机器学习实战-第三章（MNIST）

最新推荐文章于 2022-12-11 00:09:13 发布

b05410505

最新推荐文章于 2022-12-11 00:09:13 发布

阅读量964

点赞数 1

文章标签：《机器学习实战——基于Scikit-Learning和Tensorfl

本文链接：https://blog.csdn.net/b05410505/article/details/90814539

版权

import pandas as pd  
df = pd.read_csv('F:/MNIST/mnist.csv', header=None)
df.head()

	0	...
0	5	...
1	0	...
2	4	...
3	1	...
4	9	...

5 rows × 785 columns

import numpy as np

df = df.reindex(np.random.permutation(df.index)) # 打乱顺序的方法
x = df.drop(0, axis=1)
y = df[0]
print('X的概况：', x.shape)
print('Y的概况：', y.shape)
x_train, x_test, y_train, y_test = x[:60000], x[60000:], y[:60000], y[60000:]

X的概况： (70000, 784)
Y的概况： (70000,)

"""
data.irow(0)   #取data的第一行
data.icol(0)   #取data的第一列
data.ix[1:2] #返回第2行的第三种方法，返回的是DataFrame，跟data[1:2]同
data[1:2]  #返回第2行，从0计，返回的是单行，通过有前后值的索引形式，
       #如果采用data[1]则报错

"""

print(y[36000])
%matplotlib inline
import matplotlib
import matplotlib.pyplot as plt
some_digit = x.loc[[36000]].values
some_digit_image = some_digit.reshape(28, 28)
plt.imshow(some_digit_image, cmap=matplotlib.cm.binary, interpolation='nearest')
plt.axis('off')
plt.show()

在这里插入图片描述

print(y_train.head())
y_test

41266    1
719      5
69105    0
58965    1
61484    1
Name: 0, dtype: int64





18193    9
8664     1
24326    3
59176    1
28477    5
40056    1
44939    5
32688    3
30398    3
1460     4
64379    4
66269    7
26930    0
33416    1
5443     1
61529    4
54202    5
64308    1
51409    7
10533    7
10417    2
54426    7
55812    4
65474    4
51762    5
8364     7
8601     9
69603    2
1043     7
28068    2
        ..
48235    4
30295    6
20432    2
12970    7
9458     0
65275    5
8466     4
62936    4
21244    5
3176     2
34733    4
2974     0
66701    0
12519    5
46747    1
34662    0
50322    5
39379    6
33823    0
60482    7
48927    9
7176     1
49476    5
33086    1
58856    8
1062     5
63524    6
65577    4
26155    9
55798    6
Name: 0, Length: 10000, dtype: int64

y_train_9 = (y_train==9)
y_test_9 = (y_test==9)


#print(y_test[[2797]])
#y_test_9

# 使用随机梯度下降（SGD）训练模型

from sklearn.linear_model import SGDClassifier

sgd_clf = SGDClassifier(random_state=42)
sgd_clf.fit(x_train, y_train_9)

SGDClassifier(alpha=0.0001, average=False, class_weight=None,
              early_stopping=False, epsilon=0.1, eta0=0.0, fit_intercept=True,
              l1_ratio=0.15, learning_rate='optimal', loss='hinge',
              max_iter=1000, n_iter_no_change=5, n_jobs=None, penalty='l2',
              power_t=0.5, random_state=42, shuffle=True, tol=0.001,
              validation_fraction=0.1, verbose=0, warm_start=False)

sgd_clf.predict(x.loc[[30069,42882,167,33531,11688,26571,16734,14324]])
# sgd_clf.predict([some_digit])
# 这种语法错误为：Found array with dim 3. Estimator expected <= 2.

array([ True,  True,  True,  True,  True,  True,  True,  True])

# 使用cross_val_score模块的3折交叉验证方法改进SGD模型
from sklearn.model_selection import cross_val_score   # 采用K-fold交叉验证发，3个折叠
cross_val_score(sgd_clf, x_train, y_train_9, cv=3, scoring='accuracy')

array([0.92680366, 0.9418    , 0.9439472 ])

# 使用混淆矩阵（cross_val_predict）方法获取预测的真OR假结果，以便计算精确率和召回率
from sklearn.model_selection import cross_val_predict  # 先返回每个折叠的预测，每个预测都是在训练期间从未见过的
y_train_pred = cross_val_predict(sgd_clf, x_train, y_train_9, cv=3)

y_train_pred

array([False, False, False, ..., False, False, False])

