Upload evaluate.py
Browse files- evaluate.py +89 -0
evaluate.py
ADDED
@@ -0,0 +1,89 @@
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
1 |
+
import numpy as np
|
2 |
+
from sklearn.metrics import accuracy_score, precision_score, recall_score, f1_score, roc_auc_score, average_precision_score
|
3 |
+
import pandas as pd
|
4 |
+
import matplotlib.pyplot as plt
|
5 |
+
import seaborn as sns
|
6 |
+
|
7 |
+
def evaluate_metrics(labels, probs, target_cols, threshold=0.5):
|
8 |
+
"""
|
9 |
+
Evaluate metrics for each label
|
10 |
+
|
11 |
+
Parameters:
|
12 |
+
-----------
|
13 |
+
labels : numpy.ndarray
|
14 |
+
Ground truth labels (0 or 1)
|
15 |
+
probs : numpy.ndarray
|
16 |
+
Prediction probabilities
|
17 |
+
target_cols : list
|
18 |
+
List of target disease names
|
19 |
+
threshold : float
|
20 |
+
Threshold for converting probabilities to binary predictions
|
21 |
+
|
22 |
+
Returns:
|
23 |
+
--------
|
24 |
+
pandas.DataFrame
|
25 |
+
Metrics for each label
|
26 |
+
"""
|
27 |
+
# Convert probabilities to binary predictions
|
28 |
+
preds = (probs >= threshold).astype(int)
|
29 |
+
|
30 |
+
# Dictionary to store metrics
|
31 |
+
metrics_dict = {
|
32 |
+
'Disease': target_cols,
|
33 |
+
'Positive_Samples': [],
|
34 |
+
'Accuracy': [],
|
35 |
+
'Precision': [],
|
36 |
+
'Recall': [],
|
37 |
+
'F1': [],
|
38 |
+
'AUC-ROC': [],
|
39 |
+
'AP': [] # Average Precision
|
40 |
+
}
|
41 |
+
|
42 |
+
# Calculate metrics for each label
|
43 |
+
for i in range(len(target_cols)):
|
44 |
+
metrics_dict['Positive_Samples'].append(np.sum(labels[:, i]))
|
45 |
+
metrics_dict['Accuracy'].append(accuracy_score(labels[:, i], preds[:, i]))
|
46 |
+
metrics_dict['Precision'].append(precision_score(labels[:, i], preds[:, i], zero_division=0))
|
47 |
+
metrics_dict['Recall'].append(recall_score(labels[:, i], preds[:, i], zero_division=0))
|
48 |
+
metrics_dict['F1'].append(f1_score(labels[:, i], preds[:, i], zero_division=0))
|
49 |
+
metrics_dict['AUC-ROC'].append(roc_auc_score(labels[:, i], probs[:, i]))
|
50 |
+
metrics_dict['AP'].append(average_precision_score(labels[:, i], probs[:, i]))
|
51 |
+
|
52 |
+
# Convert to DataFrame
|
53 |
+
metrics_df = pd.DataFrame(metrics_dict)
|
54 |
+
|
55 |
+
# Round numerical values to 3 decimal places
|
56 |
+
numeric_cols = ['Accuracy', 'Precision', 'Recall', 'F1', 'AUC-ROC', 'AP']
|
57 |
+
metrics_df[numeric_cols] = metrics_df[numeric_cols].round(3)
|
58 |
+
|
59 |
+
return metrics_df
|
60 |
+
|
61 |
+
def plot_metrics_heatmap(metrics_df, metric_cols=['Precision', 'Recall', 'F1', 'AUC-ROC', 'AP']):
|
62 |
+
"""
|
63 |
+
Plot a heatmap of evaluation metrics
|
64 |
+
|
65 |
+
Parameters:
|
66 |
+
-----------
|
67 |
+
metrics_df : pandas.DataFrame
|
68 |
+
Output from evaluate_metrics function
|
69 |
+
metric_cols : list
|
70 |
+
Metrics to display in the heatmap
|
71 |
+
"""
|
72 |
+
plt.figure(figsize=(12, len(target_cols)//2))
|
73 |
+
|
74 |
+
# Prepare data for heatmap
|
75 |
+
heatmap_data = metrics_df[metric_cols].values
|
76 |
+
|
77 |
+
# Draw heatmap
|
78 |
+
sns.heatmap(heatmap_data,
|
79 |
+
annot=True,
|
80 |
+
fmt='.3f',
|
81 |
+
cmap='YlOrRd',
|
82 |
+
xticklabels=metric_cols,
|
83 |
+
yticklabels=metrics_df['Disease'],
|
84 |
+
vmin=0,
|
85 |
+
vmax=1)
|
86 |
+
|
87 |
+
plt.title('Evaluation Metrics Heatmap')
|
88 |
+
plt.tight_layout()
|
89 |
+
plt.show()
|