Visualization Highlights Summary
Every highlight bullet from the 9 pages in this section, gathered on one page and grouped by the page it came from.
- Every chart is a mapping from data columns to visual encodings - position, color, size, shape
- Imperative libraries (matplotlib) draw artists; declarative ones (Altair) describe mappings and let a renderer draw
- Tidy, long-format data is what makes the grammar-of-graphics model work
- Audience and delivery medium (static report vs. interactive dashboard) should pick the library before chart type does
- Match chart type to the analytic question
- Label axes and units - charts lie by omission
- Start with matplotlib for control; layer seaborn for stats
- Save figures at dpi suited to medium (screen vs print)
- Figure is canvas; Axes is where data draws
- Use fig, ax = plt.subplots() in production code
- tight_layout prevents label clipping
- Close figures in loops to avoid memory leaks
- data= parameter expects tidy DataFrames
- set_theme gives consistent aesthetics
- hue/size/style encode extra dimensions
- errorbar replaces deprecated ci= in modern seaborn
- Plotly Express maps DataFrame columns to roles
- write_html embeds without a server
- Hover templates customize tooltips
- Downsample before plotting millions of points
- Charts compose from mark + encode + data
- Vega-Lite JSON serializes for git review
- transform_* keeps logic in the spec
- Aggregate before plotting large datasets
- Streamlit is fastest for data scientist prototypes
- Dash fits multi-page Plotly enterprise apps
- Gradio targets ML model demos with inputs/outputs
- Cache expensive loads with st.cache_data
- Never send a million points to the browser raw
- Aggregate or bin first; plot the summary
- datashader rasterizes huge scatter to images
- LTTB preserves visual shape when downsampling lines
- Start y-axis at zero for bar charts unless log scale is labeled
- Color is not the only channel - use labels and shapes
- Title states the insight; axes state units
- Aggregate before plotting large datasets
Revisado por Chris St. John·Última actualización: 31 jul 2026