Update app.py
Browse files
app.py
CHANGED
|
@@ -478,6 +478,10 @@ def hr_classify():
|
|
| 478 |
for k, v in fi],
|
| 479 |
}
|
| 480 |
|
|
|
|
|
|
|
|
|
|
|
|
|
| 481 |
# ── Best model predictions ──
|
| 482 |
best_model_name = model_type if model_type != 'all' else 'xgboost'
|
| 483 |
if best_model_name not in results:
|
|
@@ -494,7 +498,31 @@ def hr_classify():
|
|
| 494 |
|
| 495 |
best_clf.fit(X_train, y_train)
|
| 496 |
all_probs = best_clf.predict_proba(X)[:, 1]
|
| 497 |
-
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
| 498 |
|
| 499 |
# ── Employee predictions ──
|
| 500 |
predictions = []
|
|
@@ -562,6 +590,8 @@ def hr_classify():
|
|
| 562 |
"top_at_risk": sorted(predictions,
|
| 563 |
key=lambda x: x['resign_probability'],
|
| 564 |
reverse=True)[:10],
|
|
|
|
|
|
|
| 565 |
"disclaimer": "ผลการพยากรณ์ใช้เพื่อสนับสนุนการตัดสินใจ HR เท่านั้น"
|
| 566 |
})
|
| 567 |
|
|
|
|
| 478 |
for k, v in fi],
|
| 479 |
}
|
| 480 |
|
| 481 |
+
# ── Threshold ──
|
| 482 |
+
threshold = float(body.get("threshold", 0.5))
|
| 483 |
+
threshold = max(0.1, min(0.9, threshold))
|
| 484 |
+
|
| 485 |
# ── Best model predictions ──
|
| 486 |
best_model_name = model_type if model_type != 'all' else 'xgboost'
|
| 487 |
if best_model_name not in results:
|
|
|
|
| 498 |
|
| 499 |
best_clf.fit(X_train, y_train)
|
| 500 |
all_probs = best_clf.predict_proba(X)[:, 1]
|
| 501 |
+
# ใช้ threshold ที่กำหนด แทน default 0.5
|
| 502 |
+
all_preds = (all_probs >= threshold).astype(int)
|
| 503 |
+
|
| 504 |
+
# ── Threshold analysis (สำหรับ chart) ──
|
| 505 |
+
thresholds_range = [round(t/10, 1) for t in range(1, 10)]
|
| 506 |
+
threshold_analysis = []
|
| 507 |
+
if has_label:
|
| 508 |
+
test_probs = best_clf.predict_proba(X_test)[:, 1]
|
| 509 |
+
for t in thresholds_range:
|
| 510 |
+
t_preds = (test_probs >= t).astype(int)
|
| 511 |
+
tp = int(((t_preds == 1) & (y_test == 1)).sum())
|
| 512 |
+
fp = int(((t_preds == 1) & (y_test == 0)).sum())
|
| 513 |
+
fn = int(((t_preds == 0) & (y_test == 1)).sum())
|
| 514 |
+
tn = int(((t_preds == 0) & (y_test == 0)).sum())
|
| 515 |
+
prec = tp/(tp+fp) if (tp+fp) > 0 else 0
|
| 516 |
+
rec = tp/(tp+fn) if (tp+fn) > 0 else 0
|
| 517 |
+
f1 = 2*prec*rec/(prec+rec) if (prec+rec) > 0 else 0
|
| 518 |
+
threshold_analysis.append({
|
| 519 |
+
"threshold": t,
|
| 520 |
+
"precision_pct": round(prec*100, 1),
|
| 521 |
+
"recall_pct": round(rec*100, 1),
|
| 522 |
+
"f1_pct": round(f1*100, 1),
|
| 523 |
+
"flagged": int(t_preds.sum()),
|
| 524 |
+
"tp": tp, "fp": fp, "fn": fn, "tn": tn
|
| 525 |
+
})
|
| 526 |
|
| 527 |
# ── Employee predictions ──
|
| 528 |
predictions = []
|
|
|
|
| 590 |
"top_at_risk": sorted(predictions,
|
| 591 |
key=lambda x: x['resign_probability'],
|
| 592 |
reverse=True)[:10],
|
| 593 |
+
"threshold_used": threshold,
|
| 594 |
+
"threshold_analysis": threshold_analysis if has_label else [],
|
| 595 |
"disclaimer": "ผลการพยากรณ์ใช้เพื่อสนับสนุนการตัดสินใจ HR เท่านั้น"
|
| 596 |
})
|
| 597 |
|