sync scripts
Browse files- scripts/best_tables.py +110 -0
- scripts/build_report.py +165 -0
scripts/best_tables.py
ADDED
|
@@ -0,0 +1,110 @@
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
| 1 |
+
"""Summarise the performance-first (``best_*``) runs.
|
| 2 |
+
|
| 3 |
+
Produces ``results/tables/tab_best.csv`` with, per architecture:
|
| 4 |
+
|
| 5 |
+
* per-fold stress R2 averaged over seeds (mean +- sd over folds/seeds)
|
| 6 |
+
* pooled out-of-fold stress / shape / primary R2 for seed 0 and for the
|
| 7 |
+
seed-averaged predictions (all seeds share one fold split by construction)
|
| 8 |
+
|
| 9 |
+
and prints the markdown table used in the report.
|
| 10 |
+
"""
|
| 11 |
+
from __future__ import annotations
|
| 12 |
+
|
| 13 |
+
import json
|
| 14 |
+
import sys
|
| 15 |
+
from pathlib import Path
|
| 16 |
+
|
| 17 |
+
import numpy as np
|
| 18 |
+
import pandas as pd
|
| 19 |
+
|
| 20 |
+
ROOT = Path(__file__).resolve().parents[1]
|
| 21 |
+
sys.path.insert(0, str(ROOT / "code"))
|
| 22 |
+
|
| 23 |
+
from pimamba.data import TARGETS, TARGET_KEYS # noqa: E402
|
| 24 |
+
from pimamba.metrics import PRIMARY, _r2, regression_metrics # noqa: E402
|
| 25 |
+
|
| 26 |
+
RUNS = ROOT / "results" / "runs"
|
| 27 |
+
TABLES = ROOT / "results" / "tables"
|
| 28 |
+
|
| 29 |
+
|
| 30 |
+
def runs_of(prefix: str):
|
| 31 |
+
out = {}
|
| 32 |
+
for d in RUNS.glob(f"{prefix}_*_f*_s*"):
|
| 33 |
+
try:
|
| 34 |
+
m, f, s = d.name[len(prefix) + 1:].rsplit("_", 2)
|
| 35 |
+
f = int(f[1:]); s = int(s[1:])
|
| 36 |
+
except ValueError:
|
| 37 |
+
continue
|
| 38 |
+
if (d / "metrics.json").exists():
|
| 39 |
+
out.setdefault(m, {}).setdefault(s, {})[f] = d
|
| 40 |
+
return out
|
| 41 |
+
|
| 42 |
+
|
| 43 |
+
def pooled(tags: dict, seeds: list[int]):
|
| 44 |
+
"""Pool the predictions of one model over the given seeds."""
|
| 45 |
+
P, Y, M = [], [], []
|
| 46 |
+
for f in sorted(tags[seeds[0]]):
|
| 47 |
+
arrs = [np.load(tags[s][f] / "predictions.npz", allow_pickle=True)
|
| 48 |
+
for s in seeds if f in tags.get(s, {})]
|
| 49 |
+
if not arrs:
|
| 50 |
+
continue
|
| 51 |
+
P.append(np.mean([a["pred"] for a in arrs], 0))
|
| 52 |
+
Y.append(arrs[0]["true"]); M.append(arrs[0]["mask"])
|
| 53 |
+
if not P:
|
| 54 |
+
return None
|
| 55 |
+
return (np.concatenate(P), np.concatenate(Y), np.concatenate(M))
|
| 56 |
+
|
| 57 |
+
|
| 58 |
+
def score(P, Y, M):
|
| 59 |
+
pt = regression_metrics(Y.reshape(-1, len(TARGETS)), P.reshape(-1, len(TARGETS)),
|
| 60 |
+
M.reshape(-1, len(TARGETS)))
|
| 61 |
+
j = TARGET_KEYS.index("sigma")
|
| 62 |
+
mm = M[:, :, j] > 0
|
| 63 |
+
sig = _r2(Y[:, :, j][mm], P[:, :, j][mm])
|
| 64 |
+
ys, ps = [], []
|
| 65 |
+
for i in range(P.shape[0]):
|
| 66 |
+
k = M[i, :, j] > 0
|
| 67 |
+
if k.sum() < 4:
|
| 68 |
+
continue
|
| 69 |
+
ys.append(Y[i, k, j] - Y[i, k, j].mean())
|
| 70 |
+
