[人工智能]Task05：模型建立和评估

Task05：模型搭建和评估

• 本文参考datawhale开源学习资料

# 导入本节相关数据包
import pandas as pd
import numpy as np
import seaborn as sns
import matplotlib.pyplot as plt
from IPython.display import Image

%matplotlib inline

plt.rcParams['font.sans-serif'] = ['SimHei']  # 用来正常显示中文标签
plt.rcParams['axes.unicode_minus'] = False  # 用来正常显示负号
plt.rcParams['figure.figsize'] = (10, 6)  # 设置输出图片大小

# 读取训练数集
train = pd.read_csv('train.csv')
train.shape

(891, 12)

train.head(2)

特征工程

任务一：缺失值填充

• 对分类变量缺失值：填充某个缺失值字符(NA)、用最多类别的进行填充
• 对连续变量缺失值：填充均值、中位数、众数

# 对分类变量进行填充
train['Cabin'] = train['Cabin'].fillna('NA')
train['Embarked'] = train['Embarked'].fillna('S')

# 对连续变量进行填充
train['Age'] = train['Age'].fillna(train['Age'].mean())

# 检查缺失值比例
train.isnull().sum().sort_values(ascending=False)

Embarked       0
Cabin          0
Fare           0
Ticket         0
Parch          0
SibSp          0
Age            0
Sex            0
Name           0
Pclass         0
Survived       0
PassengerId    0
dtype: int64

任务二：编码分类变量

# 取出所有的输入特征
data = train[['Pclass','Sex','Age','SibSp','Parch','Fare', 'Embarked']]

# 进行虚拟变量转换
data = pd.get_dummies(data)

• pandas.get_dummies()：利用pandas实现one_hot_encode的方式；
• 可以参考pandas.get_dummies 的用法

data.head(2)

模型搭建

• 处理完前面的数据我们就得到建模数据，下一步是选择合适模型
• 在进行模型选择之前我们需要先知道数据集最终是进行监督学习还是无监督学习
• 除了根据我们任务来选择模型外，还可以根据数据样本量以及特征的稀疏性来决定
• 刚开始我们总是先尝试使用一个基本的模型来作为其baseline，进而再训练其他模型做对比，最终选择泛化能力或性能比较好的模型

载入我们提供清洗之后的数据(clear_data.csv)，大家也将原始数据载入（train.csv），说说他们有什么不同

# 读取原数据数集
train = pd.read_csv('train.csv')
train.shape

(891, 12)

train.head(2)

#读取清洗过的数据集
data = pd.read_csv('clear_data.csv')

data.head(2)

• 'Name’和’Survived’一列被删去，'Name’对是否存活无影响
• 对分类变量进行了编码处理

思考0

• 数据集哪些差异会导致模型在拟合数据是发生变化

• 数据异常值
• 数据噪音
• 可以参考欠拟合和过拟合以及如何选择模型进行阅读

# sklearn模型算法选择路径图
Image('20170624105439491.png')

任务一：切割训练集和测试集

• 按比例切割训练集和测试集(一般测试集的比例有30%、25%、20%、15%和10%)
• 按目标变量分层进行等比切割
• 设置随机种子以便结果能复现

提示1

• 切割数据集是为了后续能评估模型泛化能力
• sklearn中切割数据集的方法为train_test_split
• 查看函数文档可以在jupyter noteboo里面使用train_test_split?后回车即可看到
• 分层和随机种子在参数里寻找

思考1

• 什么情况下切割数据集的时候不用进行随机选取

• 数据集是时序类型的，需要按照时间顺序来排列；
• 数据集需要使用分层拆分，等等。

from sklearn.model_selection import train_test_split

# 一般先取出X和y后再切割，有些情况会使用到未切割的，这时候X和y就可以用
X = data
y = train['Survived']

# 对数据集进行切割
X_train, X_test, y_train, y_test = train_test_split(X, y, stratify=y, random_state=0)

# 查看数据形状
X_train.shape, X_test.shape

((668, 11), (223, 11))

