File size: 4,538 Bytes
c22b544
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
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
#!/usr/bin/env python3
"""
Plot score distributions (REMOVED vs PRESENT) for each model.

Two figures — one per model — each with 3 subplots (SI-SNR, NXCorr, CLAP sim).
REMOVED and PRESENT distributions are overlaid with a vertical Youden T* line.
"""

from pathlib import Path

import numpy as np
import pandas as pd
import matplotlib
matplotlib.use("Agg")
import matplotlib.pyplot as plt
from sklearn.metrics import roc_auc_score, roc_curve

BASE_DIR = Path(__file__).parent

MODELS = {
    "combined_v1":          BASE_DIR / "experiments_final/combined_v1/eval_outputs_test_3k/event_detection_scores_gt.csv",
    "no_TSDL_old_mixtures": BASE_DIR / "experiments_final/no_TSDL_old_mixtures/eval_outputs_test_3k/event_detection_scores_gt.csv",
}

METRICS = [
    ("si_snr_db", "SI-SNR (dB)",      (-40, 20),  1.0),
    ("nxcorr",    "NXCorr",           (0,   1.0),  0.02),
    ("clap_sim",  "CLAP similarity",  (0,   1.0),  0.02),
]

COLORS = {"REMOVED": "#E05C5C", "PRESENT": "#4C9BE8"}
ALPHA  = 0.55


def load(path):
    df = pd.read_csv(path)
    df = df[df["error"].isna() | (df["error"] == "")]
    for col, *_ in METRICS:
        df[col] = pd.to_numeric(df[col], errors="coerce")
    df["gt_binary"] = df["gt_label"].map({"PRESENT": 1, "REMOVED": 0})
    return df


def youden_t(df, col):
    valid = df[col].notna() & df["gt_binary"].notna()
    fpr, tpr, thresholds = roc_curve(
        df.loc[valid, "gt_binary"], df.loc[valid, col]
    )
    j = tpr + (1 - fpr) - 1
    return float(thresholds[np.argmax(j)])


def dprime(present, removed):
    sigma_pooled = np.sqrt((present.std()**2 + removed.std()**2) / 2)
    return (present.mean() - removed.mean()) / sigma_pooled


def pooled_youden_t(dfs, col):
    all_s = pd.concat([df[col].dropna() for df in dfs.values()])
    all_l = pd.concat([
        df.loc[df[col].notna(), "gt_binary"].dropna()
        for df in dfs.values()
    ])
    valid = all_s.notna() & all_l.notna()
    fpr, tpr, thresholds = roc_curve(all_l[valid].values, all_s[valid].values)
    j = tpr + (1 - fpr) - 1
    return float(thresholds[np.argmax(j)])


def plot_model(df, model_name, pooled_thresholds, out_path):
    fig, axes = plt.subplots(1, 3, figsize=(15, 5.0))
    fig.suptitle(
        f"Score distributions: REMOVED vs PRESENT  |  Model: {model_name}",
        fontsize=13, fontweight="bold", y=1.01,
    )

    for ax, (col, xlabel, xlim, bw) in zip(axes, METRICS):
        present = df.loc[df["gt_label"] == "PRESENT", col].dropna()
        removed = df.loc[df["gt_label"] == "REMOVED",  col].dropna()

        bins = np.arange(xlim[0], xlim[1] + bw, bw)
        for vals, label in [(removed, "REMOVED"), (present, "PRESENT")]:
            ax.hist(vals, bins=bins, color=COLORS[label], alpha=ALPHA,
                    label=f"{label} (n={len(vals)})", density=True, edgecolor="none")

        # Youden T* line
        t = pooled_thresholds[col]
        ax.axvline(t, color="black", linewidth=1.5, linestyle="--",
                   label=f"Youden T*={t:.3f}")

        # AUC and d' annotation
        valid = df[col].notna() & df["gt_binary"].notna()
        auc = roc_auc_score(df.loc[valid, "gt_binary"], df.loc[valid, col])
        dp  = dprime(present.values, removed.values)
        ax.text(0.97, 0.97, f"AUC = {auc:.4f}\nd′  = {dp:.4f}",
                transform=ax.transAxes, fontsize=9,
                verticalalignment="top", horizontalalignment="right",
                bbox=dict(boxstyle="round,pad=0.3", facecolor="white",
                          edgecolor="grey", alpha=0.8))

        ax.set_xlabel(xlabel, fontsize=11)
        ax.set_ylabel("Density", fontsize=10)
        ax.set_xlim(xlim)
        ax.legend(fontsize=8.5, framealpha=0.7)
        ax.set_title(xlabel, fontsize=11)
        ax.spines[["top", "right"]].set_visible(False)

    plt.tight_layout()
    fig.savefig(out_path, dpi=150, bbox_inches="tight")
    plt.close(fig)
    print(f"Saved: {out_path}")


def main():
    dfs = {name: load(path) for name, path in MODELS.items() if path.exists()}
    if not dfs:
        print("No CSVs found.")
        return

    # Compute pooled Youden thresholds (same for both plots)
    pooled_t = {col: pooled_youden_t(dfs, col) for col, *_ in METRICS}
    print("Pooled Youden T*:")
    for col, label, *_ in METRICS:
        print(f"  {label:<20}: {pooled_t[col]:.4f}")

    for model_name, df in dfs.items():
        out = BASE_DIR / f"gt_score_hist_{model_name}.png"
        plot_model(df, model_name, pooled_t, out)


if __name__ == "__main__":
    main()