Hank Roark, Data Scientist, H2O
summary
Claim: IoT is about massive sensor data and practical ML wins over fancy transforms. Uses the NASA turbofan dataset (100 train, 100 test, 23 unlabeled sensors) and applies simple signal processing (smoothers/Kalman-style), careful temporal/spatial cross‑validation (hold out 10 engines to avoid autocorrelation), and an H2O GBM grid search plus Random Forests. Resulting model scores 1044 on the competition metric (would place ~6th–7th). Lesson: smoothing + proper validation + GBMs often work and uncertainty modeling matters.
AI-written summary · about
chapters (17)
- 0:00Intro
- 0:40Hanks background
- 1:25IOT
- 2:35Brontobite
- 2:57Zettabyte
- 5:20Jet Engines
- 7:25Simple Things Work
- 8:15Data Scientist
- 9:12Feature Engineering
- 11:30Crossvalidation
- 11:55Takeaways
- 13:35Random Forest
- 16:18Flow User Interface
- 16:52Cross Validation
- 17:13End Result
- 17:48Summary
- 19:04Conclusion
related talks
50mScalable Data Science & Deep Learning with H2O
27mDeploying distributed AI and machine learning in financial services
01h 18mIntroduction to H20 AutoML with Python
01h 10mGetting started with H2O on Python
01h 26mMachine Learning Algorithms with H2o and PowerShell
44mScalable Automatic Machine Learning with H2O
30mScalable Automatic Machine Learning with H2O
01h 39mApplying Rigorous Machine Learning Methods in Business Strategy