任务二：模型创建

• 创建基于线性模型的分类模型（逻辑回归）
• 创建基于树的分类模型（决策树、随机森林）
• 查看模型的参数，并更改参数值，观察模型变化

提示2

• 逻辑回归不是回归模型而是分类模型，不要与LinearRegression混淆
• 随机森林其实是决策树集成为了降低决策树过拟合的情况
• 线性模型所在的模块为sklearn.linear_model
• 树模型所在的模块为sklearn.ensemble

思考2

• 为什么线性模型可以进行分类任务，背后是怎么的数学关系
• 对于多分类问题，线性模型是怎么进行分类的

• 可以参考https://blog.csdn.net/robin_xu_shuai/article/details/77246481；
• 对于多分类而言，将多个二分类的逻辑回归组合，即可实现多分类。

from sklearn.linear_model import LogisticRegression
from sklearn.ensemble import RandomForestClassifier

# 默认参数逻辑回归模型
lr = LogisticRegression()
lr.fit(X_train, y_train)

E:\Coding\Anaconda3\envs\my-rdkit-env\lib\site-packages\sklearn\linear_model\_logistic.py:764: ConvergenceWarning: lbfgs failed to converge (status=1):
STOP: TOTAL NO. of ITERATIONS REACHED LIMIT.

Increase the number of iterations (max_iter) or scale the data as shown in:
    https://scikit-learn.org/stable/modules/preprocessing.html
Please also refer to the documentation for alternative solver options:
    https://scikit-learn.org/stable/modules/linear_model.html#logistic-regression
  extra_warning_msg=_LOGISTIC_SOLVER_CONVERGENCE_MSG)





LogisticRegression()

# 查看训练集和测试集score值
print("Training set score: {:.2f}".format(lr.score(X_train, y_train)))
print("Testing set score: {:.2f}".format(lr.score(X_test, y_test)))

Training set score: 0.80
Testing set score: 0.79

# 调整参数后的逻辑回归模型
lr2 = LogisticRegression(C=100)
lr2.fit(X_train, y_train)

E:\Coding\Anaconda3\envs\my-rdkit-env\lib\site-packages\sklearn\linear_model\_logistic.py:764: ConvergenceWarning: lbfgs failed to converge (status=1):
STOP: TOTAL NO. of ITERATIONS REACHED LIMIT.

Increase the number of iterations (max_iter) or scale the data as shown in:
    https://scikit-learn.org/stable/modules/preprocessing.html
Please also refer to the documentation for alternative solver options:
    https://scikit-learn.org/stable/modules/linear_model.html#logistic-regression
  extra_warning_msg=_LOGISTIC_SOLVER_CONVERGENCE_MSG)





LogisticRegression(C=100)

print("Training set score: {:.2f}".format(lr2.score(X_train, y_train)))
print("Testing set score: {:.2f}".format(lr2.score(X_test, y_test)))

Training set score: 0.79
Testing set score: 0.78

# 默认参数的随机森林分类模型
rfc = RandomForestClassifier()
rfc.fit(X_train, y_train)

RandomForestClassifier()

print("Training set score: {:.2f}".format(rfc.score(X_train, y_train)))
print("Testing set score: {:.2f}".format(rfc.score(X_test, y_test)))

Training set score: 1.00
Testing set score: 0.83

# 调整参数后的随机森林分类模型
# n_estimators：分类器数量，max_depth：决策树的深度
rfc2 = RandomForestClassifier(n_estimators=100, max_depth=5)
rfc2.fit(X_train, y_train)

RandomForestClassifier(max_depth=5)

print("Training set score: {:.2f}".format(rfc2.score(X_train, y_train)))
print("Testing set score: {:.2f}".format(rfc2.score(X_test, y_test)))

Training set score: 0.87
Testing set score: 0.81

任务三：输出模型预测结果

• 输出模型预测分类标签
• 输出不同分类标签的预测概率

提示3

• 一般监督模型在sklearn里面有个predict能输出预测标签，predict_proba则可以输出标签概率

# 预测标签
pred = lr.predict(X_train)

# 此时我们可以看到0和1的数组
pred[:10]

array([0, 1, 1, 1, 0, 0, 1, 0, 1, 1], dtype=int64)

