Spaces:
Sleeping
Sleeping
File size: 11,931 Bytes
3d2a487 | 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 99 100 101 102 103 104 105 106 107 108 109 110 111 112 113 114 115 116 117 118 119 120 121 122 123 124 125 126 127 128 129 130 131 132 133 134 135 136 137 138 139 140 141 142 143 144 145 146 147 148 149 150 151 152 153 154 155 156 157 158 159 160 161 162 163 164 165 166 167 168 169 170 171 172 173 174 175 176 177 178 179 180 181 182 183 184 185 186 187 188 189 190 191 192 193 194 195 196 197 198 199 200 201 202 203 204 205 206 207 208 209 210 211 212 213 214 215 216 217 218 219 220 221 222 223 224 225 226 227 228 229 230 231 232 233 234 235 236 237 238 239 240 241 242 243 244 | """Feature selection analysis β combine permutation + saliency scores,
compute rank correlation, and produce selection recommendations.
Run directly:
python -m src.pipeline.eval.feature_selection_analysis
"""
from __future__ import annotations
import json
import sys
from pathlib import Path
import matplotlib
matplotlib.use("Agg")
import matplotlib.pyplot as plt
import matplotlib.gridspec as gridspec
import numpy as np
from scipy.stats import spearmanr
if __package__ is None or __package__ == "":
sys.path.append(str(Path(__file__).resolve().parents[4]))
from src.config import load_default_config, resolve_project_path
# βββββββββββββββββββββββββββββββββββββββββββββββββββββββββββββββββββββββββββββ
# Analysis
# βββββββββββββββββββββββββββββββββββββββββββββββββββββββββββββββββββββββββββββ
def analyse(fi_path: Path, graphs_dir: Path) -> dict:
data = json.loads(fi_path.read_text(encoding="utf-8"))
perm: dict[str, float] = data["permutation"]
sal: dict[str, float] = data["saliency"]
features = list(perm.keys()) # already sorted by permutation score desc
perm_scores = np.array([perm[f] for f in features])
sal_scores = np.array([sal[f] for f in features])
# ββ Rank-normalise both methods to [0, 1] βββββββββββββββββββββββββββββββββ
def rank_norm(arr: np.ndarray) -> np.ndarray:
ranks = arr.argsort().argsort().astype(float)
return ranks / (len(ranks) - 1)
perm_norm = rank_norm(perm_scores)
sal_norm = rank_norm(sal_scores)
# ββ Ensemble score (equal weight average of normalised ranks) βββββββββββββ
ensemble = (perm_norm + sal_norm) / 2.0
ensemble_order = np.argsort(ensemble)[::-1] # descending
features_ranked = [features[i] for i in ensemble_order]
ensemble_ranked = ensemble[ensemble_order]
# ββ Spearman rank correlation between methods βββββββββββββββββββββββββββββ
rho, pval = spearmanr(perm_scores, sal_scores)
# ββ Tier classification βββββββββββββββββββββββββββββββββββββββββββββββββββ
n = len(features_ranked)
top_n = max(1, round(n * 0.30)) # top 30 % β keep
bottom_n = max(1, round(n * 0.25)) # bottom 25 % β candidates for removal
top_features = features_ranked[:top_n]
middle_features = features_ranked[top_n: n - bottom_n]
bottom_features = features_ranked[n - bottom_n:]
result = {
"spearman_rho": round(float(rho), 4),
"spearman_pval": round(float(pval), 6),
"agreement_note": _agreement_note(rho),
"ensemble_ranking": {f: round(float(s), 6) for f, s in zip(features_ranked, ensemble_ranked)},
"tiers": {
"top_keep": {"features": top_features, "count": len(top_features)},
"middle_watch": {"features": middle_features, "count": len(middle_features)},
"bottom_drop_candidates": {"features": bottom_features, "count": len(bottom_features)},
},
"recommendation": _recommendation(top_features, bottom_features, rho),
}
# ββ Plots ββββββββββββββββββββββββββββββββββββββββββββββββββββββββββββββββββ
graphs_dir.mkdir(parents=True, exist_ok=True)
_plot_combined(features, perm_scores, sal_scores, ensemble,
ensemble_order, top_n, bottom_n, rho, graphs_dir)
_plot_scatter(features, perm_scores, sal_scores, rho, pval, graphs_dir)
# ββ Save JSON ββββββββββββββββββββββββββββββββββββββββββββββββββββββββββββββ
out_path = graphs_dir / "feature_selection_analysis.json"
out_path.write_text(json.dumps(result, indent=2), encoding="utf-8")
print(f"\n[analysis] Saved β '{out_path}'")
return result
# βββββββββββββββββββββββββββββββββββββββββββββββββββββββββββββββββββββββββββββ
# Helpers
# βββββββββββββββββββββββββββββββββββββββββββββββββββββββββββββββββββββββββββββ
def _agreement_note(rho: float) -> str:
if rho >= 0.85:
return "Strong agreement β both methods rank features consistently."
