-
Notifications
You must be signed in to change notification settings - Fork 0
/
evaluation.py
56 lines (44 loc) · 1.43 KB
/
evaluation.py
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
import torch
import numpy as np
import tqdm
import sklearn.metrics as skm
def compute_metrics(y_true: np.ndarray, y_pred: np.ndarray):
y_labels = (y_pred > 0.5).astype(int)
accuracy = sum(y_labels == y_true) / len(y_true) * 100
tn, fp, fn, tp = skm.confusion_matrix(y_true, y_labels).ravel()
precision = tp/(tp + fp + 1e-7)
recall = tp/(tp + fn + 1e-7)
f1_score = (2 * precision * recall) / (precision + recall + 1e-7)
roc_auc = skm.roc_auc_score(y_true, y_pred)
result = {
"Accuracy": accuracy,
"Precision": precision,
"Recall": recall,
"F1 Score": f1_score,
"TN": tn,
"FP": fp,
"FN": fn,
"TP": tp,
"ROC_AUC": roc_auc
}
return result
def evaluate(model, X_test, y_test, batch_size=64, prefixes=None):
model.eval()
test_loader = torch.utils.data.DataLoader(
range(len(X_test)),
batch_size=batch_size,
shuffle=False,
)
y_pred = []
y_true = y_test.detach().cpu().numpy()
for i, indices in tqdm.tqdm(enumerate(test_loader)):
with torch.no_grad():
if prefixes is not None:
prefixes_slice = prefixes[indices]
else:
prefixes_slice = None
y_pred.extend(model(X_test[indices], prefixes=prefixes_slice).tolist())
y_pred = np.array(y_pred)
result = compute_metrics(y_true, y_pred)
print(result)
return result