# 预测标签概率
pred_proba = lr.predict_proba(X_train)

pred_proba[:10]

array([[0.60880184, 0.39119816],
       [0.17696991, 0.82303009],
       [0.40684332, 0.59315668],
       [0.18910076, 0.81089924],
       [0.87978247, 0.12021753],
       [0.91379826, 0.08620174],
       [0.13287649, 0.86712351],
       [0.90556364, 0.09443636],
       [0.05281642, 0.94718358],
       [0.10934627, 0.89065373]])

思考3

• 预测标签的概率对我们有什么帮助

• 用于模型其他指标的评估：如AUC
• 可以知道模型作出预测的"可信度"

模型评估

• 模型评估是为了知道模型的泛化能力。
• 交叉验证（cross-validation）是一种评估泛化性能的统计学方法，它比单次划分训练集和测试集的方法更加稳定、全面。
• 在交叉验证中，数据被多次划分，并且需要训练多个模型。
• 最常用的交叉验证是 k 折交叉验证（k-fold cross-validation），其中 k 是由用户指定的数字，通常取 5 或 10。
• 准确率（precision）度量的是被预测为正例的样本中有多少是真正的正例
• 召回率（recall）度量的是正类样本中有多少被预测为正类
• f-分数是准确率与召回率的调和平均

• 可以参考datawhale公众号的文章：搞懂机器学习的常用评价指标！

任务一：交叉验证

• 用10折交叉验证来评估逻辑回归模型
• 计算交叉验证精度的平均值

Image('Snipaste_2020-01-05_16-37-56.png')

提示4

• 交叉验证在sklearn中的模块为sklearn.model_selection

思考4

• k折越多的情况下会带来什么样的影响？

• k折最多n折，又称为留一法，是K折的另一种极端情况。
• 可以参考「交叉验证」到底如何选择K值？

from sklearn.model_selection import cross_val_score

lr = LogisticRegression(C=100)
scores = cross_val_score(lr, X_train, y_train, cv=10)

E:\Coding\Anaconda3\envs\my-rdkit-env\lib\site-packages\sklearn\linear_model\_logistic.py:764: ConvergenceWarning: lbfgs failed to converge (status=1):
STOP: TOTAL NO. of ITERATIONS REACHED LIMIT.

Increase the number of iterations (max_iter) or scale the data as shown in:
    https://scikit-learn.org/stable/modules/preprocessing.html
Please also refer to the documentation for alternative solver options:
    https://scikit-learn.org/stable/modules/linear_model.html#logistic-regression
  extra_warning_msg=_LOGISTIC_SOLVER_CONVERGENCE_MSG)
E:\Coding\Anaconda3\envs\my-rdkit-env\lib\site-packages\sklearn\linear_model\_logistic.py:764: ConvergenceWarning: lbfgs failed to converge (status=1):
STOP: TOTAL NO. of ITERATIONS REACHED LIMIT.

Increase the number of iterations (max_iter) or scale the data as shown in:
    https://scikit-learn.org/stable/modules/preprocessing.html
Please also refer to the documentation for alternative solver options:
    https://scikit-learn.org/stable/modules/linear_model.html#logistic-regression
  extra_warning_msg=_LOGISTIC_SOLVER_CONVERGENCE_MSG)
E:\Coding\Anaconda3\envs\my-rdkit-env\lib\site-packages\sklearn\linear_model\_logistic.py:764: ConvergenceWarning: lbfgs failed to converge (status=1):
STOP: TOTAL NO. of ITERATIONS REACHED LIMIT.

Increase the number of iterations (max_iter) or scale the data as shown in:
    https://scikit-learn.org/stable/modules/preprocessing.html
Please also refer to the documentation for alternative solver options:
    https://scikit-learn.org/stable/modules/linear_model.html#logistic-regression
  extra_warning_msg=_LOGISTIC_SOLVER_CONVERGENCE_MSG)
E:\Coding\Anaconda3\envs\my-rdkit-env\lib\site-packages\sklearn\linear_model\_logistic.py:764: ConvergenceWarning: lbfgs failed to converge (status=1):
STOP: TOTAL NO. of ITERATIONS REACHED LIMIT.

