paper-publication

name	paper-publication
description	ICML 2025 paper visuals, figures, and LaTeX tables. Use when generating figures, formatting tables, or preparing appendix materials. Triggers include figure, M1, M2, M3, publication, LaTeX, consolidated, ICML, paper.

Paper Publication Assets

Quick Reference

Document	Purpose
references/FIGURE_SPECS.md	Main (M1-M3) and appendix (A1-A6) figure specs
references/LATEX_TABLES.md	Table templates

When to Use This Skill

• You need to regenerate consolidated figures
• You need to update LaTeX tables from new results
• You need to generate Table 1 (dataset overview) or Table 2 (accuracy summary)

Figure & Table Naming

Output	Content	Location
Table 1	Dataset overview (n, hierarchy cols, features)	paper/paper_table1_dataset_overview.*
Table 2	Accuracy summary (RMSE, R², improvements)	paper/paper_table2_accuracy_summary.*
M1	RMSE Improvement (bar chart)	paper/paper_m1_rmse_improvement.*
M2	Feature Contributions (stacked bars)	paper/paper_m2_feature_contributions.*
M3	Strategy Ranking (heatmap, no grid lines)	paper/paper_m3_strategy_ranking.*
A1-A6	Per-dataset appendix figures	appendix/appendix_{dataset}_A{1-6}.*

Procedure

1. Run search experiments to generate latest results (python scripts/run.py --all --search).
2. Run generate_paper_appendix_figures.py to create figures.
3. Use flags: --paper-only, --appendix-only, --tables-only as needed.
4. Export PDF/PNG for the paper (both formats generated automatically).
5. LaTeX tables (.tex) are also generated alongside figures.

Script Usage

# Generate all figures (paper + appendix)
python scripts/generate_paper_appendix_figures.py

# Generate only main paper figures (M1-M3, Tables 1-2)
python scripts/generate_paper_appendix_figures.py --paper-only

# Generate only appendix figures (A1-A6 per dataset)
python scripts/generate_paper_appendix_figures.py --appendix-only

# Generate only Tables 1 and 2
python scripts/generate_paper_appendix_figures.py --tables-only

Output Directories

• docs/figures/paper/ - Main paper figures (M1-M3, Tables 1-2)
• docs/figures/appendix/ - Per-dataset appendix figures

SCE Feature Classification

Features are classified as Base (blue) or SCE Context (red) based on suffixes:

• SCE suffixes: _mean, _median, _std, _fold_lower, _fold_upper, _q05, _q10, etc.
• The is_sce_feature() function handles both simple and compound suffixes.