MLS-C01 Modeling Practice Test 27 — 15 Questions

Question 1

Which THREE of the following are best practices for training a deep learning model on Amazon SageMaker?

Accepted Answer

Use Pipe mode for large datasets to reduce I/O overhead. Option A is correct because Pipe mode streams data directly from Amazon S3 into the training container instead of downloading the full dataset to local disk first, which reduces I/O overhead and startup time for large datasets. Option C is correct because SageMaker automatic model tuning (hyperparameter optimization) runs multiple training jobs with different hyperparameter combinations to find the best-performing model, which is a recommended practice for improving model accuracy. Option D is correct because SageMaker Debugger can capture and analyze system and framework metrics, including GPU utilization, helping identify resource bottlenecks during training. Option B is not correct because SageMaker Debugger detects and reports training issues such as vanishing gradients or overfitting, but it does not automatically fix training errors. Option E is not correct because training on a single instance is not a best practice for large-scale workloads; distributed training across multiple instances is often used to reduce training time, and SageMaker supports it natively.

Answer

Use SageMaker Debugger to automatically fix training errors

Answer

Train on a single instance to avoid distributed training overhead

Question 2

A data scientist is training a neural network on image data using TensorFlow with GPU instances on SageMaker. The training is slow because the GPU utilization is low. The data pipeline uses tf.data with a large number of preprocessing operations. Which action would most likely increase GPU utilization?

Accepted Answer

Increase the prefetch buffer size in the tf.data pipeline.. Low GPU utilization with a heavy tf.data preprocessing pipeline indicates the GPU is starving for data — the input pipeline cannot feed batches fast enough. Increasing the prefetch buffer size allows the CPU-side pipeline to prepare more batches ahead of time and overlap preprocessing with GPU compute, keeping the accelerator busy. This is the standard TensorFlow performance fix for input-bound training.

Answer

Increase the learning rate to converge faster.

Answer

Reduce the batch size to speed up each step.

Answer

Increase the number of CPU instances in the training job.

Answer

Use smaller image sizes to reduce computation.

Question 3

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 4

A data science team is deploying a machine learning model to production using SageMaker. The model is a PyTorch model that requires custom inference logic including image preprocessing. The team needs to ensure that the endpoint can handle variable batch sizes and has low latency. Which deployment approach should the team use?

Accepted Answer

Use SageMaker Inference Pipelines with a preprocessing container followed by the PyTorch model container.. SageMaker Inference Pipelines allow chaining multiple containers — such as a preprocessing container and a model container — behind a single endpoint, which is ideal for custom preprocessing plus PyTorch inference. This approach supports variable batch sizes and low-latency real-time inference, matching the team's requirements.

Answer

Deploy the model as an AWS Lambda function and use API Gateway.

Answer

Use the SageMaker Python SDK's Predictor class with the model artifact.

Answer

Use a SageMaker multi-model endpoint to host the model with a custom container.

Answer

Use SageMaker Batch Transform for real-time inference.

Question 5

A data scientist trains a convolutional neural network (CNN) for image classification. The training loss decreases steadily, but the validation loss starts increasing after 10 epochs. Which technique should the data scientist use to address this problem?

Accepted Answer

Use early stopping to halt training when validation loss stops decreasing.. Early stopping is the appropriate technique because it halts training when validation loss stops decreasing, preventing overfitting. The scenario describes classic overfitting: training loss decreases while validation loss increases after 10 epochs. Early stopping would stop training at the point of best validation performance, saving time and improving generalization.

Answer

Add more data augmentation to the training set.

Answer

Increase the number of training epochs.

Answer

Add more convolutional layers to increase model capacity.

Answer

Increase the learning rate.

Question 6

A data scientist is using Amazon SageMaker to train a deep learning model for natural language processing. The training job is taking too long to converge. The data scientist wants to speed up training without significantly sacrificing model accuracy. Which THREE strategies should the data scientist consider? (Choose three.)