Increase the number of iterations (max_iter) or scale the data as shown in:
    https://scikit-learn.org/stable/modules/preprocessing.html
Please also refer to the documentation for alternative solver options:
    https://scikit-learn.org/stable/modules/linear_model.html#logistic-regression
  extra_warning_msg=_LOGISTIC_SOLVER_CONVERGENCE_MSG)
E:\Coding\Anaconda3\envs\my-rdkit-env\lib\site-packages\sklearn\linear_model\_logistic.py:764: ConvergenceWarning: lbfgs failed to converge (status=1):
STOP: TOTAL NO. of ITERATIONS REACHED LIMIT.

Increase the number of iterations (max_iter) or scale the data as shown in:
    https://scikit-learn.org/stable/modules/preprocessing.html
Please also refer to the documentation for alternative solver options:
    https://scikit-learn.org/stable/modules/linear_model.html#logistic-regression
  extra_warning_msg=_LOGISTIC_SOLVER_CONVERGENCE_MSG)
E:\Coding\Anaconda3\envs\my-rdkit-env\lib\site-packages\sklearn\linear_model\_logistic.py:764: ConvergenceWarning: lbfgs failed to converge (status=1):
STOP: TOTAL NO. of ITERATIONS REACHED LIMIT.

Increase the number of iterations (max_iter) or scale the data as shown in:
    https://scikit-learn.org/stable/modules/preprocessing.html
Please also refer to the documentation for alternative solver options:
    https://scikit-learn.org/stable/modules/linear_model.html#logistic-regression
  extra_warning_msg=_LOGISTIC_SOLVER_CONVERGENCE_MSG)
E:\Coding\Anaconda3\envs\my-rdkit-env\lib\site-packages\sklearn\linear_model\_logistic.py:764: ConvergenceWarning: lbfgs failed to converge (status=1):
STOP: TOTAL NO. of ITERATIONS REACHED LIMIT.

Increase the number of iterations (max_iter) or scale the data as shown in:
    https://scikit-learn.org/stable/modules/preprocessing.html
Please also refer to the documentation for alternative solver options:
    https://scikit-learn.org/stable/modules/linear_model.html#logistic-regression
  extra_warning_msg=_LOGISTIC_SOLVER_CONVERGENCE_MSG)
E:\Coding\Anaconda3\envs\my-rdkit-env\lib\site-packages\sklearn\linear_model\_logistic.py:764: ConvergenceWarning: lbfgs failed to converge (status=1):
STOP: TOTAL NO. of ITERATIONS REACHED LIMIT.

Increase the number of iterations (max_iter) or scale the data as shown in:
    https://scikit-learn.org/stable/modules/preprocessing.html
Please also refer to the documentation for alternative solver options:
    https://scikit-learn.org/stable/modules/linear_model.html#logistic-regression
  extra_warning_msg=_LOGISTIC_SOLVER_CONVERGENCE_MSG)
E:\Coding\Anaconda3\envs\my-rdkit-env\lib\site-packages\sklearn\linear_model\_logistic.py:764: ConvergenceWarning: lbfgs failed to converge (status=1):
STOP: TOTAL NO. of ITERATIONS REACHED LIMIT.

Increase the number of iterations (max_iter) or scale the data as shown in:
    https://scikit-learn.org/stable/modules/preprocessing.html
Please also refer to the documentation for alternative solver options:
    https://scikit-learn.org/stable/modules/linear_model.html#logistic-regression
  extra_warning_msg=_LOGISTIC_SOLVER_CONVERGENCE_MSG)
E:\Coding\Anaconda3\envs\my-rdkit-env\lib\site-packages\sklearn\linear_model\_logistic.py:764: ConvergenceWarning: lbfgs failed to converge (status=1):
STOP: TOTAL NO. of ITERATIONS REACHED LIMIT.

