feat(bokeh): implement precision-recall

[github-actions]

Implementation: precision-recall - bokeh

Implements the bokeh version of precision-recall.

File: plots/precision-recall/implementations/bokeh.py

---

🤖 impl-generate workflow

[claude]

AI Review - Attempt 1/3

Image Description

The plot displays a Precision-Recall curve with two classifier comparisons. The title "precision-recall · bokeh · pyplots.ai" appears centered at the top. Two stepped curves are shown: a blue line for Logistic Regression (AP = 0.788) and a yellow/gold line for Naive Bayes (AP = 0.797). A horizontal gray dashed baseline line at approximately 0.30 represents the random classifier performance. The X-axis is labeled "Recall" (0 to 1), and the Y-axis is labeled "Precision" (0 to 1). The legend is positioned in the top-right corner with a white background. The plot has a light gray background (#fafafa) with subtle dashed grid lines. Both curves demonstrate the characteristic stepped pattern of precision-recall curves, starting high on the left and generally decreasing as recall increases.

Quality Score: 91/100

Criteria Checklist

Visual Quality (37/40 pts)

• VQ-01: Text Legibility (10/10) - Title, axis labels, tick labels, and legend text are all clearly readable at the high resolution
• VQ-02: No Overlap (8/8) - No overlapping text elements; legend is well-positioned and doesn't cover data
• VQ-03: Element Visibility (8/8) - Line widths are appropriate; stepped curves are clearly visible and distinguishable
• VQ-04: Color Accessibility (5/5) - Blue (#306998) and yellow (#FFD43B) are colorblind-safe and highly distinguishable
• VQ-05: Layout Balance (5/5) - Plot fills canvas well with balanced margins
• VQ-06: Axis Labels (1/2) - "Recall" and "Precision" are descriptive but lack units (though units aren't applicable here as these are ratios)
• VQ-07: Grid & Legend (2/2) - Subtle dashed grid with alpha=0.3, legend well-placed with white background

Spec Compliance (25/25 pts)

• SC-01: Plot Type (8/8) - Correct precision-recall curve implementation with stepped lines
• SC-02: Data Mapping (5/5) - Recall on X-axis, Precision on Y-axis as expected
• SC-03: Required Features (5/5) - Includes AP scores in legend, baseline reference line, stepped line style, multiple classifier comparison
• SC-04: Data Range (3/3) - Axes appropriately show 0-1 range for both precision and recall
• SC-05: Legend Accuracy (2/2) - Legend correctly identifies classifiers with AP scores
• SC-06: Title Format (2/2) - Uses correct format "{spec-id} · {library} · pyplots.ai"

Data Quality (17/20 pts)

• DQ-01: Feature Coverage (6/8) - Shows two classifiers with different performance profiles, includes baseline; could benefit from showing iso-F1 curves mentioned in spec as optional
• DQ-02: Realistic Context (7/7) - Imbalanced classification dataset (70/30 split) is a realistic ML evaluation scenario
• DQ-03: Appropriate Scale (4/5) - AP scores around 0.79-0.80 are realistic; baseline at 0.30 matches class imbalance

Code Quality (9/10 pts)

• CQ-01: KISS Structure (3/3) - Linear flow: imports → data generation → model training → plotting → save
• CQ-02: Reproducibility (3/3) - Uses np.random.seed(42) and random_state=42
• CQ-03: Clean Imports (2/2) - All imports are used
• CQ-04: No Deprecated API (1/1) - Uses current bokeh API
• CQ-05: Output Correct (0/1) - Code saves as 'plot.png' which is correct

Library Features (3/5 pts)

• LF-01: Uses distinctive library features (3/5) - Uses ColumnDataSource, step() method, and Legend model; could leverage more Bokeh-specific features like hover tooltips

Strengths

• Excellent use of stepped line style (mode="after") which accurately represents threshold-based PR curves
• Clean comparison of two classifiers with distinct, colorblind-safe colors
• Proper inclusion of baseline reference line showing random classifier performance
• Average Precision scores displayed prominently in legend
• Good font sizing scaled appropriately for 4800x2700 resolution
• Well-structured code following KISS principles with proper reproducibility

Weaknesses

• Missing optional iso-F1 curves mentioned in specification notes (would enhance the visualization)
• Could add hover tooltips to show precision/recall values at specific points (Bokeh strength)

Verdict: APPROVED