feat(bokeh): implement acf-pacf

[github-actions]

Implementation: acf-pacf - bokeh

Implements the bokeh version of acf-pacf.

File: plots/acf-pacf/implementations/bokeh.py

Parent Issue: #4663

---

🤖 impl-generate workflow

[claude]

AI Review - Attempt 1/3

Image Description

The plot displays two vertically stacked subplots showing ACF (top) and PACF (bottom) for a simulated AR(2) process. Both panels use blue (#306998) vertical stem lines with circular dot markers extending from a gray zero baseline to each correlation value. Red dashed horizontal lines mark the 95% confidence bounds at approximately ±0.139. The ACF panel shows lag 0 at 1.0, a strong positive correlation at lag 1 (0.42), a negative dip at lag 3 (-0.25), and exponential decay pattern typical of AR processes. The PACF panel starts from lag 1, showing a significant positive spike at lag 1 (0.43) and a sharp negative spike at lag 2 (-0.27), with all subsequent lags falling within the confidence bounds — the classic AR(2) signature. X-axes are labeled "Lag" (shared range 0–35), y-axes labeled "ACF" and "PACF". The title reads "acf-pacf · bokeh · pyplots.ai". The background is clean with subtle dashed y-grid lines and no x-grid. Overall layout is professional but the two panels occupy 4800×2600 pixels (slightly short of the 4800×2700 target).

Score: 81/100

Category	Score	Max
Visual Quality	27	30
Design Excellence	11	20
Spec Compliance	15	15
Data Quality	12	15
Code Quality	10	10
Library Mastery	6	10
Total	81	100

Visual Quality (27/30)

• VQ-01: Text Legibility (8/8) — All font sizes explicitly set: title 36pt, axis labels 28pt, tick labels 22pt. Perfectly readable.
• VQ-02: No Overlap (6/6) — No overlapping text or elements anywhere.
• VQ-03: Element Visibility (5/6) — Stem lines (width 5) and markers (size 18) are clearly visible. At higher lags where values are small, stems become very short but markers remain visible.
• VQ-04: Color Accessibility (4/4) — Blue stems vs red confidence bounds. Not a red-green pair. Good contrast against white background.
• VQ-05: Layout & Canvas (3/4) — Canvas is 4800×2600 (slightly under 4800×2700 target). Two subplots fill the space reasonably but there is some vertical wasted space between and around the panels.
• VQ-06: Axis Labels & Title (1/2) — "Lag", "ACF", "PACF" are descriptive but terse. No units, though correlation and lag count are arguably unitless.

Design Excellence (11/20)

• DE-01: Aesthetic Sophistication (4/8) — Clean color scheme with Python Blue and red for bounds. Looks like a well-configured default — professional but not exceptional. No custom typography or color hierarchy beyond the basics.
• DE-02: Visual Refinement (4/6) — Good: x-grid removed, y-grid subtle and dashed (alpha 0.2), outline removed, toolbar hidden. Some intentional refinement beyond defaults.
• DE-03: Data Storytelling (3/6) — The AR(2) data choice creates the classic textbook pattern: exponential ACF decay + sharp PACF cutoff at lag 2. This tells a meaningful statistical story. However, no visual emphasis (color, annotation, or highlight) draws attention to the significant lags.

Spec Compliance (15/15)

• SC-01: Plot Type (5/5) — Correct stem/bar plots with two vertically stacked subplots.
• SC-02: Required Features (4/4) — All present: stem lines from zero baseline, 95% confidence bounds as dashed lines, lag 0 in ACF (=1.0), PACF starts from lag 1, shared x-axis, 35 lags.
• SC-03: Data Mapping (3/3) — X=lag number, Y=correlation value. Both axes show full data range.
• SC-04: Title & Legend (3/3) — Title "acf-pacf · bokeh · pyplots.ai" in correct format. No legend needed for single-series plot.

Data Quality (12/15)

• DQ-01: Feature Coverage (5/6) — AR(2) process demonstrates both significant and insignificant lags, positive and negative correlations, and the characteristic ACF/PACF patterns. Could show seasonal patterns for additional coverage.
• DQ-02: Realistic Context (3/5) — Simulated AR(2) process is plausible for time series analysis but generic. Not tied to a real-world scenario like airline passengers or stock returns as suggested in the spec.
• DQ-03: Appropriate Scale (4/4) — 200 observations, 35 lags, correlation values in [-1, 1]. All sensible and realistic.

