SVM

import numpy as np
import os
from sklearn import svm
from PIL import Image

def load_data(path, index):
    path = path + str(index)
    counter = 0
    images = np.array([[]])

    for root,dirs,files in os.walk(path):
        for f in files:
            abs_path = os.path.join(root,f)
            image = Image.open(abs_path)
            image_array = np.array(image)
            # flatten to 1-D vector; image_f: 784
            image_f = image_array.flatten()
            # add the new flattened image to our images set for training
            if counter == 0:
                images = np.concatenate((images, np.array([image_f])),axis=1)
            else:
                images = np.concatenate((images, np.array([image_f])))
            # images' shape: (num_img, 784) # (1, num_img*784)
            counter += 1
        print(images.shape)

    return images, counter                            # images: [num_image, 784]


def SVM(test_file_name):
    train_images = np.array([[]])
    train_label = np.array([])

    for i in range(0,10):
        label = i
        images, counter = load_data("D:\\HW4\\dataset\\train\\",i)

        if i == 0:
            train_images = images
        else:
            train_images = np.concatenate((train_images, images))
            # train_images' shape: (total_train_img, 784)
        train_label_i = np.array([label for j in range(0, counter)])
        train_label = np.append(train_label, train_label_i)

        print("label %d has been loaded"%(i))

    test_images = np.array([[]])
    test_label = np.array([])
    for i in range(0,10):
        label = i
        images, counter = load_data(test_file_name,i)
        if i == 0:
            test_images = images
        else:
            test_images = np.concatenate((test_images, images))
            # test_images' shape: (total_test_img, 784)
        test_label_i = np.array([label for j in range(0, counter)])
        test_label = np.append(test_label, test_label_i)

    print("Data loaded successfully")
    print(train_images.shape)  # output: (60000, 784)
    print(test_images.shape)   # output: (10000, 784)

    counter = 0
    error_times = 0

    lsvc = svm.LinearSVC(C=0.001,max_iter=10000,dual=False)
    lsvc.fit(train_images, train_label)
    predict_label = lsvc.predict(test_images)

    for label in predict_label:
        print("Test "+str(counter+1)+": Predict: "+str(label)+" "+"True: "+str(test_label[counter]))
        if label != test_label[counter]:
            error_times += 1
        counter += 1

    return error_times/counter



error_rate = SVM("D:\\HW4\\dataset\\test\\")
print("The error rate is %f"%(error_rate))