Flink provides a stateful, low-latency stream processing engine that can consume from Kinesis Data Streams, perform feature engineering in real-time, and invoke SageMaker endpoints with exactly-once processing semantics. This eliminates Lambda timeouts and data loss by using a long-running, scalable application instead of a short-lived function.

Answer

Use the Kinesis Client Library (KCL) to process the stream in an Amazon EC2 instance, and store the predictions in Amazon DynamoDB.

Answer

Increase the Lambda function timeout to 15 minutes and allocate more memory to reduce processing time.

Answer

Configure Amazon Kinesis Firehose to deliver the stream to an Amazon S3 bucket, then trigger a Lambda function to process the data in batches.

Question 18

Refer to the exhibit. A data scientist examines a sample of data and notices that all columns are numeric. The scientist wants to check for multicollinearity. Which statistic should be computed from this sample?

Accepted Answer

Correlation matrix (Pearson). The correlation matrix shows pairwise Pearson correlations, which can indicate high collinearity. Option B is wrong because chi-square is for categorical variables, not numeric. Option C is wrong because VIF requires more variables than observations (or typically used after regression). Option D is wrong because covariance alone is scale-dependent.

Answer

Chi-square test of independence

Answer

Variance Inflation Factor (VIF)

Answer

Covariance matrix

Question 19

A machine learning engineer is evaluating a multi-class classification model that predicts product categories. The model outputs probabilities for 10 classes. The engineer wants to improve the model's calibration so that the predicted probabilities reflect the true likelihood of each class. Which THREE techniques can help?

Accepted Answer

Use temperature scaling. Platt scaling and isotonic regression are common calibration methods for classification models. Temperature scaling is a variant of Platt scaling for neural networks. Using a different loss function like cross-entropy helps but is not a calibration technique per se.

Answer

Increase model complexity

Answer

Use focal loss

Question 20

A data scientist is analyzing a dataset with many features and wants to identify which features are most correlated with the target variable. Which EDA technique should be used?

Accepted Answer

Correlation matrix. A correlation matrix displays pairwise Pearson correlation coefficients between all numeric features and the target variable, enabling quick identification of the most correlated features. Option A (box plots grouped by target) is useful for visualizing feature distributions across target categories but does not directly measure correlation strength. Option B (scatter plot matrix) can show pairwise relationships but becomes impractical with many features and lacks a single quantitative correlation measure. Option C (histogram) only shows the distribution of a single feature, not its relationship with the target. Therefore, the correlation matrix is the appropriate EDA technique for identifying features most correlated with the target.

Answer

Box plots grouped by target

Answer

Scatter plot matrix

Answer

Histogram of each feature