Overview
Having an up-close view of the Analytics Industry evolve has been mesmerizing and anxiety-inducing. I love to learn, but keeping pace is pretty damn challenging. I've studied a great deal across statistics, machine learning, deep learning, and Bayesian (my all-time favorite book) over the years.
My primary interest has been and continues to be in Data Mining. How do you generate information quickly & efficiently? How do we create a smart system to alert appropriately? I sense AI agents will be the vanguard of analytics. Hence, the learning continues at pace!
Time Series Project
Placeholder Discuss a pure SQL implemention in BigQuery that preprocessed and detected points of interest for over 2 million times series in a few minutes.
Explainable AI not tutorials
- Explainable AI for Pracitioners
- Interpretable Machine Learning with Python
- Interpretable Machine Learning
- Introduction to Vertex Explainable AI
- InterpretML/ My notes in GitHub are a bit rough
Summary: In short, several Glass-Box models have been developed (e.g. Explainable Boosting Machines) that create interpretable models by design.
The key idea to remember is it's explaining the model, not the data. Processes exist to inspect Black-Box models to discern why a model makes predictions globally and locally (per prediction). The models and techniques are baked into BigQuery and VertexAI, so it's just automated now.
Feature Selection
Summary: Feature selection is the process of removing columns that are not adding additional information to a model. Modern algorithms can take a holistic approach to tease out interactive effects and shared information across several columns.