if rho >= 0.60:
return "Moderate agreement β ensemble score is reliable."
return "Low agreement β treat rankings with caution; run more repeats."
def _recommendation(top: list[str], bottom: list[str], rho: float) -> str:
top_str = ", ".join(top)
bottom_str = ", ".join(bottom)
caveat = (
" Note: scores are very close β this dataset may not benefit much from "
"feature dropping; retrain with the reduced set to confirm."
if rho > 0.9 else ""
)
return (
f"KEEP (highest combined importance): {top_str}. "
f"CONSIDER DROPPING (lowest combined importance): {bottom_str}.{caveat}"
)
def _plot_combined(features, perm_scores, sal_scores, ensemble,
ensemble_order, top_n, bottom_n, rho, graphs_dir):
n = len(features)
fig = plt.figure(figsize=(16, 10))
gs = gridspec.GridSpec(1, 3, figure=fig, wspace=0.45)
# ββ Left: Permutation βββββββββββββββββββββββββββββββββββββββββββββββββββββ
ax1 = fig.add_subplot(gs[0])
feats_p = list(perm_scores.argsort()[::-1])
colors_p = ["#2196F3"] * n
for i in range(top_n): colors_p[feats_p[i]] = "#4CAF50"
for i in range(1, bottom_n + 1): colors_p[feats_p[-i]] = "#F44336"
ax1.barh([features[i] for i in feats_p][::-1],
[perm_scores[i] for i in feats_p][::-1],
color=[colors_p[i] for i in feats_p][::-1])
ax1.set_title("Permutation\n(Ξ PR-AUC)", fontsize=11)
ax1.set_xlabel("Score")
ax1.tick_params(axis="y", labelsize=8)
ax1.grid(axis="x", alpha=0.3)
# ββ Middle: Saliency ββββββββββββββββββββββββββββββββββββββββββββββββββββββ
ax2 = fig.add_subplot(gs[1])
feats_s = list(sal_scores.argsort()[::-1])
colors_s = ["#2196F3"] * n
for i in range(top_n): colors_s[feats_s[i]] = "#4CAF50"
for i in range(1, bottom_n + 1): colors_s[feats_s[-i]] = "#F44336"
ax2.barh([features[i] for i in feats_s][::-1],
[sal_scores[i] for i in feats_s][::-1],
color=[colors_s[i] for i in feats_s][::-1])
ax2.set_title("Gradient Saliency\n(mean |βout/βin|)", fontsize=11)
ax2.set_xlabel("Score")
ax2.tick_params(axis="y", labelsize=8)
ax2.grid(axis="x", alpha=0.3)
# ββ Right: Ensemble βββββββββββββββββββββββββββββββββββββββββββββββββββββββ
ax3 = fig.add_subplot(gs[2])
colors_e = []
for i, idx in enumerate(ensemble_order):
if i < top_n: colors_e.append("#4CAF50")
elif i >= n - bottom_n: colors_e.append("#F44336")
else: colors_e.append("#2196F3")
ax3.barh([features[i] for i in ensemble_order][::-1],
ensemble[ensemble_order][::-1],
color=colors_e[::-1])
ax3.set_title("Ensemble Score\n(rank-normalised avg)", fontsize=11)
ax3.set_xlabel("Score [0β1]")
ax3.tick_params(axis="y", labelsize=8)
ax3.grid(axis="x", alpha=0.3)
# Legend
from matplotlib.patches import Patch
legend_elements = [
