• High Level Questions
• Getting Good Signals
• Managing Training Data
• Things the User Might Want to Know / Fix
• Is any particular training sample any good?
• Do I need to keep collecting data? How much impact has recently-collected data had?
• Are these two classes feasible to distinguish?
• How can I improve the differentiation between these two classes?
• Scores We Can Calculate
• Information Gain of a Sample / Confusion with Other Classes
• Changes to Classifier Distribution from a Training Sample
• Configuring Pipeline
• Evaluation / Testing
• Things the User Might Want to Know / Fix
• False negative: an action was missed by the system
• False positive: an action was detected that shouldn't have been
• Scores We Can Calculate
• Predictions for Live Data
• Class Likelihoods for Live Data
• Class Distances for Live Data
• Scoring test data
• Other Approaches
• Integration Into Project