#首先定義一個 分類數*分類數 的空混淆矩陣
 conf_matrix = torch.zeros(Emotion_kinds, Emotion_kinds)
 # 使用torch.no_grad()可以顯著降低測試用例的GPU占用
    with torch.no_grad():
        for step, (imgs, targets) in enumerate(test_loader):
            # imgs:     torch.Size([50, 3, 200, 200])   torch.FloatTensor
            # targets:  torch.Size([50, 1]),     torch.LongTensor  多了一維，所以我們要把其去掉
            targets = targets.squeeze()  # [50,1] ----->  [50]

            # 將變量轉為gpu
            targets = targets.cuda()
            imgs = imgs.cuda()
            # print(step,imgs.shape,imgs.type(),targets.shape,targets.type())
            
            out = model(imgs)
            #記錄混淆矩陣參數
            conf_matrix = confusion_matrix(out, targets, conf_matrix)
            conf_matrix=conf_matrix.cpu()

def confusion_matrix(preds, labels, conf_matrix):
    preds = torch.argmax(preds, 1)
    for p, t in zip(preds, labels):
        conf_matrix[p, t] += 1
    return conf_matrix

conf_matrix=np.array(conf_matrix.cpu())# 將混淆矩陣從gpu轉到cpu再轉到np
corrects=conf_matrix.diagonal(offset=0)#抽取對角線的每種分類的識別正確個數
per_kinds=conf_matrix.sum(axis=1)#抽取每個分類數據總的測試條數

 print("混淆矩陣總元素個數：{0},測試集總個數:{1}".format(int(np.sum(conf_matrix)),test_num))
 print(conf_matrix)

 # 獲取每種Emotion的識別準確率
 print("每種情感總個數：",per_kinds)
 print("每種情感預測正確的個數：",corrects)
 print("每種情感的識別準確率為：{0}".format([rate*100 for rate in corrects/per_kinds]))

# 繪制混淆矩陣
Emotion=8#這個數值是具體的分類數，大家可以自行修改
labels = ['neutral', 'calm', 'happy', 'sad', 'angry', 'fearful', 'disgust', 'surprised']#每種類別的標簽

# 顯示數據
plt.imshow(conf_matrix, cmap=plt.cm.Blues)

# 在圖中標注數量/概率信息
thresh = conf_matrix.max() / 2	#數值顏色閾值，如果數值超過這個，就顏色加深。
for x in range(Emotion_kinds):
    for y in range(Emotion_kinds):
        # 注意這里的matrix[y, x]不是matrix[x, y]
        info = int(conf_matrix[y, x])
        plt.text(x, y, info,
                 verticalalignment='center',
                 horizontalalignment='center',
                 color="white" if info > thresh else "black")
                 
plt.tight_layout()#保證圖不重疊
plt.yticks(range(Emotion_kinds), labels)
plt.xticks(range(Emotion_kinds), labels,rotation=45)#X軸字體傾斜45°
plt.show()
plt.close()