Patch(facecolor="#4CAF50", label=f"Top {top_n} β Keep"),
Patch(facecolor="#2196F3", label="Middle β Watch"),
Patch(facecolor="#F44336", label=f"Bottom {bottom_n} β Drop candidates"),
]
fig.legend(handles=legend_elements, loc="lower center", ncol=3,
fontsize=9, bbox_to_anchor=(0.5, -0.02))
fig.suptitle(f"Feature Importance Analysis | Spearman Ο = {rho:.3f}", fontsize=13)
path = graphs_dir / "feature_selection_combined.png"
fig.savefig(path, dpi=130, bbox_inches="tight")
plt.close(fig)
print(f"[analysis] Chart saved β '{path}'")
def _plot_scatter(features, perm_scores, sal_scores, rho, pval, graphs_dir):
fig, ax = plt.subplots(figsize=(7, 6))
ax.scatter(perm_scores, sal_scores, s=60, color="#1565C0", zorder=3)
for i, f in enumerate(features):
ax.annotate(f, (perm_scores[i], sal_scores[i]),
fontsize=6.5, xytext=(4, 2), textcoords="offset points")
# Trend line
m, b = np.polyfit(perm_scores, sal_scores, 1)
x_range = np.linspace(perm_scores.min(), perm_scores.max(), 100)
ax.plot(x_range, m * x_range + b, color="#E53935", linewidth=1.5,
linestyle="--", label=f"trend (Ο={rho:.3f}, p={pval:.4f})")
ax.set_xlabel("Permutation Importance (Ξ PR-AUC)")
ax.set_ylabel("Gradient Saliency (mean |β/βx|)")
ax.set_title("Method Agreement β Permutation vs Saliency")
ax.legend(fontsize=9)
ax.grid(alpha=0.3)
path = graphs_dir / "feature_selection_scatter.png"
fig.savefig(path, dpi=130, bbox_inches="tight")
plt.close(fig)
print(f"[analysis] Chart saved β '{path}'")
# βββββββββββββββββββββββββββββββββββββββββββββββββββββββββββββββββββββββββββββ
# Main
# βββββββββββββββββββββββββββββββββββββββββββββββββββββββββββββββββββββββββββββ
def main() -> None:
cfg = load_default_config()
graphs_dir = resolve_project_path(cfg["paths"].get("graphs_dir", "artifacts/graphs"))
fi_path = graphs_dir / "feature_importance.json"
if not fi_path.exists():
print(f"[analysis] ERROR: '{fi_path}' not found. Run dvc repro (train stage) first.")
sys.exit(1)
result = analyse(fi_path, graphs_dir)
print("\n" + "=" * 70)
print(f" Spearman Ο between methods : {result['spearman_rho']} ({result['agreement_note']})")
print("=" * 70)
print("\n ENSEMBLE RANKING (best β worst):")
for rank, (feat, score) in enumerate(result["ensemble_ranking"].items(), 1):
tier = ("β
KEEP " if feat in result["tiers"]["top_keep"]["features"]
else "β DROP? " if feat in result["tiers"]["bottom_drop_candidates"]["features"]
else " Β· ")
print(f" {rank:>2}. {tier} {feat:<42} {score:.4f}")
print("\n" + "=" * 70)
print(" RECOMMENDATION:")
print(f" {result['recommendation']}")
print("=" * 70 + "\n")
if __name__ == "__main__":
main()
|