Everyone wants to be a data scientist and build state of the art Machine Learning models. But for most companies, only a small percentage of models actually go into production. While ML can be used for research and discovery, its true ROI is realized when you make real business decisions autonomously. Kishen describes what to and what not to do when approaching a business problem with an ML lens. He covers steps to take to go from ideation to production and all the fun you can have in between.
Which customers are predicted to spend the most in the next year? Which should I select for retargeting? How often should different customers hear from my brand? Answering these requires segmenting customers but instead of applying ad hoc rules, clustering algorithms can reveal the hidden structure in data to group them on their similarity. Dirichlet Process Mixture Modeling (DPMM) is a nonparametric Bayesian method that also infers the optimal number of clusters. Raghav gives a high-level tour of Bayesian methods and DPMM followed by results on customer transaction data. Walk away with a better understanding of nonparametric Bayesian modeling and a greater appreciation for how flexible modeling with these approaches can help uncover the hidden structure in marketing datasets.
Reinforcement Learning – rewarding the machine for better outcomes – was front and center for Samsung’s email optimization. Venkata developed an agent that intelligently learns from previous campaigns and prescribes target population to maximize engagement (CTR) and thus conversions. Venkata shares how this Prescriptive Engine agent solves the Multi-Arm Bandit problem/Exploration-Exploitation dilemma.
Hundreds of Bachelors and Masters Degree programs, Certificate programs, and bootcamps have sprung up to teach the science behind Data Science, including machine learning algorithms, statistics, and coding in python, R, and SQL. Yet data scientists quickly discover that it takes more to be a data scientist than knowing the science and data doesn’t always cooperate with the analyst! Dean describes five things that are critical to success in building predictive models and creating solutions with machine learning in a business context that are rarely taught in school. Real-world examples show how these principles matter operationally. Hint: as cool as Gradient Boosted Trees, Random Forests, and Deep Learning networks are, none of these principles are related to algorithms.