Accepted Answer

Reduce the model size by using fewer layers or smaller hidden dimensions.. Option A is correct because reducing the model size (fewer layers or smaller hidden dimensions) lowers the number of parameters and FLOPs per forward/backward pass, directly decreasing per-epoch training time while typically causing only a modest accuracy trade-off. Option D is correct because mixed precision training with FP16 halves memory bandwidth and storage for activations and weights and leverages NVIDIA Tensor Core throughput, accelerating matrix multiplications on supported GPU instances with minimal accuracy loss when loss scaling is used. Option E is correct because SageMaker distributed data parallelism (e.g., the smdistributed dataparallel library) shards each mini-batch across multiple instances and all-reduces gradients, scaling throughput roughly linearly with instance count while preserving convergence behavior. Option B is not appropriate because multiplying the learning rate by 10 risks divergence or oscillation rather than reliable convergence. Option C is not appropriate because simply maximizing batch size can hurt convergence and generalization and may not fit memory, so it is not a guaranteed speedup without further tuning.

Answer

Increase the learning rate by a factor of 10 to accelerate convergence.

Answer

Increase the batch size to its maximum possible value to utilize GPU memory fully.

Question 7

A data scientist is evaluating a binary classification model that predicts whether a customer will churn. The model achieves an AUC of 0.85 on the test set. Which TWO statements about AUC are correct? (Choose two.)

Accepted Answer

AUC represents the probability that a randomly chosen positive instance is ranked higher than a randomly chosen negative instance.. Option A is correct because AUC (Area Under the ROC Curve) has a well-known probabilistic interpretation: it equals the probability that a randomly selected positive instance receives a higher predicted score than a randomly selected negative instance, which is exactly what this statement describes. Option E is correct because the ROC curve is constructed by plotting TPR against FPR at every possible classification threshold, so AUC aggregates ranking performance across all thresholds rather than depending on any single cutoff. Option B is wrong because an AUC of 0.5 corresponds to random guessing, while 0.85 indicates strong discriminative ability. Option C is wrong because average precision is the area under the precision-recall curve (AP/PR-AUC), a different metric that is more sensitive to class imbalance. Option D is wrong because AUC is threshold-independent and is not equivalent to accuracy at the 0.5 threshold, which depends on a specific decision cutoff.

Answer

An AUC of 0.85 indicates the model is no better than random guessing.

Answer

AUC is the average precision across all thresholds.

Answer

AUC is equivalent to the accuracy of the model at the default threshold of 0.5.

Question 8

A data scientist trains a linear regression model to predict house prices. The model has high bias (underfitting). Which action is most likely to reduce bias?

Accepted Answer

Increase model complexity. Increasing model complexity (e.g., adding polynomial features or using a more flexible algorithm) can reduce bias. Adding L1 regularization increases bias, reducing features reduces complexity, and lowering max_depth for a tree also increases bias.

Answer

Reduce the number of features

Answer

Decrease the maximum depth of the tree

Answer

Add L1 regularization

Question 9

A data scientist builds a Random Forest model using SageMaker. The model performs well on training data but poorly on test data. Which step is most likely to reduce overfitting?

Accepted Answer

Reduce the maximum depth of each tree. Reducing the maximum depth of each tree limits the complexity of individual decision trees, preventing them from memorizing noise and specific patterns in the training data. This directly addresses overfitting by enforcing simpler, more generalized splits, which improves performance on unseen test data.

Answer

Increase the number of trees

Answer

Switch to a linear model

Answer

Increase the number of features considered at each split

Question 10

A data scientist needs to implement a recommendation system for an e-commerce website. Which Amazon service is specifically designed for building and deploying recommendation models?

Accepted Answer

Amazon Personalize. Amazon Personalize is a fully managed machine learning service that provides real-time personalized recommendations. It is purpose-built for recommendation systems. SageMaker is a general-purpose ML platform, but Personalize is specialized for recommendations.

Answer

Amazon SageMaker

Answer

Amazon Rekognition

Answer

Amazon Forecast

Question 11

A data scientist trains a neural network using TensorFlow on SageMaker. The training job fails with a 'CUDA out of memory' error. What is the most likely cause and solution?

