Visualising Machine Learning Models
summary
Shows concrete techniques and a small prototype that actually plots models against data, inspects parameter sweeps and fitting dynamics, and maps decision boundaries. Uses Python notebooks and matplotlib seaborn bokeh to render examples: a toy trading strategy (1–10% thresholds, one‑minute vs two‑minute rules) and the UCI wine set (≈1600 rows, 11 chemical features, quality 3–8). Demonstrates visualise model in data space and mesh of decision boundaries, produces residual plots and projection‑based views; toolkit is early but usable in Jupyter.
AI-written summary · about
chapters (21)
- 0:00Visualizing ML models
- 2:55Blind men and data science
- 6:05Three layers of data visualization
- 8:12Trading model in data space
- 10:43Model performance across parameters
- 13:16Five aspects of model visualization
- 15:48ML process and visualization gaps
- 19:21Linking visualization to development
- 21:11Wine quality dataset exploration
- 23:41Alcohol and wine quality
- 26:31Predicted data in further exploration
- 27:53Model residuals and interactions
- 30:04Wine quality classification
- 31:30Projection pursuit for understanding
- 32:15Decision boundaries in models
- 34:42Model boundaries with mesh grids
- 35:47Automation versus human insight
- 39:01Visualization for prediction explanations
- 40:26Dimensionality in high-dimensional data
- 41:49Tools for interactive visualization
- 42:17Closing remarks
related talks
22mVisualizing machine learning algorithms in the Jupyter notebook
39mVisual diagnostics for more informed machine learning
01h 38mEffective Data Visualization
43mAdvanced Data Visualizations In Jupyter Notebooks
02h 23mEffective Data Visualization
35mVisual Diagnostics for More Effective Machine Learning
01h 28mPython Visualization for Exploration of Data
21mTurn Data into Action: How visualization aids decision intelligence