make_bundle.py

"""Assemble the reproduction bundle and register it as a Trackio artifact."""
from __future__ import annotations

import argparse
import json
import os
import shutil

EXCLUDE_DIRS = {"__pycache__", ".venv", ".cache", ".git", "node_modules"}
EXCLUDE_FILES = {".env", "ICML_token.txt"}


def copy_tree(src, dst):
    os.makedirs(dst, exist_ok=True)
    n = 0
    for root, dirs, files in os.walk(src):
        dirs[:] = [d for d in dirs if d not in EXCLUDE_DIRS]
        rel = os.path.relpath(root, src)
        target = os.path.join(dst, rel) if rel != "." else dst
        os.makedirs(target, exist_ok=True)
        for f in files:
            if f in EXCLUDE_FILES or f.endswith((".pyc", ".pt")):
                continue
            shutil.copy2(os.path.join(root, f), os.path.join(target, f))
            n += 1
    return n


def prune_dashboard_page(root, project):
    import shutil as sh
    pages = os.path.join(root, ".trackio", "logbook", "pages")
    page_dir = os.path.join(pages, project)
    if os.path.isdir(page_dir):
        sh.rmtree(page_dir)
        print(f"removed stray sidebar page '{project}'")
    # trackio.init() also drops an empty dashboard cell on whichever page was
    # last active; the bundle run logs no metrics, so it renders as noise.
    for page in ("executive-summary", "conclusion"):
        page_md = os.path.join(pages, page, "page.md")
        if os.path.isfile(page_md):
            import re as _re
            with open(page_md, encoding="utf-8") as f:
                text = f.read()
            chunks = _re.split(r"(?=\n---\n<!-- trackio-cell)", text)
            kept_chunks = [c for c in chunks if '"type": "dashboard"' not in c]
            if len(kept_chunks) != len(chunks):
                with open(page_md, "w", encoding="utf-8") as f:
                    f.write("".join(kept_chunks).rstrip() + "\n")
                print(f"removed stray dashboard cell from '{page}'")

    index = os.path.join(pages, "index.md")
    if os.path.isfile(index):
        with open(index, encoding="utf-8") as f:
            lines = f.readlines()
        kept = [l for l in lines if f"(#/{project})" not in l]
        if len(kept) != len(lines):
            with open(index, "w", encoding="utf-8") as f:
                f.writelines(kept)
            print("removed its row from the index Pages table")


if __name__ == "__main__":
    ap = argparse.ArgumentParser()
    ap.add_argument("--root", default=os.path.expanduser("~/repro_spectral_flow_matching"))
    ap.add_argument("--bundle", default=None)
    ap.add_argument("--project", default="repro-spectral-flow-matching-stabilizing-stochastic-gflownets-via-frequency-domain-regularizatio")
    args = ap.parse_args()
    bundle = args.bundle or os.path.join(args.root, "repro_bundle")

    if os.path.exists(bundle):
        shutil.rmtree(bundle)
    total = 0
    for sub in ["src", "outputs", "figures", "poster"]:
        s = os.path.join(args.root, sub)
        if os.path.isdir(s):
            total += copy_tree(s, os.path.join(bundle, sub))
    for f in ["paper.md", "README_code.md"]:
        p = os.path.join(args.root, f)
        if os.path.exists(p):
            shutil.copy2(p, os.path.join(bundle, f))
            total += 1

    manifest = {
        "paper": "Spectral Flow Matching: Stabilizing Stochastic GFlowNets via Frequency-Domain Regularization",
        "openreview_id": "QQRCqb2PDj",
        "contents": {
            "src/": "reimplementation: envs, ST-GFN + 10 baselines, analysis, diagnostics",
            "outputs/": "raw per-run JSON (metrics + training curves), summary tables, diagnostics",
            "figures/": "Plotly HTML figures with their raw CSV data",
            "poster/": "posterly poster source, rendered PNG/PDF, poster_embed.html",
            "paper.md": "OCR transcript of the paper used for the reimplementation",
        },
        "n_files": total,
    }
    with open(os.path.join(bundle, "MANIFEST.json"), "w") as f:
        json.dump(manifest, f, indent=2)
    print(f"bundle at {bundle} with {total} files")

    try:
        import trackio
        trackio.init(project=args.project, name="repro-bundle", resume="allow")
        trackio.log_artifact(bundle, name="repro-bundle", type="dataset")
        trackio.finish()
        print("registered trackio artifact 'repro-bundle'")
    except Exception as exc:
        print(f"trackio artifact registration skipped/failed: {type(exc).__name__}: {exc}")

    # trackio.init() adds a dashboard page for the project to the logbook sidebar.
    # The ICML profile requires exactly Executive summary / Claim N / Conclusion,
    # so drop that page and its Pages-table row again.
    prune_dashboard_page(args.root, args.project)