Increase the number of iterations (max_iter) or scale the data as shown in:
    https://scikit-learn.org/stable/modules/preprocessing.html
Please also refer to the documentation for alternative solver options:
    https://scikit-learn.org/stable/modules/linear_model.html#logistic-regression
  extra_warning_msg=_LOGISTIC_SOLVER_CONVERGENCE_MSG)

# k折交叉验证分数
scores

array([0.82089552, 0.74626866, 0.74626866, 0.79104478, 0.86567164,
       0.8358209 , 0.76119403, 0.80597015, 0.74242424, 0.75757576])

# 平均交叉验证分数
print("Average cross-validation score: {:.2f}".format(scores.mean()))

Average cross-validation score: 0.79

任务二：混淆矩阵

• 计算二分类问题的混淆矩阵
• 计算精确率、召回率以及f-分数

Image('Snipaste_2020-01-05_16-38-26.png')

Image('Snipaste_2020-01-05_16-39-27.png')

提示5

• 混淆矩阵的方法在sklearn中的sklearn.metrics模块
• 混淆矩阵需要输入真实标签和预测标签

思考5

• 如果自己实现混淆矩阵的时候该注意什么问题

• 计算混淆矩阵需要输入真实标签和预测标签
• 灵敏度、特异度的概念
• 查准率、查全率的概念

from sklearn.metrics import confusion_matrix

# 训练模型
lr = LogisticRegression(C=100)
lr.fit(X_train, y_train)

E:\Coding\Anaconda3\envs\my-rdkit-env\lib\site-packages\sklearn\linear_model\_logistic.py:764: ConvergenceWarning: lbfgs failed to converge (status=1):
STOP: TOTAL NO. of ITERATIONS REACHED LIMIT.

Increase the number of iterations (max_iter) or scale the data as shown in:
    https://scikit-learn.org/stable/modules/preprocessing.html
Please also refer to the documentation for alternative solver options:
    https://scikit-learn.org/stable/modules/linear_model.html#logistic-regression
  extra_warning_msg=_LOGISTIC_SOLVER_CONVERGENCE_MSG)





LogisticRegression(C=100)

# 模型预测结果
pred = lr.predict(X_train)

# 混淆矩阵
confusion_matrix(y_train, pred)

array([[354,  58],
       [ 83, 173]], dtype=int64)

from sklearn.metrics import classification_report

# 精确率、召回率以及f1-score
print(classification_report(y_train, pred))

              precision    recall  f1-score   support

           0       0.81      0.86      0.83       412
           1       0.75      0.68      0.71       256

    accuracy                           0.79       668
   macro avg       0.78      0.77      0.77       668
weighted avg       0.79      0.79      0.79       668

任务三：ROC曲线

• 绘制ROC曲线

提示6

• ROC曲线在sklearn中的模块为sklearn.metrics
• ROC曲线下面所包围的面积越大越好

思考6

• 对于多分类问题如何绘制ROC曲线

• 可以参考多分类ROC曲线及AUC计算
• ①方法一：每种类别下，都可以得到m个测试样本为该类别的概率（矩阵P中的列）。所以，根据概率矩阵P和标签矩阵L中对应的每一列，可以计算出各个阈值下的假正例率（FPR）和真正例率（TPR），从而绘制出一条ROC曲线。这样总共可以绘制出n条ROC曲线。最后对n条ROC曲线取平均，即可得到最终的ROC曲线。
• ②方法二：首先，对于一个测试样本：1）标签只由0和1组成，1的位置表明了它的类别（可对应二分类问题中的‘’正’’），0就表示其他类别（‘’负‘’）；2）要是分类器对该测试样本分类正确，则该样本标签中1对应的位置在概率矩阵P中的值是大于0对应的位置的概率值的。基于这两点，将标签矩阵L和概率矩阵P分别按行展开，转置后形成两列，这就得到了一个二分类的结果。所以，此方法经过计算后可以直接得到最终的ROC曲线。

from sklearn.metrics import roc_curve

fpr, tpr, thresholds = roc_curve(y_test, lr.decision_function(X_test))
plt.plot(fpr, tpr, label="ROC Curve")
plt.xlabel("FPR")
plt.ylabel("TPR (recall)")
# 找到最接近于0的阈值
close_zero = np.argmin(np.abs(thresholds))
plt.plot(fpr[close_zero], tpr[close_zero], 'o', markersize=10, label="threshold zero", fillstyle="none", c='k', mew=2)
plt.legend(loc=4)

<matplotlib.legend.Legend at 0x1969e573d30>