How to Design Feature Vectors for Model Inputs and
summary
Covers how to choose and design input feature representations and output formats for machine learning models, with concrete examples across different problem areas. Discusses problems caused by poor feature or output representations (e.g., slow or failed convergence) and shows creative, practical strategies data scientists can use to produce well-performing models.
AI-written summary · about
chapters (18)
- 0:00Intro
- 1:54Common View of Machine Learning
- 8:07Remember
- 10:50Model Selection is important
- 11:48Convolution
- 15:49Time Series / Prediction
- 19:47Example with Categorical Values
- 22:32Imprecision with Categorical Values
- 24:31One-Hot Encoding
- 26:39Mathematically...
- 29:28Preparation
- 31:47Normalization
- 35:30Transforms
- 40:42Variable-Width Input
- 43:24Encoding using RNNS
- 45:37Engineering Features
- 48:37Dimensionality Reduction
- 52:09Model Output
related talks
39mFeature Engineering: Unlocking Potential for Predictive Models
48mThe Missing Step in Moving from Data to Models
43mDesigning our own in-house Feature Platform
50mVector Ops: run vector embedding-powered apps in prod
31mLife cycle of data becoming a model
02h 17mFeature Engineering is for Everyone!
30mHow Feature Stores Enable Operational ML
43mDesigning for Guidance in Machine Learning