Modeling: free practice questions

Linear Learner is correct because it natively supports binary classification, automatically normalizes input features during preprocessing, provides a 'positive_example_weight_mult' hyperparameter for class imbalance, and is highly optimized for fast training on large tabular datasets. XGBoost is a strong classifier but does not perform automatic feature normalization and requires manual handling of class imbalance (e.g., via 'scale_pos_weight'). K-Means is an unsupervised clustering algorithm and cannot perform supervised binary classification. Random Cut Forest is an unsupervised anomaly detection algorithm, not a binary classifier.

Amazon Bedrock provides fully managed access to large foundation models such as Anthropic Claude and Amazon Titan via a simple API call with no model training or infrastructure management. It supports long-context summarization and can return results within the latency budget for interactive use — making it the best fit. Option A requires training a custom model from scratch, which contradicts 'no model training required,' and the Seq2Seq built-in algorithm has limited context length for 10,000-word documents. Option C uses key phrase extraction, not true abstractive summarization, and the quality would be insufficient for legal documents. Option D (JumpStart BART) requires deploying and managing an endpoint; while it avoids training, it is not 'fully managed' in the same sense as Bedrock and BART models typically have shorter max token limits than what's needed for 10,000-word documents.

With severe class imbalance, accuracy is misleading because a model can achieve high accuracy simply by predicting the majority class. Macro-average F1 averages the F1 score across all classes equally, penalizing models that ignore minority classes. Model C achieves the highest macro-average F1 (0.87), indicating it performs best across all classes including the underrepresented ones. — Model A has a lower macro-F1 than Model C and is therefore inferior in terms of balanced class performance. — Model B has the highest accuracy (91%) but the lowest macro-F1 (0.72), strongly suggesting it is biased toward the majority class and poorly serves minority classes. — Lower accuracy alone does not imply less overfitting to the majority class; the F1 metric is the appropriate evidence, and Model C leads on that metric.

Narrowing search ranges around top-performing configurations focuses the Bayesian optimizer on the most promising region of the hyperparameter space, improving efficiency when the plateau suggests the global region has been identified but not fully exploited. Enabling warm starting with 'IDENTICAL_DATA_AND_ALGORITHM' transfers the surrogate model's learned knowledge from previous tuning jobs to a new job, allowing Bayesian optimization to continue refining rather than starting from scratch. Switching to random search discards all the probabilistic knowledge Bayesian optimization has accumulated; random search is less efficient than Bayesian optimization after many trials. Adding more hyperparameters increases the dimensionality of the search space, making it harder (not easier) to find optima. Increasing parallel jobs only affects throughput/speed; it does not improve the quality of exploration and may actually reduce Bayesian optimization's effectiveness since parallel jobs cannot incorporate each other's results sequentially.

BlazingText in Word2Vec mode (unsupervised) generates high-quality word embeddings that can be used as features for downstream NLP tasks like NER. While BlazingText also has a supervised text classification mode, the Word2Vec mode produces word-level vector representations as explicitly requested. Sequence-to-Sequence is designed for sequence transformation tasks like machine translation and text summarization, not for producing word embeddings or performing NER. Neural Topic Model (NTM) is a topic modeling algorithm that discovers abstract topics in a document corpus; it does not produce word-level embeddings suitable for NER. Latent Dirichlet Allocation (LDA) is also a topic modeling algorithm focused on document-level topic distributions, not word-level representations for sequence labeling tasks.