python：mnistデータで最適なkを探すプログラム

###前提・実現したいこと

MNIST手書き数字データセットを使って最適なkを探すプログラムを作成しているのですが、
現在の状況からどのように改善すればいいか教えてください。
また、ｋの使う近傍数ごとの正解率を折れ線グラフで可視化するプログラムも教えてください。

###発生している問題

データの容量が大きいこと

###該当のソースコード

python
1from sklearn import datasets, model_selection, svm, metrics
2
3mnist = datasets.fetch_mldata('MNIST original', data_home='data/src/download/')
4
5print(type(mnist))
6print(mnist.keys())
7
8mnist_data = mnist.data / 255
9mnist_label = mnist.target
10
11print(mnist_data.shape)
12print(mnist_label.shape)
13
14train_size = 500
15test_size = 100
16data_train, data_test, label_train, label_test = model_selection.train_test_split(mnist_data, mnist_label, test_size=test_size, train_size=train_size)
17
18clf = svm.SVC()
19clf.fit(data_train, label_train)
20pre = clf.predict(data_test)
21
22ac_score = metrics.accuracy_score(label_test, pre)
23print(ac_score)
24
25import timeit
26
27num = 10
28
29train_size = 500
30test_size = 100
31data_train, data_test, label_train, label_test = model_selection.train_test_split(mnist_data, mnist_label, test_size=test_size, train_size=train_size)
32
33clf = svm.SVC()
34print(timeit.timeit(lambda: clf.fit(data_train, label_train), number=num) / num)
35pre = clf.predict(data_test)
36ac_score = metrics.accuracy_score(label_test, pre)
37print(ac_score)
38
39
40clf = svm.LinearSVC()
41print(timeit.timeit(lambda: clf.fit(data_train, label_train), number=num) / num)
42pre = clf.predict(data_test)
43ac_score = metrics.accuracy_score(label_test, pre)
44print(ac_score)
45
46train_size = 10000
47test_size = 2000
48data_train, data_test, label_train, label_test = model_selection.train_test_split(mnist_data, mnist_label, test_size=test_size, train_size=train_size)
49
50clf = svm.LinearSVC()
51print(timeit.timeit(lambda: clf.fit(data_train, label_train), number=num) / num)
52pre = clf.predict(data_test)
53
54ac_score = metrics.accuracy_score(label_test, pre)
55print(ac_score)
56# 5.980748841702007
57# 0.8945
58
59co_mat = metrics.confusion_matrix(label_test, pre)
60print(co_mat)