Engineer Features for Models SNOWPRO-ADVANCED-DATA-SCIENTIST Practice Questions (Page 6)

Questions 26–30

1. 26application · medium

   A data scientist has trained a random forest model with 200 features and wants to reduce the feature set for a simpler, faster model. The scientist needs a ranking of feature importance that accounts for both individual feature performance and interactions with other features. Which method should be used?

   • ACompute the Pearson correlation of each feature with the target and rank by absolute value.
   • BRank features by their variance and keep the top 50.
   • CUse a filter method like chi-squared test on each feature independently.
   • DUse the random forest's built-in feature_importances_ attribute.
   Reveal Solution Discussion Select an answer first
2. 27application · medium

   A data scientist is building a sentiment analysis model on product reviews. The scientist wants to capture the semantic meaning of words beyond simple frequency counts. The model will be a neural network. Which text representation is most appropriate?

   • ATF-IDF vectors of the reviews.
   • BPre-trained word embeddings like Word2Vec or GloVe.
   • CCharacter-level n-gram counts.
   • DBag-of-words with binary presence indicators.
   Reveal Solution Discussion Select an answer first
4. 28foundation · easy

   What is an interaction feature?

   • AA feature that is imputed using the mean of the column
   • BA feature that is selected using a filter method
   • CA feature that is scaled to have zero mean and unit variance
   • DA feature created by combining two or more original features to capture their joint effect
   Reveal Solution Discussion Select an answer first
5. 29foundation · easy

   Which text representation technique creates a vector of word counts for each document?

   • ATF-IDF
   • BBag-of-words
   • CWord embeddings (e.g., Word2Vec)
   • DOne-hot encoding of words
   Reveal Solution Discussion Select an answer first
6. 30application · medium

   A data scientist is building a spam classifier using a linear SVM on a corpus of 100,000 emails. The scientist wants to represent the text as numeric features while down-weighting common words that appear in almost every document. Which text representation should be used?

   • AWord2vec embeddings averaged across all words in each email.
   • BBag-of-words with raw term counts for all unique terms.
   • CTF-IDF with a vocabulary limited to the top 10,000 terms by document frequency.
   • DBinary bag-of-words indicating only presence or absence of each term.
   Reveal Solution Discussion Select an answer first