ps.append(P[i, k, j] - P[i, k, j].mean())
|
| 71 |
+
shape = _r2(np.concatenate(ys), np.concatenate(ps))
|
| 72 |
+
prim = [pt[k]["r2"] for k in PRIMARY if np.isfinite(pt[k]["r2"])]
|
| 73 |
+
return dict(sigma=sig, shape=shape, primary=float(np.mean(prim)))
|
| 74 |
+
|
| 75 |
+
|
| 76 |
+
def main(prefix="best"):
|
| 77 |
+
found = runs_of(prefix)
|
| 78 |
+
rows = []
|
| 79 |
+
for m, per_seed in sorted(found.items()):
|
| 80 |
+
seeds = sorted(per_seed)
|
| 81 |
+
folds = sorted(per_seed[seeds[0]])
|
| 82 |
+
nparams = json.loads((per_seed[seeds[0]][folds[0]] / "metrics.json").read_text())["n_params"]
|
| 83 |
+
fold_r2 = []
|
| 84 |
+
for s in seeds:
|
| 85 |
+
for f, d in per_seed[s].items():
|
| 86 |
+
r = json.loads((d / "metrics.json").read_text())
|
| 87 |
+
fold_r2.append(r["per_target"]["sigma"]["r2"])
|
| 88 |
+
row = {"model": m, "seeds": len(seeds), "folds": len(folds),
|
| 89 |
+
"params": nparams,
|
| 90 |
+
"fold_r2_mean": float(np.mean(fold_r2)),
|
| 91 |
+
"fold_r2_sd": float(np.std(fold_r2)),
|
| 92 |
+
"n_runs": sum(len(v) for v in per_seed.values())}
|
| 93 |
+
full = all(len(per_seed[s]) == 5 for s in seeds)
|
| 94 |
+
if full:
|
| 95 |
+
for tag, ss in (("seed0", [seeds[0]]), ("seedavg", seeds)):
|
| 96 |
+
g = pooled(per_seed, ss)
|
| 97 |
+
sc = score(*g)
|
| 98 |
+
row[f"{tag}_sigma"] = round(sc["sigma"], 3)
|
| 99 |
+
row[f"{tag}_shape"] = round(sc["shape"], 3)
|
| 100 |
+
row[f"{tag}_primary"] = round(sc["primary"], 3)
|
| 101 |
+
rows.append(row)
|
| 102 |
+
df = pd.DataFrame(rows)
|
| 103 |
+
TABLES.mkdir(parents=True, exist_ok=True)
|
| 104 |
+
df.to_csv(TABLES / f"tab_{prefix}.csv", index=False)
|
| 105 |
+
print(df.to_string(index=False))
|
| 106 |
+
return df
|
| 107 |
+
|
| 108 |
+
|
| 109 |
+
if __name__ == "__main__":
|
| 110 |
+
main(sys.argv[1] if len(sys.argv) > 1 else "best")
|
scripts/build_report.py
ADDED
|
@@ -0,0 +1,165 @@
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
| 1 |
+
"""Build the HTML and PDF versions of the report from ``report/paper.md``.
|
| 2 |
+
|
| 3 |
+
Run on a machine that has WeasyPrint and a CJK font installed::
|
| 4 |
+
|
| 5 |
+
python scripts/build_report.py
|
| 6 |
+
|
| 7 |
+
The HTML is self-contained apart from the figure files (``report/figures/``),
|
| 8 |
+
so it can be opened directly in a browser or printed to PDF by hand.
|
| 9 |
+
"""
|
| 10 |
+
from __future__ import annotations
|
| 11 |
+
|
| 12 |
+
import shutil
|
| 13 |
+
import sys
|
| 14 |
+
from pathlib import Path
|
| 15 |
+
|
| 16 |
+
import markdown
|
| 17 |
+
from weasyprint import CSS, HTML
|
| 18 |
+
|
| 19 |
+
ROOT = Path(__file__).resolve().parents[1]
|
| 20 |
+
REPORT = ROOT / "report"
|
| 21 |
+
SRC = REPORT / "paper.md"
|
| 22 |
+
OUT_HTML = REPORT / "paper.html"
|
| 23 |
+
OUT_PDF = REPORT / "paper.pdf"
|
| 24 |
+
FIG_SRC = ROOT / "results" / "figures"
|
| 25 |
+
FIG_DST = REPORT / "figures"
