feat(altair): implement line-loss-training

[github-actions]

Implementation: line-loss-training - altair

Implements the altair version of line-loss-training.

File: plots/line-loss-training/implementations/altair.py

Parent Issue: #2860

---

🤖 impl-generate workflow

[claude]

AI Review - Attempt 1/3

Image Description

The plot displays two line curves on a white background with a subtle grid. The title "line-loss-training · altair · pyplots.ai" is centered at the top in large black text. The x-axis is labeled "Epoch" (ranging from 0 to 50) and the y-axis is labeled "Cross-Entropy Loss" (ranging from 0.0 to 2.8). Two curves are shown: a blue line (Training Loss) starting at ~2.5 and exponentially decaying to ~0.15, and a yellow/gold line (Validation Loss) starting at ~2.7, decaying then showing overfitting behavior by rising slightly after epoch 25. A red diamond marker at approximately epoch 30 indicates "Min Val Loss (Epoch 30)" with a red text annotation. The legend in the top-right corner shows "Curve Type" with Training Loss (blue) and Validation Loss (yellow). Points are marked along both lines.

Quality Score: 91/100

Criteria Checklist

Visual Quality (36/40 pts)

• VQ-01: Text Legibility (10/10) - Title is large (28pt), axis labels are 22pt, tick labels 18pt - all clearly readable
• VQ-02: No Overlap (8/8) - No overlapping text elements, all labels are clear
• VQ-03: Element Visibility (7/8) - Lines are thick (strokeWidth=3), points are visible, but points at 50 epochs create slight visual density
• VQ-04: Color Accessibility (5/5) - Blue (#306998) and yellow (#FFD43B) are colorblind-safe and have excellent contrast
• VQ-05: Layout Balance (5/5) - Plot fills canvas well, good margins, legend is appropriately positioned
• VQ-06: Axis Labels (2/2) - Y-axis specifies "Cross-Entropy Loss" as required by spec, X-axis is "Epoch"
• VQ-07: Grid & Legend (1/2) - Grid opacity is subtle (0.3), legend is well-placed but labeled "Curve Type" instead of something more specific

Spec Compliance (25/25 pts)

• SC-01: Plot Type (8/8) - Correct line chart showing training/validation loss curves
• SC-02: Data Mapping (5/5) - X=Epoch, Y=Loss values correctly mapped
• SC-03: Required Features (5/5) - Has distinct colors, legend, both curves, and optional minimum validation loss marker
• SC-04: Data Range (3/3) - All data visible, axes show complete range
• SC-05: Legend Accuracy (2/2) - Legend correctly labels Training Loss and Validation Loss
• SC-06: Title Format (2/2) - Title follows required format "line-loss-training · altair · pyplots.ai"

Data Quality (18/20 pts)

• DQ-01: Feature Coverage (8/8) - Shows exponential decay, overfitting behavior (val loss rising after epoch 25), optimal stopping point
• DQ-02: Realistic Context (5/7) - Neural network training scenario is plausible, but 50 epochs is on the shorter end for typical training
• DQ-03: Appropriate Scale (5/5) - Loss values (0.15 to 2.7) are realistic for cross-entropy loss

Code Quality (9/10 pts)

• CQ-01: KISS Structure (3/3) - Simple structure: imports → data → plot → save
• CQ-02: Reproducibility (3/3) - Uses np.random.seed(42)
• CQ-03: Clean Imports (2/2) - Only necessary imports (altair, numpy, pandas)
• CQ-04: No Deprecated API (1/1) - Uses current Altair API
• CQ-05: Output Correct (0/1) - Saves as "plot.png" which is correct, but also saves HTML

Library Features (3/5 pts)

• LF-01: Uses distinctive library features (3/5) - Uses layered grammar (lines + points + marker + text), proper encoding types, but doesn't use interactivity in a meaningful way for the saved output

Strengths

• Excellent implementation of overfitting visualization with validation loss diverging from training loss after epoch 25
• Clean use of Altair's layered composition with separate chart objects for lines, points, and annotations
• Red diamond marker with annotation clearly indicates the optimal stopping point at minimum validation loss
• Colorblind-safe color scheme with strong contrast between blue training and yellow validation curves
• Proper title format and axis labeling including loss function type as specified

Weaknesses

• Legend title "Curve Type" is generic; could be more descriptive
• Points on every epoch (50 points per curve) add visual noise; could use opacity or fewer markers

Verdict: APPROVED