Code Quality (10/10)

• CQ-01: KISS Structure (3/3) — Clean linear flow: imports → data generation → compute ACF/PACF → plot ACF → plot PACF → style → save.
• CQ-02: Reproducibility (2/2) — np.random.seed(42) set.
• CQ-03: Clean Imports (2/2) — All imports are used: numpy, bokeh components, statsmodels acf/pacf.
• CQ-04: Code Elegance (2/2) — Appropriate complexity. Good use of loop for shared styling. No over-engineering.
• CQ-05: Output & API (1/1) — Saves as plot.png via export_png. Also saves HTML. Current API usage.

Library Mastery (6/10)

• LM-01: Idiomatic Usage (4/5) — Good use of ColumnDataSource, figure, segment+scatter for stems, Span for reference lines, column layout, x_range sharing. Idiomatic Bokeh patterns.
• LM-02: Distinctive Features (2/5) — Uses ColumnDataSource and Span (Bokeh-specific), plus HTML export alongside PNG. However, doesn't leverage Bokeh's interactive strengths like HoverTool for showing exact correlation values on hover, or Band for confidence intervals.

Score Caps Applied

• None — no caps triggered (DE-01=4, DE-02=4, both above threshold)

Strengths

• Perfect spec compliance — all required features implemented correctly
• Clean, well-structured code with good reproducibility
• Appropriate AR(2) data choice that creates meaningful ACF/PACF patterns
• Good visual refinement: subtle grid, removed outlines, clean styling

Weaknesses

• No interactive features leveraging Bokeh's strengths (e.g., HoverTool for correlation values)
• Generic data context — could use a real-world scenario instead of abstract AR(2) simulation
• No visual emphasis on significant lags to guide the viewer's eye
• Canvas slightly undersized (2600px vs 2700px target height)

Issues Found

1. LM-02 LOW: Doesn't leverage Bokeh's interactive capabilities
   • Fix: Add HoverTool to display exact correlation values and lag numbers on hover
2. DE-03 MEDIUM: No visual storytelling emphasis
   • Fix: Highlight significant lags (those exceeding confidence bounds) with a different color to create a clear focal point
3. DE-01 MEDIUM: Professional but not exceptional aesthetics
   • Fix: Consider a more refined color palette, add subtle differentiation between ACF and PACF panels
4. DQ-02 MEDIUM: Generic simulated data
   • Fix: Use a real-world context (e.g., monthly airline passengers, daily temperature) to make the example more relatable

AI Feedback for Next Attempt

Add HoverTool to show correlation values on hover — this is Bokeh's key strength. Highlight significant lags (exceeding confidence bounds) with a contrasting color to create visual storytelling. Consider using a real-world data context instead of abstract AR(2). Fix canvas height to 2700px (use height=1350 per subplot). These changes would significantly improve Library Mastery and Design Excellence scores.

Verdict: REJECTED

[github-actions]

🔧 Repair Attempt 1/3

Applied fixes based on AI review feedback.

Status: Repair completed, re-triggering review...

---

🤖 impl-repair

[claude]

AI Review - Attempt 2/3

Image Description

The plot displays two vertically stacked subplots on a light gray (#FAFAFA) background. The top subplot shows ACF values with the title "acf-pacf · bokeh · pyplots.ai" in dark gray. The bottom subplot shows PACF values. Both use vertical stem lines from a zero baseline to correlation values, topped with circular markers (size 18). Significant lags (outside 95% CI) are rendered in darker blue (#306998), while non-significant lags appear in muted blue (#8FAEC4), creating immediate visual differentiation. Red dashed horizontal lines mark the 95% confidence interval bounds (~±0.139). A small "95% CI" label in red sits at the right edge of the ACF plot. The x-axis is labeled "Lag" (0–35, shared between subplots), y-axes are labeled "ACF" and "PACF". The ACF shows lag 0 at 1.0, strong positive correlation at lag 1 (0.42), negative at lag 3 (-0.25), with exponential decay — consistent with the AR(2) generative process. The PACF shows sharp cutoff after lag 2 (positive lag 1 ~0.43, negative lag 2 ~-0.3), correctly identifying the AR(2) order. Grid lines are subtle dashed on y-axis only. No toolbar is visible.

