机器学习实战第3章笔记

发表于 |
字数统计: 747

《机器学习实战》一书第3章笔记。

MNIST数据集

用sklearn下载流行数据集，数据保存在C:\Users\<username>\scikit_learn_data文件夹中，不想更改用户名，可以用%HOMEPATH%\scikit_learn_data直接访问。

1
2
3
4
from sklearn.datasets import fetch_openml
mnist = fetch_openml('mnist_784', version=1)
mnist.keys()
# dict_keys(['data', 'target', 'feature_names', 'DESCR', 'details', 'categories', 'url'])

二元分类器

SGDClassifier是用随机梯度下降法进行参数训练的线性分类器。

1
2
3
4
5
6
7
from sklearn.linear_model import SGDClassifier

sgd_clf = SGDClassifier(max_iter=1000, tol=1e-3, random_state=42)
sgd_clf.fit(X_train, y_train_5)

# 随机分类器
from sklearn.dummy import DummyClassifier

性能考核

交叉验证

1
2
3
4
5
6
7
8
9
10
11
12
13
14
15
16
17
18
19
20
from sklearn.model_selection import cross_val_score
cross_val_score(sgd_clf, X_train, y_train_5, cv=3, scoring="accuracy")

# 手动实现交叉验证
from sklearn.model_selection import StratifiedKFold
from sklearn.base import clone

skfolds = StratifiedKFold(n_splits=3, random_state=42)

for train_index, test_index in skfolds.split(X_train, y_train_5):
    clone_clf = clone(sgd_clf)
    X_train_folds = X_train[train_index]
    y_train_folds = y_train_5[train_index]
    X_test_fold = X_train[test_index]
    y_test_fold = y_train_5[test_index]

    clone_clf.fit(X_train_folds, y_train_folds)
    y_pred = clone_clf.predict(X_test_fold)
    n_correct = sum(y_pred == y_test_fold)
    print(n_correct / len(y_pred))

混淆矩阵

image

所有的Negetive和Positive都是指的预测，True和False指预测对错，对角线上的元素都是预测对的。

1
2
3
4
5
from sklearn.model_selection import cross_val_predict
from sklearn.metrics import confusion_matrix

y_train_pred = cross_val_predict(sgd_clf, X_train, y_train_5, cv=3)
confusion_matrix(y_train_5, y_train_pred)

精度和召回率

1
2
3
4
5
6
from sklearn.metrics import precision_score, recall_score
precision_score(y_train_5, y_train_pred)
recall_score(y_train_5, y_train_pred)

from sklearn.metrics import f1_score
f1_score(y_train_5, y_train_pred)

$F_1$公式如下：
$$
F_1 = \frac{2}{\frac{1}{Precison}+\frac{1}{Recall}} = \frac{2TP}{2TP+FN+FP}
$$

精度/召回率权衡(PR曲线)

1
2
3
4
5
y_scores = cross_val_predict(sgd_clf, X_train, y_train_5, cv=3,method="decision_function")
# precision_recall_curve需要传入预测分数，而不是预测结果
from sklearn.metrics import precision_recall_curve
precisions, recalls, thresholds = precision_recall_curve(y_train_5, y_scores)
plt.plot(recalls, precisions, "b-", linewidth=2)

ROC曲线

ROC曲线绘制的是真正类率(灵敏度，TPR, Recall)和假正类率(FPR, 1-Specificity)的关系。

1
2
3
4
5
6
from sklearn.metrics import roc_curve
fpr, tpr, thresholds = roc_curve(y_train_5, y_scores)

# ROC AUC
from sklearn.metrics import roc_auc_score
roc_auc_score(y_train_5, y_scores)

多类别分类器

多类别分类器可以由二元分类器扩展，OVR(one-versus-the-rest, or one-versus-all, OVA)或者OVO(one-versus-one)， sklearn会自动选择运行OVR还是OVO，多数时候都是OVR，SVM选择OVO。你也可以强制指定选用哪种方法。

1
2
3
4
from sklearn.multiclass import OneVsOneClassifier
ovo_clf = OneVsOneClassifier(sgd_clf)
ovo_clf.fit(X_train, y_train)
len(ovo_clf.estimators_) # 45

但是有些分类可以直接进行多分类，无需OVR和OVO。SGD就可以。

错误分析

通过混淆矩阵，分析某些错误实例。

多标签分类

结果输出多个标签，而不是一个。KNN支持多标签分类。

1
2
3
4
5
6
7
8
9
from sklearn.neighbors import KNeighborsClassifier

y_train_large = (y_train >= 7)
y_train_odd = (y_train % 2 == 1)
y_multilabel = np.c_[y_train_large, y_train_odd]

knn_clf = KNeighborsClassifier()
knn_clf.fit(X_train, y_multilabel)
knn_clf.predict([some_digit]) # array([[False,  True]])

多输出分类

多标签的泛化，标签可以多种类别。

1
2
3
4
5
6
7
8
9
# 图片降噪示例，输出是图片的向量
noise = np.random.randint(0, 100, (len(X_train), 784))
X_train_mod = X_train + noise
noise = np.random.randint(0, 100, (len(X_test), 784))
X_test_mod = X_test + noise
y_train_mod = X_train
y_test_mod = X_test
knn_clf.fit(X_train_mod, y_train_mod)
clean_digit = knn_clf.predict([X_test_mod[0]])

-------------本文结束感谢您的阅读-------------
0%