# 使用生成的y_train_pred（预测结果正确与否的列表）与真实值计算精确率和召回率
# 手工算法
from sklearn.metrics import confusion_matrix
print(confusion_matrix(y_train_9, y_train_pred))
"""
真负类TN  假正类FP
假负类FN  真正类TP
"""
jingdu = 3637 / (3637+2281)
zhaohuilv = 3637 / (3637+1468)
print('精度：', jingdu)
print('召回率：', zhaohuilv)

[[52614  1468]
 [ 2281  3637]]
精度： 0.6145657316661034
召回率： 0.7124387855044074

# 使用sklearn.metrics中的precision_score和recall_score自动计算精确率和召回率

from sklearn.metrics import precision_score,recall_score
precision_score(y_train_9, y_train_pred)

0.7124387855044074

recall_score(y_train_9, y_train_pred)

0.6145657316661034

# 使用sklearn.metrics中的f1_score模块，自动计算F1分数（2*精度*召回率 / (精度 + 召回率)）
from sklearn.metrics import f1_score
f1_score(y_train_9, y_train_pred)

0.6598929511022407

# 设定阈值来改变精度和召回率，主意参数：method='decision_function'
# method='decision_function'这个参数返回的是计算得出的分数。可以用来设定阈值

#from sklearn.model_selection import cross_val_predict  # 先返回每个折叠的预测，每个预测都是在训练期间从未见过的
#y_train_pred = cross_val_predict(sgd_clf, x_train, y_train_9, cv=3)

y_scores = cross_val_predict(sgd_clf, x_train, y_train_9, method='decision_function')
y_scores

D:\Anaconda3\lib\site-packages\sklearn\model_selection\_split.py:1978: FutureWarning: The default value of cv will change from 3 to 5 in version 0.22. Specify it explicitly to silence this warning.
  warnings.warn(CV_WARNING, FutureWarning)





array([-42620.62981929, -51693.91890067, -39176.69145394, ...,
       -22254.03321689, -53437.33278592, -33654.56993053])

# 画出精确率和召回率的图
from sklearn.metrics import precision_recall_curve
precisions, recalls, thresholds = precision_recall_curve(y_train_9, y_scores)

def plot_pre_recall_vs_thre(precisions, recalls, thresholds):
    plt.plot(thresholds, precisions[:-1], 'b--',label='Precision')
    plt.plot(thresholds, recalls[:-1],'g--', label='recall')
    plt.xlabel('Threshold')
    plt.legend(loc='upper left')
    plt.ylim([0,1])

plot_pre_recall_vs_thre(precisions, recalls, thresholds)
plt.show()

在这里插入图片描述

y_train_pred_90 = (y_scores > 10000)
precision_score(y_train_9, y_train_pred_90)

0.8898305084745762

recall_score(y_train_9, y_train_pred_90)

0.01774248056775938

# ROC曲线：特异度和灵敏度的关系

from sklearn.metrics import roc_curve
fpr, tpr, thresholds = roc_curve(y_train_9, y_scores)

def plot_roc_curve(fpr, tpr, label=None):
    plt.plot(fpr, tpr, linewidth=2, label=label)
    plt.plot([0,1], [0,1], 'k--')
    plt.axis([0,1,0,1])
    plt.xlabel('False Positive Rate')
    plt.ylabel('True Positive Rate')
    
plot_roc_curve(fpr, tpr)
plt.show()

在这里插入图片描述

# 训练一个随机森林，与SGD分类在ROC曲线上进行对比

from sklearn.ensemble import RandomForestClassifier
forest_clf = RandomForestClassifier(random_state=42)
y_probas_forest = cross_val_predict(forest_clf, x_train, y_train_9, cv=10, method='predict_proba')