Score: 88/100

Category	Score	Max
Visual Quality	28	30
Design Excellence	13	20
Spec Compliance	15	15
Data Quality	14	15
Code Quality	10	10
Library Mastery	8	10
Total	88	100

Visual Quality (28/30)

• VQ-01: Text Legibility (8/8) — All font sizes explicitly set: title 36pt, axis labels 28pt, tick labels 22pt. All perfectly readable at 4800px width.
• VQ-02: No Overlap (6/6) — No overlapping text elements anywhere. Clean spacing between all labels.
• VQ-03: Element Visibility (5/6) — Stem lines (line_width=5) and markers (size=18) are well-visible. Muted non-significant stems are slightly hard to distinguish at higher lags where values are small.
• VQ-04: Color Accessibility (4/4) — Blue intensity differentiation is colorblind-safe. Red confidence bounds used sparingly with good contrast.
• VQ-05: Layout & Canvas (3/4) — Two subplots at 4800×1350 each fill the 4800×2700 canvas well. Minor wasted space between subplots and at margins.
• VQ-06: Axis Labels & Title (2/2) — "Lag", "ACF", "PACF" are descriptive and appropriate. Correlation coefficients and lag counts are unitless, so no units needed.

Design Excellence (13/20)

• DE-01: Aesthetic Sophistication (5/8) — Custom color palette with significant/non-significant differentiation, custom background color, removed toolbar, refined axis colors. Above well-configured defaults but not publication-level polish.
• DE-02: Visual Refinement (4/6) — X-grid removed, y-grid subtle dashed at 0.15 alpha, outline removed, minor ticks hidden, axis lines lightened to #CCCCCC. Good refinement but could benefit from a confidence band fill.
• DE-03: Data Storytelling (4/6) — Color differentiation between significant and non-significant lags immediately guides the viewer to important lags. The AR(2) structure is clearly visible with the PACF cutoff after lag 2. Viewer can identify model order at a glance.

Spec Compliance (15/15)

• SC-01: Plot Type (5/5) — Correct ACF/PACF stem plots in two vertically stacked subplots.
• SC-02: Required Features (4/4) — All features present: vertical stem lines from zero, 95% CI dashed lines at ±1.96/√N, lag 0 in ACF (=1.0), PACF starts from lag 1, shared x-axis, 35 lags.
• SC-03: Data Mapping (3/3) — X=lag number, Y=correlation values. Axes show full data range correctly.
• SC-04: Title & Legend (3/3) — Title format "acf-pacf · bokeh · pyplots.ai" is correct. No legend needed for single-series stem plot; color meaning is self-evident.

Data Quality (14/15)

• DQ-01: Feature Coverage (6/6) — AR(2) process shows both significant and non-significant lags, positive and negative correlations, exponential ACF decay, and sharp PACF cutoff — all key features of ACF/PACF analysis.
• DQ-02: Realistic Context (4/5) — Monthly retail sales with inventory cycles is plausible and neutral. Context described in code comments but not visible in the plot itself.
• DQ-03: Appropriate Scale (4/4) — 200 observations, 35 lags, correlation values in [-1, 1], AR coefficients (0.6, -0.3) produce realistic patterns.

Code Quality (10/10)

• CQ-01: KISS Structure (3/3) — Clean Imports → Data → ACF Plot → PACF Plot → Style → Save structure. No functions or classes.
• CQ-02: Reproducibility (2/2) — np.random.seed(42) set.
• CQ-03: Clean Imports (2/2) — All imports are used: numpy, bokeh components, statsmodels acf/pacf.
• CQ-04: Code Elegance (2/2) — Clean and well-organized. Shared styling loop for both plots avoids duplication.
• CQ-05: Output & API (1/1) — Saves as plot.png via export_png, also saves HTML. Current API used.

Library Mastery (8/10)

• LM-01: Idiomatic Usage (4/5) — Good use of ColumnDataSource, segment+scatter for stems, Span for reference lines, Label for annotations, column layout for stacking. Could use Band glyph for confidence intervals.
• LM-02: Distinctive Features (4/5) — HoverTool with tooltips (interactive feature distinctive to Bokeh), HTML export alongside PNG, shared x_range between figures. These features couldn't easily be replicated in static libraries.

Score Caps Applied

• None — no score caps triggered.

