Python Visuals in Power BI (PBIR)

Report modification requires tooling. Two paths exist:

1. pbir CLI (preferred) -- use the pbir command and the pbir-cli skill. Install with uv tool install pbir-cli or pip install pbir-cli. Check availability with pbir --version.
2. Direct JSON modification -- if pbir is not available, use the pbir-format skill (pbip plugin) for PBIR JSON structure and patterns. Validate every change with jq empty <file.json>.

If neither the pbir-cli skill nor the pbir-format skill is loaded, ask the user to install the appropriate plugin before proceeding with report modifications.

Python visuals execute matplotlib/seaborn scripts to render static PNG images on the Power BI canvas. Prefer seaborn over raw matplotlib for cleaner syntax and better defaults -- it handles most chart types with less code.

Visual Identity

• visualType: pythonVisual
• Data role: Values (columns and measures, multiple allowed)
• Data variable: dataset (pandas DataFrame, auto-injected)
• Row limit: 150,000 rows
• Output: Static PNG at 72 DPI -- no interactivity

Workflow: Creating a Python Visual

Step 1: Add the Visual

Create the visual.json file manually (see pbir-format skill in the pbip plugin for JSON structure) with visualType: pythonVisual, field bindings for the columns and measures you need (use Values:Table.Column or Values:Table.Measure format), and position/size as required.

Step 2: Write the Script

import matplotlib.pyplot as plt

fig, ax = plt.subplots(figsize=(8, 4))
ax.bar(dataset["Date"], dataset["Sales"], color="#5B8DBE")
ax.spines["top"].set_visible(False)
ax.spines["right"].set_visible(False)
plt.tight_layout()
plt.show()  # MANDATORY

Critical rules:

• plt.show() is mandatory as the final line -- nothing renders without it
• dataset is auto-injected as a pandas DataFrame; do not create it
• Column names match the nativeQueryRef (display name) from field bindings
• Only the last plt.show() call renders; multiple figures not supported

Step 2b: Review

Before presenting the script to the user, dispatch the python-reviewer agent to validate correctness and provide design feedback.

Step 3: Inject the Script

Set the script content in the visual's objects.script[0].properties.source literal value (see PBIR Format section below).

Escaping rules for visual.json injection:

The script must be encoded as a single-quoted DAX literal string inside expr.Literal.Value:

• Newlines in the script become \n in the JSON string
• Double quotes inside the script (e.g., "#5B8DBE") become \" in the JSON string
• The entire script is wrapped in single quotes: 'import matplotlib...\nplt.show()'
• See examples/visual/ for a complete real-world visual.json showing this encoding

Step 4: Validate

Validate JSON syntax with jq empty <visual.json> and inspect the visual.json to confirm script content and field bindings.

PBIR Format

Scripts are stored in visual.objects.script[0].properties:

{
  "source": {"expr": {"Literal": {"Value": "'import matplotlib.pyplot as plt\\n...\\nplt.show()'"}}},
  "provider": {"expr": {"Literal": {"Value": "'Python'"}}}
}

The CLI handles all escaping automatically.

Supported Libraries

Power BI Service (Python 3.11)

Not supported: plotly, bokeh, altair (networking blocked in Service).

Full package list: https://learn.microsoft.com/power-bi/connect-data/service-python-packages-support

Desktop

Any locally installed package works without restriction.

Best Practices

1. Always call plt.show() -- mandatory, must be the final line
2. Use figsize=(w, h) to match container aspect ratio (72 DPI output)
3. Remove chart chrome -- ax.spines["top"].set_visible(False) etc.
4. Use hex colors matching the report theme
5. Keep scripts simple -- 5-min timeout Desktop, 1-min Service
6. Minimize transforms -- do heavy computation in DAX/Power Query instead
7. Use try/except for robustness in production scripts
8. Copy data first -- data = dataset.copy() before manipulation

Limitations

Script Structure Template

import matplotlib.pyplot as plt
import numpy as np

# 1. Guard against empty data
if dataset.empty:
    fig, ax = plt.subplots(1, 1, figsize=(6, 4))
    ax.text(0.5, 0.5, "No data available", ha='center', va='center', fontsize=14, color='#888888')
    ax.axis('off')
    plt.show()
else:
    # 2. Data preparation (dataset is auto-injected)
    data = dataset.copy()

    # 3. Create figure with explicit size
    fig, ax = plt.subplots(figsize=(8, 4))

    # 4. Plot
    ax.plot(data["X"], data["Y"], color="#5B8DBE", linewidth=2)

    # 5. Style
    ax.spines["top"].set_visible(False)
    ax.spines["right"].set_visible(False)
    ax.grid(axis="y", alpha=0.3)

    # 6. Layout and render
    plt.tight_layout()
    plt.show()

When to Use a Script Visual

Reach for a Python visual only when all of the following hold:

• The chart has no native equivalent and no reasonable Deneb spec
• The value is in a statistical computation that must run at render time (model fit, kernel density, forecast band), not just a shape Vega could draw
• The visual does not need to be a cross-filter source, hover tooltips, publish-to-web, or app-owns-data embed
• The report is served in a Pro/PPU or higher capacity with a Fabric-enabled region

If interactivity or cross-filtering matters, use Deneb (a static PNG cannot be a selection source). If the need is a small inline mark (sparkline, bar, status pill), use an SVG measure (no row cap, no timeout, no licensing/region gate, renders under publish-to-web). The script visual's niche is narrow: compute-at-render statistical plots for internal or org consumption.

Python vs R once a script visual is the right call: use Python when the computation leans on scikit-learn, statsmodels, or scipy, or when surrounding report logic is already Python. Use R for publication-quality statistical defaults and packages with no Python peer (forecast, corrplot, pheatmap, ridgeline/violin). Where equal, default to whichever language the report's other scripts use; mixing doubles the publish-time package surface to validate.

Do not default to a script visual because a chart type "looks statistical." A box plot, lollipop, or dumbbell is an SVG-measure or Deneb job; reserve scripts for charts that genuinely compute.

References

• references/data-model.md -- dataset grouping mechanic, the row/byte caps, and how to force per-row input
• references/community-examples.md -- seaborn gallery examples organized by chart type, plus matplotlib and Python Graph Gallery links
• references/chart-patterns.md -- Common matplotlib/seaborn chart patterns (bar, heatmap, donut, KPI, area)
• examples/script/ -- Standalone Python scripts (bar-chart, trend-line) -- ready to inject into visual.json after escaping
• examples/visual/bar-chart.json -- PBIR visual.json: horizontal stacked bar with PY comparison lines and % change labels
• examples/visual/kpi-card.json -- PBIR visual.json: text-based KPI with value, % change indicator, and PY comparison
• examples/visual/trend-line.json -- PBIR visual.json: area chart with line plot and monthly x-axis

Fetching Docs

To retrieve current Python visual / package support docs, use microsoft_docs_search + microsoft_docs_fetch (MCP) if available, otherwise mslearn search + mslearn fetch (CLI). Search based on the user's request and run multiple searches as needed to ensure sufficient context before proceeding.

Related Skills

• pbi-report-design -- Layout and design best practices
• r-visuals -- R Script visuals (same concept, different language)
• deneb-visuals -- Vega/Vega-Lite visuals (interactive, vector-based alternative)
• svg-visuals -- SVG via DAX measures (lightweight inline graphics)
• pbir-format (pbip plugin) -- PBIR JSON format reference