def RGB_test(x_test,x_adv,y_test,classifier, pred_net):
x_test, x_adv, y_test = get_testing_data(x_test, x_adv, y_test, classifier)
x_adv_shape = x_adv.shape[0]
outputs_adv = pred_net(x_adv)
outputs = pred_net(x_test)
adv_pre = outputs_adv[:, 1]
x_pre = outputs[:, 1]
outputs_adv = tf.argmax(outputs_adv, 1)
outputs = tf.argmax(outputs, 1)
adv_score = tf.reduce_sum(outputs_adv)
x_score = tf.reduce_sum(outputs)
print('total_number:', x_adv_shape)
print('x_score:', x_score.numpy())
print('x_adv_score:', x_adv_shape - adv_score.numpy())
print('P:', x_score.numpy() / (x_score.numpy() + adv_score.numpy()))
print('R:', x_score.numpy() / x_adv_shape)
fpr, tpr, auc_score = compute_roc(adv_pre, x_pre)
print('auc:', auc_score)
concat = np.vstack((fpr, tpr))
return concat
print("Total Model Runtime: {}min, {:0.2f}sec".format(
int(time_elapsed // 60), time_elapsed % 60))
probabilities = np.squeeze(np.array(probabilities))
predictions = np.argmax(probabilities, axis=1)
labels = np.argmax(test_target, axis=1)
test_accuracy = np.sum(np.equal(predictions, labels)) / labels.size
print("Final Accuracy: {:0.4f}".format(test_accuracy))
# CNV [1, 0, 0 ,0]
# DRUSEN [0, 1, 0 ,0]
# DME [0, 0, 1 ,0]
# NORMAL [0, 0, 0 ,1]
LIST_OF_POS_IDX = [0]
auc_0, se_0, sp_0, acc_0 = utils.compute_roc(probabilities, labels,
LIST_OF_POS_IDX)
print(
"POS_IDX:{}, Final Model AUC: {:0.4f}, SE: {:0.4f}, SP: {:0.4f}, ACC: {:0.4f}"
.format(LIST_OF_POS_IDX, auc_0, se_0, sp_0, acc_0))
LIST_OF_POS_IDX = [1]
auc_1, se_1, sp_1, acc_1 = utils.compute_roc(probabilities, labels,
LIST_OF_POS_IDX)
print(
"POS_IDX:{}, Final Model AUC: {:0.4f}, SE: {:0.4f}, SP: {:0.4f}, ACC: {:0.4f}"
.format(LIST_OF_POS_IDX, auc_1, se_1, sp_1, acc_1))
LIST_OF_POS_IDX = [2]
auc_2, se_2, sp_2, acc_2 = utils.compute_roc(probabilities, labels,
LIST_OF_POS_IDX)
print(