Accepted Answer

The model is too large for the GPU. Use a smaller batch size.. CUDA out of memory indicates that the GPU memory is insufficient for the batch size or model size. Reducing the batch size is a common fix. Switching to CPU is not ideal for deep learning. Increasing the number of instances may help but requires distributed training setup. Upgrading to a larger instance type is another option, but reducing batch size is simpler.

Answer

The dataset is too large. Use SageMaker Pipe mode.

Answer

The training script has a bug. Use SageMaker Debugger.

Answer

The instance type is insufficient. Use distributed training across multiple instances.

Question 12

Which TWO metrics are suitable for evaluating a regression model? (Select TWO.)

Accepted Answer

Root Mean Squared Error (RMSE). Root Mean Squared Error (RMSE) (B) is correct because it is a standard regression metric that measures the average magnitude of prediction errors by taking the square root of the mean of squared differences between predicted and actual continuous values, expressed in the same units as the target. R-squared (C) is also correct because it quantifies the proportion of variance in the dependent variable explained by the model, making it a common goodness-of-fit measure for regression. Accuracy (A) is not suitable because it is a classification metric that requires exact matches of discrete labels, which is meaningless for continuous predictions. F1-score (D) and Precision (E) are classification metrics derived from the confusion matrix (true/false positives and negatives) and do not apply to regression tasks.

Answer

Accuracy

Answer

F1-score

Answer

Precision

Question 13

Which THREE techniques help reduce overfitting in a neural network? (Select THREE.)

Accepted Answer

Dropout. Dropout (A) is correct because randomly deactivating a fraction of neurons during each training step prevents the network from relying on any single neuron's co-adaptation, effectively acting as an ensemble and reducing overfitting. L2 Regularization (B) is correct because adding a penalty proportional to the sum of squared weights (lambda * sum(w^2)) to the loss function constrains weight magnitudes, discouraging the model from fitting noise in the training data. Early Stopping (E) is correct because monitoring validation loss and halting training when it stops improving prevents the model from continuing to memorize the training set and overfit. Increasing the number of layers (C) is not correct because adding depth increases model capacity, which typically worsens overfitting rather than reducing it. Using a larger batch size (D) is not correct because batch size primarily affects gradient estimate variance and training speed, not the model's tendency to overfit in the way these regularization techniques do.

Answer

Increasing the number of layers

Answer

Using a larger batch size

Question 14

Which TWO services can be used to serve machine learning models for real-time inference? (Select TWO.)

Accepted Answer

Amazon SageMaker. Amazon SageMaker (C) is correct because it provides fully managed endpoints (real-time inference endpoints) that host trained models and serve low-latency predictions over HTTPS, which is exactly what real-time ML inference requires. AWS Lambda (E) is also correct because it can run inference code in response to events or API Gateway requests, invoking a model (e.g., from S3 or a container image) and returning predictions synchronously with low latency. Amazon Rekognition (A) and Amazon Comprehend (D) are pre-built AI services for specific tasks like image analysis and NLP, not general-purpose services for serving your own ML models for real-time inference. Amazon Athena (B) is an interactive query service for analyzing data in S3 with SQL, not a model-serving service.

Answer

Amazon Rekognition

Answer

Amazon Athena

Answer

Amazon Comprehend

Question 15

A data scientist is training a binary classifier on an imbalanced dataset (95% negative, 5% positive). The model achieves 99% accuracy but only correctly identifies 2% of the positive samples. Which metric should the data scientist focus on to improve the model's performance?

Accepted Answer

Recall. (Recall). Recall measures the proportion of actual positive samples correctly identified, which is critical for an imbalanced dataset where the model fails to detect positives. Option A (Precision) is not the primary focus because it measures the accuracy of positive predictions, not the ability to find all positives. Option B (RMSE) is a regression metric, not suitable for binary classification. Option D (Accuracy) is misleading because a model can achieve high accuracy by simply predicting the majority class, as seen here.

Answer

Precision

Answer

RMSE

Answer

Accuracy