D:\Anaconda3\lib\site-packages\sklearn\ensemble\forest.py:245: FutureWarning: The default value of n_estimators will change from 10 in version 0.20 to 100 in 0.22.
  "10 in version 0.20 to 100 in 0.22.", FutureWarning)
D:\Anaconda3\lib\site-packages\sklearn\ensemble\forest.py:245: FutureWarning: The default value of n_estimators will change from 10 in version 0.20 to 100 in 0.22.
  "10 in version 0.20 to 100 in 0.22.", FutureWarning)
D:\Anaconda3\lib\site-packages\sklearn\ensemble\forest.py:245: FutureWarning: The default value of n_estimators will change from 10 in version 0.20 to 100 in 0.22.
  "10 in version 0.20 to 100 in 0.22.", FutureWarning)
D:\Anaconda3\lib\site-packages\sklearn\ensemble\forest.py:245: FutureWarning: The default value of n_estimators will change from 10 in version 0.20 to 100 in 0.22.
  "10 in version 0.20 to 100 in 0.22.", FutureWarning)
D:\Anaconda3\lib\site-packages\sklearn\ensemble\forest.py:245: FutureWarning: The default value of n_estimators will change from 10 in version 0.20 to 100 in 0.22.
  "10 in version 0.20 to 100 in 0.22.", FutureWarning)
D:\Anaconda3\lib\site-packages\sklearn\ensemble\forest.py:245: FutureWarning: The default value of n_estimators will change from 10 in version 0.20 to 100 in 0.22.
  "10 in version 0.20 to 100 in 0.22.", FutureWarning)
D:\Anaconda3\lib\site-packages\sklearn\ensemble\forest.py:245: FutureWarning: The default value of n_estimators will change from 10 in version 0.20 to 100 in 0.22.
  "10 in version 0.20 to 100 in 0.22.", FutureWarning)
D:\Anaconda3\lib\site-packages\sklearn\ensemble\forest.py:245: FutureWarning: The default value of n_estimators will change from 10 in version 0.20 to 100 in 0.22.
  "10 in version 0.20 to 100 in 0.22.", FutureWarning)
D:\Anaconda3\lib\site-packages\sklearn\ensemble\forest.py:245: FutureWarning: The default value of n_estimators will change from 10 in version 0.20 to 100 in 0.22.
  "10 in version 0.20 to 100 in 0.22.", FutureWarning)
D:\Anaconda3\lib\site-packages\sklearn\ensemble\forest.py:245: FutureWarning: The default value of n_estimators will change from 10 in version 0.20 to 100 in 0.22.
  "10 in version 0.20 to 100 in 0.22.", FutureWarning)

# 随机森林中，使用参数method='predict_proba'，返回的是概率值
y_probas_forest

array([[1., 0.],
       [1., 0.],
       [1., 0.],
       ...,
       [1., 0.],
       [1., 0.],
       [1., 0.]])

y_scores_forest = y_probas_forest[:, 1]
fpr_forest, tpr_forest, thresholds_forest = roc_curve(y_train_9, y_scores_forest)

plt.plot(fpr, tpr, 'b', label='SGD')
plot_roc_curve(fpr_forest, tpr_forest, 'Random Forest')
plt.legend(loc='bottom right')
plt.show()

D:\Anaconda3\lib\site-packages\matplotlib\legend.py:497: UserWarning: Unrecognized location "bottom right". Falling back on "best"; valid locations are
	best
	upper right
	upper left
	lower left
	lower right
	right
	center left
	center right
	lower center
	upper center
	center

  % (loc, '\n\t'.join(self.codes)))

在这里插入图片描述

from sklearn.metrics import roc_auc_score

print(roc_auc_score(y_train_9, y_scores))
print(roc_auc_score(y_train_9, y_scores_forest))

0.9401824847125175
0.9883210685077506

# 多类别分类器
# OvA——OvO

# 先使用SGD对所有标签进行训练试试
sgd_clf.fit(x_train, y_train)

SGDClassifier(alpha=0.0001, average=False, class_weight=None,
              early_stopping=False, epsilon=0.1, eta0=0.0, fit_intercept=True,
              l1_ratio=0.15, learning_rate='optimal', loss='hinge',
              max_iter=1000, n_iter_no_change=5, n_jobs=None, penalty='l2',
              power_t=0.5, random_state=42, shuffle=True, tol=0.001,
              validation_fraction=0.1, verbose=0, warm_start=False)

sgd_clf.predict(x.loc[[36000]])

array([4], dtype=int64)

# 看看sklearn中二分类器是如何进行多类别分类的，需要看decision_function()分数
# 很明显，分类为9的分数为-2712，是最大的，所以分类为9，结果也是正确的。
some_digit_scores = sgd_clf.decision_function(x.loc[[36000]])
print(some_digit_scores)
print(np.argmax(some_digit_scores))  # np.argmax() 求最大的nampy方法
print(sgd_clf.classes_) # 查看原始结果方法

