---

name: eda-visualization description: "Use AutoViz to profile tabular data and generate static or interactive EDA charts." metadata: disco-role: operating disable-model-invocation: true license: Apache 2.0

EDA Visualization

Use this sub-skill when the user wants AutoViz to visualize a CSV, TXT, Excel file, JSON-like file, or pandas DataFrame. This is the primary route for AutoViz_Class, AutoViz, chart formats, plot directories, sampling limits, target variables, and static or interactive plots.

Use this when

• The user asks for one-line EDA, automated visualization, or quick dataset profiling.
• The task names AutoViz_Class, AV.AutoViz, AutoViz_Main, AutoViz_Holo, chart_format, verbose, depVar, or dfte.
• The user needs to choose between png, svg, jpg, bokeh, server, or html outputs.
• The user is debugging missing plots, empty plots, file-loading issues, or output-directory behavior.
• The user wants to know how AutoViz chooses plot families from supervised or unsupervised data.

Core flow

1. Identify whether the input is a filename or a DataFrame.
2. If using a filename, pass the path as filename; if using a DataFrame, pass filename="" and dfte=df.
3. Set depVar to the target column for supervised EDA or "" / None for unsupervised exploration.
4. Pick a chart_format:
   • svg, png, or jpg for matplotlib/static plots.
   • bokeh, server, or html for the HoloViews/Bokeh path.
5. Use max_rows_analyzed and max_cols_analyzed to bound large datasets.
6. Use verbose=2 and save_plot_dir when the user wants files written instead of displayed.
7. After the run, summarize the most useful plot families rather than every internal helper.

Read these references

• references/workflows.md: end-to-end file and DataFrame recipes.
• references/interactive-backends.md: Bokeh/HoloViews/html/server behavior and dependency notes.
• references/troubleshooting.md: common EDA failures and how to recover.
• ../../references/chart-formats.md: compact chart-format matrix.
• ../../references/api-reference.md: exact signatures and helper notes.
• ../../references/install-and-compatibility.md: when the chart path fails because the environment is incomplete.

Use these scripts

• Run scripts/autoviz_smoke.py to verify the static AutoViz path on a safe in-memory DataFrame.
• Run scripts/autoviz_interactive_smoke.py to check that interactive backend imports and the html route are usable.
• Run ../../scripts/inspect_install.py if the environment itself is the problem.

Decision points

• Prefer static svg or png when running headless or inside automated tests.
• Prefer html when the user wants a saved interactive artifact without keeping a server process alive.
• Treat server as a browser/server workflow; do not start a long-running server unless the user asked for it.
• Use a nonempty save_plot_dir for reproducible outputs; otherwise AutoViz creates AutoViz_Plots under the current working directory when saving.
• Explain that AutoViz classifies variables before plotting, so tiny toy data can be classified as IDs, booleans, or categorical columns in surprising ways.
• If a supervised run has many columns, mention that AutoViz may use feature selection to limit the plots.

Typical plot families

• scatter plots for numeric predictors against the target
• pairwise scatter plots for numeric-variable relationships
• distribution plots for target or feature distributions
• violin plots for class or regression comparisons
• heatmaps for numeric correlation inspection
• bar and pivot plots for categorical or mixed data
• date/time plots when date variables are present
• catscatter plots when there are mostly categorical variables
• wordcloud plots when string-like variables are detected

Cross-routing

• If an AutoViz run fails inside pandas_dq or the user asks how to fix dataset issues, switch to ../data-quality-fixes/SKILL.md.
• If string columns trigger wordclouds, NLTK downloads, or text-cleaning questions, switch to ../text-wordclouds/SKILL.md.
• Do not tell future users to open the original repository notebooks. Use this skill's references and bundled scripts instead.
• Keep the response focused on the chosen chart_format and data shape; do not restate every helper function in the package.

Troubleshooting reminders

• If no plots appear, check whether the target column exists and whether AutoViz removed many columns as IDs or low-information features.
• If the run is interactive but nothing renders, try html first and confirm the required dependencies are installed.
• If plots are saved but hard to find, inspect the save_plot_dir argument and the target-named output subdirectory.
• If a tiny fixture gives odd plot families, make the test data more varied before treating the behavior as a bug.