|
| 26 |
+
|
| 27 |
+
# The text uses LaTeX only inside the first sections; converting those few
|
| 28 |
+
# expressions to Unicode keeps the PDF readable without a JS math renderer.
|
| 29 |
+
MATH = {
|
| 30 |
+
r"$\mathrm{m^{-2}}$": "m⁻²",
|
| 31 |
+
r"$10^{16}\,\mathrm{m^{-2}}$": "10¹⁶ m⁻²",
|
| 32 |
+
r"$5\times10^{-4}$": "5×10⁻⁴",
|
| 33 |
+
r"$33\times401=13233$": "33 × 401 = 13233",
|
| 34 |
+
r"$T=401$": "T = 401",
|
| 35 |
+
r"$R^2=1$": "R² = 1",
|
| 36 |
+
r"$R^2=0$": "R² = 0",
|
| 37 |
+
r"$R^2<0$": "R² < 0",
|
| 38 |
+
r"$\sigma$": "σ",
|
| 39 |
+
r"$c$": "c",
|
| 40 |
+
r"$\delta$": "δ",
|
| 41 |
+
r"$\chi$": "χ",
|
| 42 |
+
r"$\Delta H_{\mathrm{mix}}$": "ΔH_mix",
|
| 43 |
+
r"$\Delta S_{\mathrm{mix}}$": "ΔS_mix",
|
| 44 |
+
r"$\rho_{\mathrm{total}}$": "ρ_total",
|
| 45 |
+
r"$\rho_{\mathrm{twin}}$": "ρ_twin",
|
| 46 |
+
r"$\varepsilon_{1:T}$": "ε(1..T)",
|
| 47 |
+
r"$\xi_{\mathrm{BCC}}, \xi_{\mathrm{FCC}}, \xi_{\mathrm{HCP}}, \xi_{\mathrm{other}}$":
|
| 48 |
+
"ξ_BCC, ξ_FCC, ξ_HCP, ξ_other",
|
| 49 |
+
r"$\rho_{1/2},\rho_{a/2},\rho_{a/3},\rho_{\mathrm{edge}},\rho_{\mathrm{screw}},\rho_{\mathrm{perfect}},\rho_{\mathrm{other}}$":
|
| 50 |
+
"ρ_1/2, ρ_a/2, ρ_a/3, ρ_edge, ρ_screw, ρ_perfect, ρ_other",
|
| 51 |
+
}
|
| 52 |
+
|
| 53 |
+
CSS_TEXT = r"""
|
| 54 |
+
@page {
|
| 55 |
+
size: A4;
|
| 56 |
+
margin: 20mm 18mm 18mm 18mm;
|
| 57 |
+
@bottom-center {
|
| 58 |
+
content: counter(page);
|
| 59 |
+
font-family: "Noto Serif CJK SC", serif;
|
| 60 |
+
font-size: 9pt;
|
| 61 |
+
color: #555;
|
| 62 |
+
}
|
| 63 |
+
@top-center {
|
| 64 |
+
content: string(doctitle);
|
| 65 |
+
font-family: "Noto Serif CJK SC", serif;
|
| 66 |
+
font-size: 8.5pt;
|
| 67 |
+
color: #777;
|
| 68 |
+
}
|
| 69 |
+
}
|
| 70 |
+
@page :first { @top-center { content: none; } }
|
| 71 |
+
|
| 72 |
+
html { font-size: 10.5pt; }
|
| 73 |
+
body {
|
| 74 |
+
font-family: "Noto Serif CJK SC", "Songti SC", serif;
|
| 75 |
+
line-height: 1.78;
|
| 76 |
+
color: #111;
|
| 77 |
+
text-align: left;
|
| 78 |
+
max-width: 820px;
|
| 79 |
+
margin: 0 auto;
|
| 80 |
+
padding: 0 8px;
|
| 81 |
+
}
|
| 82 |
+
h1 {
|
| 83 |
+
font-size: 19pt; line-height: 1.45; text-align: center;
|
| 84 |
+
margin: 0 0 6px 0; font-weight: 700; string-set: doctitle content();
|
| 85 |
+
}
|
| 86 |
+
h2 {
|
| 87 |
+
font-size: 13.5pt; margin: 22px 0 8px 0; font-weight: 700;
|
| 88 |
+
border-bottom: 1.2px solid #333; padding-bottom: 3px;
|
| 89 |
+
page-break-after: avoid;
|
| 90 |
+
}
|
| 91 |
+
h3 { font-size: 11.5pt; margin: 16px 0 6px 0; font-weight: 700; page-break-after: avoid; }
|
| 92 |
+
p { margin: 0 0 8px 0; text-indent: 2em; }
|
| 93 |
+
p.eq { text-align: center; text-indent: 0; margin: 9px 0 11px 0; }
|
| 94 |
+
p.eq sub, td sub { font-size: 68%; }
|
| 95 |
+
p:has(> img) { text-indent: 0; text-align: center; margin: 12px 0 4px 0; }
|
| 96 |
+
img { max-width: 96%; height: auto; }
|
| 97 |
+
table {
|
| 98 |
+
border-collapse: collapse; width: 100%; margin: 10px 0 6px 0;
|
| 99 |
+
