My Three Ex’s: A Data Science Approach for Applied Machine Learning

Today, I gave a talk at QCon SF entitled “My Three Ex’s: A Data Science Approach for Applied Machine Learning”. The talk wasn’t about machine learning as such, but rather about applying machine learning to solve problems.

Hence my three ex’s:

Express: Understand your utility and inputs.

• Choose an objective function that models utility.
• Be careful how you define precision.
• Account for non-uniform inputs and costs.
• Stratified sampling is your friend.
• Express yourself in your feature vectors.

Explain: Understand your models and metrics.

• Accuracy isn’t everything.
• Less is more when it comes to explainability.
• Don’t knock linear models and decision trees!
• Start with simple models, then upgrade.

Experiment: Optimize for the speed of learning.

• Kiss lots of frogs: experiments are cheap.
• But test in good faith – don’t just flip coins.
• Optimize for the speed of learning.
• Be disciplined: test one variable at a time.

I peppered the talk with examples from my experiences working on search quality at LinkedIn and Google:

• Modeling search quality and searcher effort.
• Mapping local businesses to their official home pages.
• Segmenting search models based on searchers and queries.
• Automatically rewriting search queries to improve relevance.
• Entity-based search suggestions.

For those who weren’t able to hear the talk live, I hope the slides prove useful. And I’ll share the video as soon as it’s available.

Finally, I’d like to thank Brendan Collins, Gloria Lau, and Monica Rogati — three of my favorite ex-coworkers, to whom I had the pleasure to dedicate this talk. I learned so much from working with all of you, and I look forward to working with you again someday.