機械学習時など，モデルの性能評価で必要なROC, AUCですが
cutoff ポイントを自作したいとき，不便なので自分で作ってみました。
(もしかして誤解があったら，気づいた方は教えてください！)

For ROC curve after model construction

ROC.py


# Code for CutOff_and_ROCcurve.
Predict_arr = model.predict(test_arr)

cutoff_list = np.arange(0, 1.001, 0.001)

cut_sen_spe = []
for cutoff in cutoff_list:
    pred_answer_arr = Predict_arr.copy()
    pred_answer_arr[pred_answer_arr >= cutoff] = 1
    pred_answer_arr[pred_answer_arr < cutoff] = 0
    PP = 0
    PN = 0
    NP = 0
    NN = 0
    for i in range(len(test_answer_arr)):
        if test_answer_arr[i] == 1 and pred_answer_arr[i] == 1:
            PP += 1
        if test_answer_arr[i] == 1 and pred_answer_arr[i] == 0:
            PN += 1
        if test_answer_arr[i] == 0 and pred_answer_arr[i] == 1:
            NP += 1
        if test_answer_arr[i] == 0 and pred_answer_arr[i] == 0:
            NN += 1
    cut_sen_spe.append([cutoff, PP, PN, NP, NN])

total_list = []
for j in range(len(cut_sen_spe)):
    sen = cut_sen_spe[j][1] / (cut_sen_spe[j][1] + cut_sen_spe[j][2])
    spe = cut_sen_spe[j][4] / (cut_sen_spe[j][3] + cut_sen_spe[j][4])
    total_list.append([sen, spe])

total_arr = np.array(total_list)

sen_arr = total_arr[:, 0]
spe_arr = total_arr[:, 1]


# Visualization & Save ROCcurve

f = plt.figure(figsize = (10, 10))
plt.plot(1-spe_arr, sen_arr, c = "r", linewidth = 3.0)
plt.plot([0.0, 1.0], [0.0, 1.0], linestyle='dashed')
plt.xticks(np.arange(0, 1.01, 0.1))
plt.yticks(np.arange(0, 1.01, 0.1))
plt.xlim(-0.03, 1.03)
plt.ylim(-0.03, 1.03)
plt.savefig(ROCcurve_Save)
plt.show()

# AUC 
target_falPos_arr = 1-spe_arr
target_sen_arr = sen_arr
target_falPos_arr.sort()
target_sen_arr.sort()

AUC_segIMG = 0
for ind in range(len(target_falPos_arr[:-1])):
    X_p = target_falPos_arr[ind]
    nex_X = target_falPos_arr[ind+1]
    Y_p = target_sen_arr[ind]
    nex_Y = target_sen_arr[ind+1]
    area = (nex_X - X_p) * Y_p + (nex_X - X_p) * (nex_Y - Y_p) / 2
    AUC_segIMG += area

print(AUC_segIMG)