CDA1/clac_metric.py at master · Acceleratoreq/CDA1 · GitHub

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
import numpy as np


def get_metrics(real_score, predict_score):
    sorted_predict_score = np.array(sorted(list(set(np.array(predict_score).flatten()))))
    sorted_predict_score_num = len(sorted_predict_score)
    thresholds = sorted_predict_score[np.int32(sorted_predict_score_num * np.arange(1, 1000) / 1000)]
    thresholds = np.mat(thresholds)
    thresholds_num = thresholds.shape[1]

    predict_score_matrix = np.tile(predict_score, (thresholds_num, 1))
    negative_index = np.where(predict_score_matrix < thresholds.T)
    positive_index = np.where(predict_score_matrix >= thresholds.T)
    predict_score_matrix[negative_index] = 0
    predict_score_matrix[positive_index] = 1
    TP = predict_score_matrix.dot(real_score.T)
    FP = predict_score_matrix.sum(axis=1) - TP
    FN = real_score.sum() - TP
    TN = len(real_score.T) - TP - FP - FN

    fpr = FP / (FP + TN)
    tpr = TP / (TP + FN)
    ROC_dot_matrix = np.mat(sorted(np.column_stack((fpr, tpr)).tolist())).T
    ROC_dot_matrix.T[0] = [0, 0]
    ROC_dot_matrix = np.c_[ROC_dot_matrix, [1, 1]]
    x_ROC = ROC_dot_matrix[0].T
    y_ROC = ROC_dot_matrix[1].T
    auc = 0.5 * (x_ROC[1:] - x_ROC[:-1]).T * (y_ROC[:-1] + y_ROC[1:])

    recall_list = tpr
    precision_list = TP / (TP + FP)
    PR_dot_matrix = np.mat(sorted(np.column_stack((recall_list, precision_list)).tolist())).T
    PR_dot_matrix.T[0] = [0, 1]
    PR_dot_matrix = np.c_[PR_dot_matrix, [1, 0]]
    x_PR = PR_dot_matrix[0].T
    y_PR = PR_dot_matrix[1].T
    aupr = 0.5 * (x_PR[1:] - x_PR[:-1]).T * (y_PR[:-1] + y_PR[1:])

    f1_score_list = 2 * TP / (len(real_score.T) + TP - TN)
    accuracy_list = (TP + TN) / len(real_score.T)
    specificity_list = TN / (TN + FP)

    mcc_list = (TP * TN - FP * FN) / np.sqrt((TP + FP) * (TP + FN) * (TN + FP) * (TN + FN))
    max_index = np.argmax(f1_score_list)
    f1_score = f1_score_list[max_index]
    accuracy = accuracy_list[max_index]
    specificity = specificity_list[max_index]
    recall = recall_list[max_index]
    precision = precision_list[max_index]
    mcc = mcc_list[max_index]

    return [aupr[0, 0], auc[0, 0], f1_score, accuracy, recall, specificity, precision, mcc]


def cv_model_evaluate(interaction_matrix, predict_matrix, train_matrix):
    test_index = np.where(train_matrix == 0)
    real_score = interaction_matrix[test_index]
    predict_score = predict_matrix[test_index]
    return get_metrics(real_score, predict_score)