feat(seaborn): implement confusion-matrix

[github-actions]

Implementation: confusion-matrix - seaborn

Implements the seaborn version of confusion-matrix.

File: plots/confusion-matrix/implementations/seaborn.py

---

🤖 impl-generate workflow

[claude]

AI Review - Attempt 1/3

Image Description

The plot displays a 4x4 confusion matrix heatmap for sentiment analysis classification with four classes: Negative, Neutral, Positive, and Mixed. The matrix uses a sequential blue colormap ("Blues") where darker blues indicate higher counts. Cell annotations show the raw count values in bold white/black text. The diagonal cells (correct predictions) are prominently darker: Negative=156, Neutral=142, Positive=168, Mixed=125. Off-diagonal cells show misclassification patterns with lighter shading. The y-axis is labeled "True Label", x-axis is "Predicted Label". A colorbar on the right shows the count scale from ~0 to 168. The title reads "Sentiment Analysis Model · confusion-matrix · seaborn · pyplots.ai". White gridlines separate cells clearly. X-axis labels are rotated 45° for readability.

Quality Score: 92/100

Criteria Checklist

Visual Quality (37/40 pts)

• VQ-01: Text Legibility (10/10) - Title at 24pt, axis labels at 22pt, tick labels at 18pt, annotations at 20pt bold - all perfectly readable
• VQ-02: No Overlap (8/8) - No overlapping text anywhere, rotated x-labels prevent collision
• VQ-03: Element Visibility (8/8) - Cells are square, well-sized, annotations clearly visible with contrasting colors
• VQ-04: Color Accessibility (5/5) - Blues sequential colormap is colorblind-safe
• VQ-05: Layout Balance (4/5) - Square format appropriate for matrix, good proportions but colorbar could be slightly better integrated
• VQ-06: Axis Labels (2/2) - "True Label" and "Predicted Label" as specified
• VQ-07: Grid & Legend (0/2) - No grid needed for heatmap, but colorbar label "Count" is duplicated (in cbar_kws and manually set)

Spec Compliance (25/25 pts)

• SC-01: Plot Type (8/8) - Correct confusion matrix heatmap
• SC-02: Data Mapping (5/5) - True labels on y-axis, Predicted on x-axis as specified
• SC-03: Required Features (5/5) - Annotations present, Blues colormap, colorbar included, clear axis labels
• SC-04: Data Range (3/3) - All 4 classes visible, full range shown
• SC-05: Legend Accuracy (2/2) - Colorbar correctly shows count scale
• SC-06: Title Format (2/2) - Follows "{description} · {spec-id} · {library} · pyplots.ai" format

Data Quality (20/20 pts)

• DQ-01: Feature Coverage (8/8) - Shows strong diagonal (good model), asymmetric confusion patterns, varied misclassification rates
• DQ-02: Realistic Context (7/7) - Sentiment analysis is a real-world NLP task, 4 classes are realistic
• DQ-03: Appropriate Scale (5/5) - Counts are realistic for a test set (~700 samples total)

Code Quality (7/10 pts)

• CQ-01: KISS Structure (3/3) - Simple linear structure: imports → data → plot → save
• CQ-02: Reproducibility (0/3) - No random seed, though data is deterministic (hardcoded matrix)
• CQ-03: Clean Imports (2/2) - Only matplotlib, numpy, seaborn used
• CQ-04: No Deprecated API (1/1) - Uses current seaborn API
• CQ-05: Output Correct (1/1) - Saves as plot.png

Library Features (3/5 pts)

• LF-01: Uses seaborn's heatmap with annotations, square cells, colorbar customization - good use but could leverage mask parameter or diverging normalization for additional features

Strengths

• Excellent visual clarity with well-sized annotations and contrasting text colors on cells
• Properly follows specification with True/Predicted axis labels and Blues colormap
• Realistic sentiment analysis scenario with believable confusion patterns showing systematic errors
• Clean, readable code following KISS principles
• Square format appropriate for symmetric confusion matrix

Weaknesses

• Minor colorbar label duplication (label set both in cbar_kws and manually)
• Data is deterministic but no explicit np.random.seed() for documentation purposes

Verdict: APPROVED