font-size: 8.6pt; line-height: 1.45; page-break-inside: avoid;
|
| 100 |
+
break-before: avoid; /* keep the caption above the table on the same page */
|
| 101 |
+
}
|
| 102 |
+
th, td { border: 0.5pt solid #999; padding: 3px 5px; text-align: center; vertical-align: middle; }
|
| 103 |
+
th { background: #f0f0f0; font-weight: 700; }
|
| 104 |
+
td:first-child, th:first-child { text-align: left; }
|
| 105 |
+
tr:nth-child(even) td { background: #fafafa; }
|
| 106 |
+
code { font-family: "DejaVu Sans Mono", monospace; font-size: 8.6pt; background: #f4f4f4; padding: 0 2px; }
|
| 107 |
+
pre {
|
| 108 |
+
background: #f6f6f6; border: 0.5pt solid #ddd; padding: 6px 8px;
|
| 109 |
+
font-size: 8.2pt; line-height: 1.4; white-space: pre-wrap; page-break-inside: avoid;
|
| 110 |
+
}
|
| 111 |
+
pre code { background: none; }
|
| 112 |
+
blockquote { margin: 8px 0; padding-left: 10px; border-left: 3px solid #ccc; color: #444; }
|
| 113 |
+
hr { border: none; border-top: 0.6pt solid #ccc; margin: 16px 0; }
|
| 114 |
+
strong { font-weight: 700; }
|
| 115 |
+
#TOC { background: #f7f7f7; border: 0.5pt solid #ddd; padding: 8px 14px; margin: 14px 0 18px 0; }
|
| 116 |
+
#TOC ul { margin: 2px 0; padding-left: 18px; }
|
| 117 |
+
#TOC li { margin: 1px 0; font-size: 9.2pt; }
|
| 118 |
+
#TOC a { color: #111; text-decoration: none; }
|
| 119 |
+
a { color: #1a4f8b; text-decoration: none; }
|
| 120 |
+
"""
|
| 121 |
+
|
| 122 |
+
TEMPLATE = """<!DOCTYPE html>
|
| 123 |
+
<html lang="zh-CN">
|
| 124 |
+
<head>
|
| 125 |
+
<meta charset="utf-8">
|
| 126 |
+
<title>{title}</title>
|
| 127 |
+
<style>{css}</style>
|
| 128 |
+
</head>
|
| 129 |
+
<body>
|
| 130 |
+
{body}
|
| 131 |
+
</body>
|
| 132 |
+
</html>
|
| 133 |
+
"""
|
| 134 |
+
|
| 135 |
+
|
| 136 |
+
def main() -> int:
|
| 137 |
+
if not SRC.exists():
|
| 138 |
+
print(f"missing {SRC}")
|
| 139 |
+
return 1
|
| 140 |
+
text = SRC.read_text(encoding="utf-8")
|
| 141 |
+
for k, v in MATH.items():
|
| 142 |
+
text = text.replace(k, v)
|
| 143 |
+
left = text.count("$")
|
| 144 |
+
if left:
|
| 145 |
+
print(f"warning: {left} '$' characters remain (unconverted LaTeX?)")
|
| 146 |
+
body = markdown.markdown(
|
| 147 |
+
text, extensions=["tables", "fenced_code", "toc", "sane_lists"],
|
| 148 |
+
extension_configs={"toc": {"title": "目录", "toc_depth": "2-2"}})
|
| 149 |
+
doc = TEMPLATE.format(title="多���金 MD 轨迹的应力–应变预测", css=CSS_TEXT, body=body)
|
| 150 |
+
OUT_HTML.write_text(doc, encoding="utf-8")
|
| 151 |
+
|
| 152 |
+
if FIG_SRC.exists():
|
| 153 |
+
FIG_DST.mkdir(exist_ok=True)
|
| 154 |
+
for f in FIG_SRC.glob("*.png"):
|
| 155 |
+
shutil.copy2(f, FIG_DST / f.name)
|
| 156 |
+
|
| 157 |
+
HTML(string=doc, base_url=str(REPORT)).write_pdf(
|
| 158 |
+
str(OUT_PDF), stylesheets=[CSS(string=CSS_TEXT)])
|
| 159 |
+
print(f"html {OUT_HTML.stat().st_size/1024:.0f} KB, "
|
| 160 |
+
f"pdf {OUT_PDF.stat().st_size/1024:.0f} KB")
|
| 161 |
+
return 0
|
| 162 |
+
|
| 163 |
+
|
| 164 |
+
if __name__ == "__main__":
|
| 165 |
+
sys.exit(main())
|