File size: 6,396 Bytes
45383d0 | 1 2 3 4 5 6 7 8 9 10 11 12 13 14 15 16 17 18 19 20 21 22 23 24 25 26 27 28 29 30 31 32 33 34 35 36 37 38 39 40 41 42 43 44 45 46 47 48 49 50 51 52 53 54 55 56 57 58 59 60 61 62 63 64 65 66 67 68 69 70 71 72 73 74 75 76 77 78 79 80 81 82 83 84 85 86 87 88 89 90 91 92 93 94 95 96 97 98 99 100 101 102 103 104 105 106 107 108 109 110 111 112 113 114 115 116 117 118 119 120 121 122 123 124 125 126 127 128 129 130 131 132 133 134 135 136 137 138 139 140 141 142 143 144 145 146 147 148 149 150 151 152 153 154 155 156 157 158 159 160 161 162 163 164 165 166 167 168 169 170 171 172 173 174 | """R@k tables and plots from benchmark_grid.py output (Table 4)."""
import argparse, json
from pathlib import Path
import numpy as np
import pandas as pd
import matplotlib
matplotlib.use("Agg")
import matplotlib.pyplot as plt
KS = [5, 10, 20, 50]
STYLES = ["terse", "casual", "vague", "detailed", "notation"]
METHOD_COLORS = {"ours": "#1a6faa", "pathfinder": "#c98a1e", "s2": "#7a4fa3"}
FORM_STYLES = {"orig": "-", "expanded": "--", "keywords": ":", "fused": "-."}
def load_tidy(path):
rows = []
for line in open(path):
rec = json.loads(line)
for style, cells in rec["cells"].items():
for cell_key, c in cells.items():
method, form = cell_key.split("/")
rows.append({
"arxiv_id": rec["arxiv_id"],
"fig_index": rec["fig_index"],
"target_in_pf": rec.get("target_in_pf"),
"style": style, "method": method, "form": form,
"fig_rank": c.get("fig_rank"),
"paper_rank": c.get("paper_rank"),
"pool_size": c.get("pool_size"),
"n_arxiv_ids": c.get("n_arxiv_ids"),
})
return pd.DataFrame(rows)
def rank_column(df, level):
if level == "fig":
return df.fig_rank
if level == "paper":
return df.paper_rank
# task: figure rank for ours, paper rank for baselines
return df.fig_rank.where(df.method == "ours", df.paper_rank)
def recall_table(df, level):
df = df.copy()
df["rank"] = rank_column(df, level)
out = []
for (style, method, form), g in df.groupby(["style", "method", "form"]):
entry = {"style": style, "method": method, "form": form,
"n": len(g)}
for k in KS:
entry[f"R@{k}"] = round(
float((g["rank"].notna() & (g["rank"] <= k)).mean()), 3)
if g.n_arxiv_ids.notna().any():
entry["s2_empty"] = round(
float((g.n_arxiv_ids == 0).mean()), 3)
out.append(entry)
t = pd.DataFrame(out)
t["style"] = pd.Categorical(t["style"], STYLES)
return t.sort_values(["style", "method", "form"]).reset_index(drop=True)
def print_blocks(t):
for style in STYLES:
block = t[t["style"] == style].drop(columns="style")
if block.empty:
continue
print(f"\n=== {style} ===")
print(block.to_string(index=False))
def plot_heatmap(t, level, k, stem):
col = f"R@{k}"
piv = t.pivot_table(index=["method", "form"], columns="style",
values=col, observed=True)
piv = piv.reindex(columns=[s for s in STYLES if s in piv.columns])
fig, ax = plt.subplots(
figsize=(1.3 * len(piv.columns) + 3, 0.42 * len(piv) + 1.5))
im = ax.imshow(piv.to_numpy(), vmin=0, vmax=1, cmap="viridis",
aspect="auto")
ax.set_xticks(range(len(piv.columns)), piv.columns)
ax.set_yticks(range(len(piv)),
[f"{m}/{f}" for m, f in piv.index])
for i in range(piv.shape[0]):
for j in range(piv.shape[1]):
v = piv.iloc[i, j]
if not np.isnan(v):
ax.text(j, i, f"{v:.2f}", ha="center", va="center",
fontsize=8,
color="white" if v < 0.55 else "black")
ax.set_title(f"{col}, level={level}")
fig.colorbar(im, ax=ax, shrink=0.8)
fig.tight_layout()
out = f"{stem}_heatmap_{level}_r{k}.png"
fig.savefig(out, dpi=200)
plt.close(fig)
return out
def plot_curves(t, level, stem):
fig, axes = plt.subplots(1, len(STYLES), figsize=(3.2 * len(STYLES), 3.4),
sharey=True)
for ax, style in zip(axes, STYLES):
block = t[t["style"] == style]
for _, r in block.iterrows():
ys = [r[f"R@{k}"] for k in KS]
ax.plot(KS, ys, color=METHOD_COLORS.get(r.method, "gray"),
linestyle=FORM_STYLES.get(r.form, "-"),
marker="o", ms=3,
label=f"{r.method}/{r.form}")
ax.set_title(style)
ax.set_xscale("log")
ax.set_xticks(KS, [str(k) for k in KS])
ax.set_xlabel("k")
ax.set_ylim(-0.03, 1.03)
ax.grid(alpha=0.25)
axes[0].set_ylabel(f"recall (level={level})")
handles, labels = axes[0].get_legend_handles_labels()
fig.legend(handles, labels, loc="center left",
bbox_to_anchor=(1.0, 0.5), fontsize=8, frameon=False)
fig.tight_layout()
out = f"{stem}_curves_{level}.png"
fig.savefig(out, dpi=200, bbox_inches="tight")
plt.close(fig)
return out
def main():
ap = argparse.ArgumentParser()
ap.add_argument("jsonl")
ap.add_argument("--level", default="task",
choices=["task", "fig", "paper"])
ap.add_argument("--k", type=int, default=20,
help="k for the heatmap")
ap.add_argument("--forms", default=None,
help="comma list to restrict, e.g. orig,keywords")
ap.add_argument("--covered-only", action="store_true",
help="restrict pathfinder rows to figures whose target "
"paper exists in the pathfinder corpus")
args = ap.parse_args()
stem = str(Path(args.jsonl).with_suffix(""))
df = load_tidy(args.jsonl)
n_figs = df[["arxiv_id", "fig_index"]].drop_duplicates().shape[0]
print(f"{n_figs} figures, {len(df)} cells, level={args.level}")
if args.forms:
keep = [f.strip() for f in args.forms.split(",")]
df = df[df.form.isin(keep)]
if args.covered_only:
drop = (df.method == "pathfinder") & (df.target_in_pf == False) # noqa: E712
print(f"covered-only: dropping {int(drop.sum())} pathfinder cells "
"whose target is absent from the pathfinder corpus")
df = df[~drop]
if df.target_in_pf.notna().any():
cov = df.drop_duplicates(["arxiv_id", "fig_index"]).target_in_pf
print(f"pathfinder corpus covers {cov.mean():.1%} of sampled targets")
t = recall_table(df, args.level)
print_blocks(t)
csv = f"{stem}_recall_{args.level}.csv"
t.to_csv(csv, index=False)
h = plot_heatmap(t, args.level, args.k, stem)
c = plot_curves(t, args.level, stem)
print(f"\nwrote: {csv}\n {h}\n {c}")
if __name__ == "__main__":
main()
|