How this hub works: the vocabulary, grammar, and interview links below are hand-curated for Data Scientist & ML Engineers. The blog section is a live query — new articles tagged for this role appear here automatically as they're published, no manual edit needed. Prefer a structured essay instead? Read the full Data Scientist & ML Engineer guide. Or see the Data Scientist & ML Engineer learning path for a scenario-based walkthrough.

Vocabulary sets

Grammar & writing

Interview prep

Related deep-dive hub

Blog articles (3)

Other role hubs

Explore more

Browse every exercise category, or search the full site.

All exercises All role hubs Search the site

Frequently Asked Questions

What's the difference between feature engineering and feature selection, specifically in a time series context?

Feature engineering involves creating new features from existing ones – for example, calculating rolling averages or lagged values in time series data. Feature selection focuses on choosing the most relevant subsets of original features. In time series, you might engineer features like exponentially weighted moving averages to capture trends and then select those that best predict future values based on statistical significance.

I'm building a recommendation system; how do I handle cold start problems effectively with collaborative filtering?

Cold start occurs when you have little or no interaction data for new users or items. Techniques like content-based filtering (using item metadata) can provide initial recommendations based on similarity to existing user preferences. Hybrid approaches, combining both collaborative and content-based methods, often mitigate cold start issues by leveraging available information.

Can you explain the concept of 'drift' in model monitoring for deployed ML systems?

Drift is the change in data characteristics over time which impacts model accuracy. It can be feature drift – changes in individual features' distributions – or concept drift – a shift in the relationship between input features and the target variable. Monitoring these drifts allows for proactive model retraining to maintain performance.