By @rpinho
Refereed correctness bugs in open-source research software, with patches and links to upstream PRs
diff --git a/Code/alert_model/hpo_fpr.py b/Code/alert_model/hpo_fpr.py
index 15ac051..af937c6 100644
--- a/Code/alert_model/hpo_fpr.py
+++ b/Code/alert_model/hpo_fpr.py
@@ -5,7 +5,7 @@ import numpy as np
import optuna
import pandas as pd
from scipy.stats import loguniform
-from sklearn.metrics import recall_score
+from sklearn.metrics import recall_score, roc_curve
import lightgbm as lgb
import os
@@ -20,11 +20,10 @@ def tpr_at_fpr(labels, preds):
temp = results.sort_values(by="score", ascending=False)
FPR = 0.05
- N = (temp["true"] == 0).sum()
- FP = round(FPR * N)
- aux = temp[temp["true"] == 0]
-
- threshold = aux.iloc[FP - 1, 1]
+ fpr_values, _, thresholds = roc_curve(
+ results["true"], results["score"], drop_intermediate=False
+ )
+ threshold = thresholds[np.flatnonzero(fpr_values <= FPR)[-1]]
y_pred = np.where(results["score"] >= threshold, 1, 0)
tpr = recall_score(labels, y_pred)
@@ -184,11 +183,10 @@ class HPO:
temp = results.sort_values(by="score", ascending=False)
FPR = 0.05
- N = (temp["true"] == 0).sum()
- FP = round(FPR * N)
- aux = temp[temp["true"] == 0]
-
- threshold = aux.iloc[FP - 1, 1]
+ fpr_values, _, thresholds = roc_curve(
+ results["true"], results["score"], drop_intermediate=False
+ )
+ threshold = thresholds[np.flatnonzero(fpr_values <= FPR)[-1]]
y_pred = np.where(results["score"] >= threshold, 1, 0)
tpr = recall_score(y_val, y_pred)
diff --git a/Code/alert_model/training_and_predicting.py b/Code/alert_model/training_and_predicting.py
index 8337d5e..a6e7eb4 100644
--- a/Code/alert_model/training_and_predicting.py
+++ b/Code/alert_model/training_and_predicting.py
@@ -7,7 +7,7 @@ import os
import pickle
from sklearn import metrics
import matplotlib.pyplot as plt
-from sklearn.metrics import confusion_matrix
+from sklearn.metrics import confusion_matrix, roc_curve
from pathlib import Path
with open('../data/dataset_cfg.yaml', 'r') as infile:
@@ -61,11 +61,10 @@ def fpr_thresh(y_true, y_pred):
temp = results.sort_values(by="score", ascending=False)
FPR = 0.05
- N = (temp["true"] == 0).sum()
- FP = round(FPR * N)
- aux = temp[temp["true"] == 0]
-
- threshold = aux.iloc[FP - 1, 1]
+ fpr_values, _, thresholds = roc_curve(
+ results["true"], results["score"], drop_intermediate=False
+ )
+ threshold = thresholds[np.flatnonzero(fpr_values <= FPR)[-1]]
y_pred = np.where(results["score"] >= threshold, 1, 0)
tpr = metrics.recall_score(y_true, y_pred)
diff --git a/Code/data/preprocess.py b/Code/data/preprocess.py
index 9323b12..51c3c24 100644
--- a/Code/data/preprocess.py
+++ b/Code/data/preprocess.py
@@ -1,7 +1,7 @@
import yaml
import pandas as pd
import pickle
-from sklearn.metrics import recall_score
+from sklearn.metrics import recall_score, roc_curve
import numpy as np
import os
@@ -14,11 +14,10 @@ def tpr_at_fpr(labels, preds,fpr):
temp = results.sort_values(by="score", ascending=False)
FPR = fpr
- N = (temp["true"] == 0).sum()
- FP = round(FPR * N)
- aux = temp[temp["true"] == 0]
-
- threshold = aux.iloc[FP - 1, 1]
+ fpr_values, _, thresholds = roc_curve(
+ results["true"], results["score"], drop_intermediate=False
+ )
+ threshold = thresholds[np.flatnonzero(fpr_values <= FPR)[-1]]
y_pred = np.where(results["score"] >= threshold, 1, 0)
tpr = recall_score(labels, y_pred)