[[-39761.98821912 -22916.54645981 -18184.28472608   1016.42192678
    2300.61774756  -6673.18271756 -40432.41849423 -13508.94695046
   -2635.88432119  -2399.9387372 ]]
4
[0 1 2 3 4 5 6 7 8 9]

# 使用交叉验证方法评估SGD的准确率
cross_val_score(sgd_clf, x_train, y_train, cv=3, scoring='accuracy')

array([0.85907114, 0.87885   , 0.8819823 ])

# 使用多分类器“随机森林”算法再试试看

forest_clf.fit(x_train, y_train)

D:\Anaconda3\lib\site-packages\sklearn\ensemble\forest.py:245: FutureWarning: The default value of n_estimators will change from 10 in version 0.20 to 100 in 0.22.
  "10 in version 0.20 to 100 in 0.22.", FutureWarning)





RandomForestClassifier(bootstrap=True, class_weight=None, criterion='gini',
                       max_depth=None, max_features='auto', max_leaf_nodes=None,
                       min_impurity_decrease=0.0, min_impurity_split=None,
                       min_samples_leaf=1, min_samples_split=2,
                       min_weight_fraction_leaf=0.0, n_estimators=10,
                       n_jobs=None, oob_score=False, random_state=42, verbose=0,
                       warm_start=False)

# 返回概率值，发现结果是9的概率最大，所以返回结果是9
forest_clf.predict_proba(x.loc[[36000]])

array([[0., 0., 0., 0., 0., 0., 0., 0., 0., 1.]])

# 错误分析
# 使用cross_val_predict()函数进行预测，然后调用confusion_matrix()函数
y_train_pred_final = cross_val_predict(forest_clf, x_train, y_train, cv=3)
conf_mx = confusion_matrix(y_train, y_train_pred_final)
conf_mx

array([[5853,    2,   18,    6,    5,   15,   30,    2,   28,    2],
       [   1, 6618,   39,   19,   15,   10,    7,   11,   11,    5],
       [  54,   20, 5686,   39,   40,    7,   23,   57,   47,    5],
       [  36,   21,  147, 5646,    7,  119,   10,   52,   84,   36],
       [  18,   13,   34,   10, 5572,   10,   40,   19,   23,  117],
       [  50,   21,   31,  206,   35, 4948,   56,   12,   54,   30],
       [  57,   15,   26,    6,   31,   69, 5643,    1,   24,    2],
       [  13,   28,   96,   27,   82,    4,    2, 5861,   13,   79],
       [  29,   47,  104,  133,   61,  106,   44,   29, 5240,   78],
       [  33,   20,   31,   85,  193,   41,    6,   89,   65, 5355]],
      dtype=int64)

plt.matshow(conf_mx, cmap=plt.cm.gray)
plt.show()

在这里插入图片描述

row_sums = conf_mx.sum(axis=1, keepdims=True)
norm_conf_mx = conf_mx / row_sums
np.fill_diagonal(norm_conf_mx, 0)
plt.matshow(norm_conf_mx, cmap=plt.cm.gray)
plt.show()

在这里插入图片描述

# 多标签分类
# 希望分类器为每个样本产生多个类别（如一张照片里识别出多个人）
# 分类器可以输出多个二元标签。
# 这种分类叫做，多标签分类系统

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)

KNeighborsClassifier(algorithm='auto', leaf_size=30, metric='minkowski',
                     metric_params=None, n_jobs=None, n_neighbors=5, p=2,
                     weights='uniform')

knn_clf.predict(x.loc[[36000]])

array([[ True,  True]])

# 多标签分类的评估
y_train_knn_pred = cross_val_predict(knn_clf, x_train, y_train, cv=3)
f1_score(y_train, y_train_knn_pred, average='macro')

b05410505

关注

1
点赞
踩
5

收藏

觉得还不错? 一键收藏
3
评论
复制链接

分享到 QQ

分享到新浪微博

扫一扫

	0	...
0	5	...
1	0	...
2	4	...
3	1	...
4	9	...

	0	...
0	5	...
1	0	...
2	4	...
3	1	...
4	9	...

	0	...
0	5	...
1	0	...
2	4	...
3	1	...
4	9	...