Strengths

• Excellent spec compliance — all required features implemented correctly with proper ACF/PACF computation
• Strong data storytelling through significant/non-significant color differentiation that immediately highlights important lags
• Good visual refinement with subtle grid, removed toolbar, and custom axis styling
• Effective use of Bokeh-specific features (HoverTool, HTML export, shared ranges)
• Clean, well-organized code with proper reproducibility

Weaknesses

• Design could be elevated with a semi-transparent confidence band fill (BoxAnnotation) instead of just dashed lines
• The "95% CI" label is small and positioned at the far right edge — could be more prominent or positioned near the band
• No annotation calling out the identified AR(2) structure (e.g., highlighting lags 1-2 in PACF)

Issues Found

1. DE-01 MODERATE: Aesthetic sophistication is good but not publication-ready — confidence bounds are plain dashed lines without fill
   • Fix: Add semi-transparent BoxAnnotation between upper and lower confidence bounds to create a confidence band
2. DE-02 MINOR: "95% CI" label is small and edge-positioned
   • Fix: Reposition label or increase visibility; consider adding it to both subplots
3. LM-01 MINOR: Could use Bokeh's Band or BoxAnnotation glyph for confidence interval shading
   • Fix: Use BoxAnnotation(bottom=-conf_bound, top=conf_bound, fill_alpha=0.08, fill_color=RED) for a polished confidence band

AI Feedback for Next Attempt

To reach 90+: (1) Add a semi-transparent BoxAnnotation between the confidence bounds on both subplots to create a shaded confidence band — this is a Bokeh-distinctive feature that elevates both aesthetics and data storytelling. (2) Make the "95% CI" label more prominent and add it to the PACF subplot as well. (3) Consider adding subtle annotations or visual emphasis on the first 2-3 significant PACF lags to highlight the AR(2) identification story. These changes would push DE-01 to 6+ and LM-01 to 5.

Verdict: REJECTED

[github-actions]

🔧 Repair Attempt 2/3

Applied fixes based on AI review feedback.

Status: Repair completed, re-triggering review...

---

🤖 impl-repair

[claude]

AI Review - Attempt 3/3

Image Description

The plot displays two vertically stacked subplots for ACF (top) and PACF (bottom) of a simulated AR(2) retail sales time series. The title "acf-pacf · bokeh · pyplots.ai" appears at the top left. Both plots use vertical stem lines from a zero baseline to correlation values, with circular markers at each tip. Significant lags (outside the 95% confidence interval) are rendered in darker Python Blue (#306998), while non-significant lags appear in a muted blue (#8FAEC4). A semi-transparent pink/red shaded band marks the 95% confidence interval on both subplots, bounded by dashed red lines. Red "95% Confidence Interval" labels appear on each subplot. The ACF shows lag 0 = 1.0, lag 1 ≈ 0.42, with gradual exponential decay — characteristic of an AR process. The PACF shows a sharp cutoff after lag 2, with lags 1 (≈0.43) and 2 (≈-0.28) highlighted by gold-outlined markers and a gold "AR(2) identified" annotation. The background is light gray (#FAFAFA) with subtle dashed y-grid lines. X-axes share the same range (0–35) labeled "Lag"; y-axes are labeled "ACF" and "PACF" respectively.

Score: 92/100

Category	Score	Max
Visual Quality	29	30
Design Excellence	15	20
Spec Compliance	15	15
Data Quality	15	15
Code Quality	10	10
Library Mastery	8	10
Total	92	100

Visual Quality (29/30)

• VQ-01: Text Legibility (8/8) — All font sizes explicitly set: title 36pt, axis labels 28pt, tick labels 22pt, annotation labels 20-22pt. All clearly readable at full resolution.
• VQ-02: No Overlap (6/6) — No overlapping text or elements. Stems well-spaced across 35 lags. Labels and annotations positioned cleanly.
• VQ-03: Element Visibility (6/6) — Stem lines at line_width=5 and markers at size=18 are perfectly visible. Significant/non-significant color differentiation aids scanning.
• VQ-04: Color Accessibility (4/4) — Blue/muted-blue/gold/red palette is colorblind-safe. No red-green distinction issues. Good contrast against light background.
• VQ-05: Layout & Canvas (3/4) — Two subplots fill the 4800×2700 canvas well, but some vertical whitespace wasted in ACF subplot above lag-0's value of 1.0 where no other data reaches.
• VQ-06: Axis Labels & Title (2/2) — "Lag" for x-axis, "ACF"/"PACF" for y-axes are standard and appropriate for correlation plots (no units needed for dimensionless coefficients).

