File size: 6,491 Bytes
f28d994
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
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
"""Data loading and audit helpers for figures_v2."""
from __future__ import annotations

import csv
import re
from pathlib import Path

import numpy as np
import pandas as pd


def path_exists(path: Path) -> bool:
    """Explicit existence helper used by the audit and loaders."""
    return Path(path).exists()


def read_table(path: Path, **kwargs) -> pd.DataFrame | None:
    if not path_exists(path):
        return None
    return pd.read_csv(path, **kwargs)


def read_edges(path: Path) -> pd.DataFrame | None:
    if not path_exists(path):
        return None
    return pd.read_csv(path, sep=r"\s+", header=None, engine="python")


def load_dataset_degrees(root: Path) -> dict[str, np.ndarray] | None:
    dd = root / "data_and_docs"
    co = read_edges(dd / "author_file_ann.txt")
    cite = read_edges(dd / "paper_file_ann.txt")
    read = read_edges(dd / "bipartite_train_ann.txt")
    if co is None or cite is None or read is None:
        return None
    co_deg = np.bincount(np.concatenate([co[0].to_numpy(), co[1].to_numpy()]), minlength=6611)
    citation_in = np.bincount(cite[1].to_numpy(), minlength=79937)
    paper_read = np.bincount(read[1].to_numpy(), minlength=79937)
    author_read = np.bincount(read[0].to_numpy(), minlength=6611)
    return {
        "coauthor_degree": co_deg,
        "citation_indegree": citation_in,
        "paper_read_degree": paper_read,
        "author_read_degree": author_read,
    }


def load_manual_metrics(root: Path) -> pd.DataFrame:
    path = root / "figures_v2" / "data" / "manual_metrics.csv"
    if path_exists(path):
        return pd.read_csv(path)
    rows = [
        (0, "gnn_baseline", "GNN\nbaseline", 0.8850, np.nan, "reports"),
        (1, "lightgcn_ensemble", "LightGCN\nensemble", 0.938576, 0.93044, "validation_runs/dynamic_summary.csv; README"),
        (2, "graph_stack", "+ graph/meta-path\nstack", 0.95599, 0.95760, "validation_runs/stack_threshold_summary.csv; README"),
        (3, "content_bpr", "+ content\n+ BPR-MF", 0.95930, 0.95996, "reports; README"),
        (4, "deepwalk_node2vec", "+ DeepWalk\n/ Node2Vec", 0.96213, 0.96252, "error_group_calibration anchor; README"),
        (5, "rw7", "+ 7 RW\nblocks", 0.964947, np.nan, "high_order validation_summary.csv"),
        (6, "highorder_directed", "+ high-order\ncitation", 0.966874, 0.96626, "high_order validation_summary.csv; README"),
    ]
    return pd.DataFrame(rows, columns=["order", "stage", "label", "val_f1", "public_f1", "source"])


def high_order_summary(root: Path) -> pd.DataFrame:
    path = root / "validation_runs" / "dynamic_seed202" / "high_order_graph_stack" / "validation_summary.csv"
    df = read_table(path)
    if df is not None:
        return df
    return pd.DataFrame(
        [
            ("base_highorder", 0.9642697338013148, 0.4554775357246399, 0.994052111749616, 0.9653815084086069, 0.9631605169836758, 108),
            ("rich_rw7", 0.9649474248055991, 0.49044686555862427, 0.9945549026665483, 0.9663869251458722, 0.9635122065590106, 190),
            ("rich_rw7_highorder", 0.9665557233547776, 0.46933943033218384, 0.9948903494937357, 0.9671087377501321, 0.9660033410509656, 214),
            ("rich_rw7_highorder_directed", 0.966873736337297, 0.46173080801963806, 0.9949182985645343, 0.9667037764040665, 0.9670437560446645, 259),
        ],
        columns=["stage", "validation_f1", "threshold", "auc", "precision", "recall", "n_features"],
    )


def rw_ensemble_metrics(root: Path) -> tuple[list[int], list[float], list[str]]:
    rw = root / "validation_runs" / "dynamic_seed202" / "randomwalk_systematic"
    sources: list[str] = []
    single_df = read_table(rw / "small_ablation_table.csv")
    e5_df = read_table(rw / "ensemble_5_ablation.csv")
    e7_df = read_table(rw / "ensemble_7_ablation.csv")
    if single_df is None or e5_df is None or e7_df is None:
        return [1, 5, 7], [0.96310, 0.96393, 0.96492], ["reported fallback values"]
    sources = [str(rw / "small_ablation_table.csv"), str(rw / "ensemble_5_ablation.csv"), str(rw / "ensemble_7_ablation.csv")]
    return [1, 5, 7], [
        float(single_df["validation_F1"].max()),
        float(e5_df["validation_F1"].iloc[0]),
        float(e7_df["validation_F1"].iloc[0]),
    ], sources


def load_npy(path: Path) -> np.ndarray | None:
    if not path_exists(path):
        return None
    return np.load(path)


def pr_curve(y_true: np.ndarray, scores: np.ndarray) -> tuple[np.ndarray, np.ndarray, float]:
    order = np.argsort(-scores)
    y = y_true[order].astype(float)
    tp = np.cumsum(y)
    fp = np.cumsum(1.0 - y)
    precision = tp / np.maximum(tp + fp, 1.0)
    recall = tp / max(float(y.sum()), 1.0)
    precision = np.r_[1.0, precision]
    recall = np.r_[0.0, recall]
    ap = float(np.sum((recall[1:] - recall[:-1]) * precision[1:]))
    return recall, precision, ap


def numeric_bucket_key(bucket: str) -> float:
    toks = re.findall(r"-?inf|-?\d+(?:\.\d+)?(?:e-?\d+)?", str(bucket))
    if not toks:
        return 0.0
    val = toks[0]
    if val == "-inf":
        return -1e18
    if val == "inf":
        return 1e18
    return float(val)


def inventory_files(root: Path, out_csv: Path) -> pd.DataFrame:
    watched = [
        "README.md",
        "CLAUDE.md",
        "WORKSPACE_STATUS.md",
        "reports/final_report.md",
        "reports/exploration_summary.md",
        "reports/preliminary_report.md",
        "notes/experiment_history.md",
        "data_and_docs",
        "validation_runs",
        "cached_scores",
        "submissions",
        "code",
        "figures_paper",
    ]
    rows = []
    for rel in watched:
        p = root / rel
        if not path_exists(p):
            rows.append({"path": rel, "exists": False, "kind": "missing", "size_bytes": ""})
            continue
        if p.is_file():
            rows.append({"path": rel, "exists": True, "kind": "file", "size_bytes": p.stat().st_size})
        else:
            for child in p.rglob("*"):
                if child.is_file():
                    rows.append(
                        {
                            "path": child.relative_to(root).as_posix(),
                            "exists": True,
                            "kind": "file",
                            "size_bytes": child.stat().st_size,
                        }
                    )
    df = pd.DataFrame(rows)
    out_csv.parent.mkdir(parents=True, exist_ok=True)
    df.to_csv(out_csv, index=False, quoting=csv.QUOTE_MINIMAL)
    return df