親自驗證，可以解決svm的模型載入問題:

    import numpy as np
    from sklearn import datasets
     
    X,y = datasets.make_classification(n_samples=100,n_features=2,n_redundant=0,n_classes=2,random_state=7816)
     
    print(X.shape,y.shape)
    X = X.astype(np.float32)
    y = y * 2 - 1
    '''分離資料'''
    from sklearn import model_selection as ms
    X_train, X_test, y_train, y_test = ms.train_test_split(
        X, y, test_size=0.2, random_state=42
    )
    import cv2
    svm = cv2.ml.SVM_create()
    svm.setKernel(cv2.ml.SVM_LINEAR)
    '''開始訓練'''
     
    y_train = y_train.reshape(-1, 1)
    # print(y_train)
    svm.train(X_train, cv2.ml.ROW_SAMPLE, y_train)
    svm.save("svmtest.mat")
    print ("Done\n")
     
    svm2 = cv2.ml.SVM_load("svmtest.mat")
     
    # svm2.load("svmtest.mat")
    # print(svm2)
    '''開始預測'''
    _, y_pred = svm2.predict(X_test)
     
    '''用scikit-learn的metrics模組計算準確率'''
    from sklearn import metrics
    print(metrics.accuracy_score(y_test, y_pred))

關鍵程式碼如下:

建立：

    import cv2
    svm = cv2.ml.SVM_create()
    svm.setKernel(cv2.ml.SVM_LINEAR)

其它的寫法都是以前較老的版本，基本上都不行

載入：

svm2 = cv2.ml.SVM_load("svmtest.mat")