Using Modes for Time Series Classification
summary
Uses the sample mode as the primary statistic and builds a compact signature from mode strength, run ratio and GT ratio after optionally applying one order of differencing. Claims signatures form vectors that plug into standard ML; tests on historical IT monitoring series separate flats slopes and spikes cleanly. Works fast enough for offline labeling but not yet measured in production; differencing beyond one pass is not needed on provided data.
AI-written summary · about
chapters (11)
- 0:00Introduction
- 1:05Continuous Monitoring
- 1:35Time Series
- 3:05Patterns
- 4:05The Mode
- 5:08Strength
- 5:44Run Ratio
- 7:10Difference Operator
- 8:42Signature
- 11:33Plugging
- 12:05Questions
related talks
48mSignature Methods for Time Series Data
12mModel Monitoring: Detecting and Analyzing Data Issues
35mSignatures, patterns and trends: Timeseries data mining @ Etsy
45mAnomaly Detection in Time Series
41mApplying Foundational Models for Time Series Anomaly Detection
36mDrift Detection on Irregular Time Series with Multiple Non-Uniform Seasonal Patterns
24mMachine learning and financial time series analysis
36mExtracting Relevant Metrics with Spectral Clustering