Design Excellence (15/20)

• DE-01: Aesthetic Sophistication (6/8) — Thoughtful custom palette (Python Blue + muted variant + gold accent + red CI). Custom background (#FAFAFA), refined typography colors ([area-basic] plotnine implementation #333, feat(plotly): implement donut-labeled #555), intentional visual hierarchy with significant/non-significant distinction. Clearly above defaults but not quite FiveThirtyEight-level.
• DE-02: Visual Refinement (4/6) — X-grid removed, y-grid subtle (alpha 0.15, dashed), outline removed, minor ticks hidden, axis lines lightened. Good refinement, though the "95% Confidence Interval" label positioned in the data area is slightly inelegant.
• DE-03: Data Storytelling (5/6) — Strong storytelling: gold-highlighted AR(2) lags with annotation immediately draw attention to the key insight. Significant/non-significant color coding guides interpretation. Confidence band visualization contextualizes statistical significance. Viewer immediately understands the AR(2) structure.

Spec Compliance (15/15)

• SC-01: Plot Type (5/5) — Correct ACF/PACF stem plots in two vertically stacked subplots.
• SC-02: Required Features (4/4) — Vertical stem lines ✓, 95% CI dashed lines ✓, lag 0 in ACF = 1.0 ✓, PACF starts from lag 1 ✓, shared x-axis ✓, 35 lags (within 30–40 range) ✓.
• SC-03: Data Mapping (3/3) — X = lag number, Y = correlation values. Correctly mapped.
• SC-04: Title & Legend (3/3) — Title format "acf-pacf · bokeh · pyplots.ai" correct. No legend needed for this plot type (single series per subplot).

Data Quality (15/15)

• DQ-01: Feature Coverage (6/6) — AR(2) process generates clear exponential ACF decay and sharp PACF cutoff at lag 2. Both positive and negative correlations visible. Mix of significant and non-significant lags present.
• DQ-02: Realistic Context (5/5) — "Monthly retail sales with inventory cycles" is a real-world plausible, neutral scenario. AR coefficients (0.6 momentum, -0.3 correction) model realistic dynamics.
• DQ-03: Appropriate Scale (4/4) — 200 observations is appropriate for reliable correlation estimates. Correlation values in proper [-1, 1] range. 35 lags suitable for this sample size.

Code Quality (10/10)

• CQ-01: KISS Structure (3/3) — Clean imports → data generation → ACF/PACF computation → plot creation → save flow. No functions or classes.
• CQ-02: Reproducibility (2/2) — np.random.seed(42) ensures deterministic output.
• CQ-03: Clean Imports (2/2) — All imports used: numpy, bokeh components, statsmodels acf/pacf.
• CQ-04: Code Elegance (2/2) — Well-organized with clear sections. Styling loop for shared properties avoids duplication. No fake UI or over-engineering.
• CQ-05: Output & API (1/1) — Saves as plot.png via export_png. Also generates HTML for interactive viewing.

Library Mastery (8/10)

• LM-01: Idiomatic Usage (4/5) — Good use of ColumnDataSource for all data, figure for plot creation, segment+scatter for stems, Span/BoxAnnotation for reference lines/bands, column layout for subplot arrangement. Idiomatic patterns throughout.
• LM-02: Distinctive Features (4/5) — HoverTool with vline mode provides interactive tooltips (lag/value/significance) — distinctively Bokeh. BoxAnnotation for confidence band shading, HTML export for interactive version. These features couldn't easily be replicated in static libraries.

Score Caps Applied

• None — no cap conditions triggered.

Strengths

• Excellent data storytelling: AR(2) gold highlights and annotation immediately convey the statistical insight
• Strong color differentiation between significant and non-significant lags aids interpretation
• Bokeh-specific features well leveraged: HoverTool, BoxAnnotation, HTML export
• Clean, well-structured code with proper reproducibility (seed set)
• Full spec compliance with all required features present

Weaknesses

• Minor vertical whitespace waste in ACF subplot above lag-0 peak
• "95% Confidence Interval" label positioned in data area rather than margin
• Could benefit from slightly more refined label placement

Issues Found

None significant — implementation is well-polished after 3 attempts.

AI Feedback for Next Attempt

N/A — implementation meets quality threshold. Strong ACF/PACF visualization with excellent data storytelling through AR(2) highlights, good Bokeh feature usage, and full spec compliance.

Verdict: APPROVED