Python数据分析实战之：特征重要性分析

提醒

• pandas 读取 excel 文件，需要 xlrd >= 1.1.0

代码

import sklearn
import pandas as pd
import numpy as np
import matplotlib.pyplot as plt
# 混淆矩阵
import seaborn as sns
from sklearn.linear_model import LogisticRegression
from sklearn.model_selection import train_test_split
from sklearn.metrics import confusion_matrix
from sklearn.model_selection import cross_val_score
from sklearn.metrics import roc_curve, f1_score, precision_score, recall_score
from sklearn.svm import SVC

## pandas 显示全部单元格

pd.set_option('display.max_rows', None)
pd.set_option('display.max_columns', None)
pd.set_option('display.width', None)
pd.set_option('display.max_colwidth', -1)

数据处理

path = "./2019-2020年.xlsx"

# df = pd.read_csv(path, error_bad_lines=False)
df = pd.read_excel(path)

# 因为 OTT, TOAST 空值太多，暂不用于分析
df = df.drop(['OTT', 'TOAST subtypes'],axis=1)


# 其余的缺失值比较少，直接删除空值
df = df.dropna(axis=0, how='any')

df = df.reset_index()

# 有些列中的数据是字符串，要转成 int 或者 float 才能训练
# 看哪些数据是 object 类型的，全部转成 int 或者 float64 型

# 只有这一个数据有问题，转成 0 即可
df['Coronary heart disease'][74] = 0

# 数据很干净
df.head(10)

分离 data 和 label

# 看分布
label_1 = df.columns[-3]
label_2 = df.columns[-1]

# df[label_1].hist()

# df[label_2].hist()

# 得到数据集 Data 是训练数据；label1_data 是第一种标签的标签数据; label2_data 是第二种标签的标签数据

label1_data = df[label_1]
label2_data = df[label_2]


data = df[[column for column in df.columns if column not in [label_1, label_2]]]
# 去除 index 这一列，这列数据没有意义
data.drop(['index'], axis=1, inplace=True)

data.drop(['90dmRS'], axis=1, inplace=True)

data.head(10)

训练

def train(model, dataset, labelset):
    x_train, x_test, y_train, y_test = train_test_split(dataset.values
                                                        , labelset.values
                                                        , test_size=0.2
                                                        , train_size=0.8
                                                        , shuffle=True
                                                        , stratify=labelset)
    model.fit(x_train, y_train)
    score = model.score(x_test, y_test)
    accs = cross_val_score(model, dataset.values, labelset.values, verbose=0)
    print(f'validation acc is: {score}')
    print(f'cross validation accs are: {accs}')
    
    y_pre = model.predict(x_test)
    metri = confusion_matrix(y_test, y_pre)
    sns.heatmap(metri, annot=True)
    plt.show()
    
'
运行
运行

训练结果 & 混淆矩阵

svc1 = SVC(class_weight='balanced', kernel='linear')
train(svc1, data, label1_data)

validation acc is: 0.8863636363636364
cross validation accs are: [0.90909091 0.69767442 0.76744186 0.88372093 0.88372093]

svc2 = SVC(class_weight='balanced', kernel='linear')
train(svc2, data, label2_data)

validation acc is: 0.8863636363636364
cross validation accs are: [0.81818182 0.6744186  0.72093023 0.76744186 0.76744186]

lr1 = LogisticRegression(class_weight='balanced', max_iter=10000)
train(lr1, data, label1_data)

validation acc is: 0.9090909090909091
cross validation accs are: [0.81818182 0.6744186  0.76744186 0.88372093 0.90697674]

lr2 = LogisticRegression(class_weight='balanced', max_iter=10000)
train(lr2, data, label2_data)

validation acc is: 0.8863636363636364
cross validation accs are: [0.79545455 0.69767442 0.74418605 0.6744186  0.81395349]

各种 feature 的重要性

def make_coef_dictNdf(data_columns, coef):
    name_influence_dic = {string: imp for string, imp in zip(data_columns, coef.squeeze())}
    name_influence_df = pd.DataFrame(data=name_influence_dic, index=['influence']).T
    return name_influence_dic, name_influence_df
## label2 结果各个特征的重要性
'
运行
运行

def write(filename, name_df_dic):
    writer = pd.ExcelWriter(filename)
    for k,v in name_df_dic.items():
        v.to_excel(writer, sheet_name=k)
    writer.save()
    writer.close()

# writer = pd.ExcelWriter("逻辑回归.xlsx")
# label1_df.to_excel(writer, sheet_name="label1")
# label2_df.to_excel(writer, sheet_name="label2")
# writer.save()
# writer.close()

def plot(figsize, name_influence_df_lst, img_label_lst, title):
    plt.figure(figsize=figsize)
    for i in range(len(name_influence_df_lst)):
        df = name_influence_df_lst[i]
        plt.bar(x=df.index, height=df['influence'],label=img_label_lst[i])
    plt.title(title)
    plt.legend()
    plt.xticks(rotation=90)
# plt.figure(figsize=(20,10))
# plt.bar(x=label1_df.index, height=label1_df['influence'],label='label1')
# plt.bar(x=label1_df.index, height=label2_df['influence'], label='label2')
# plt.legend()
# plt.xticks(rotation=90) # 旋转90度
'
运行
运行

lr1_dic, lr1_df = make_coef_dictNdf(data.columns, lr1.coef_)
lr2_dic, lr2_df = make_coef_dictNdf(data.columns, lr2.coef_)

svc1_dic, svc1_df = make_coef_dictNdf(data.columns, svc1.coef_)
svc2_dic, svc2_df = make_coef_dictNdf(data.columns, svc2.coef_)

write("逻辑回归.xlsx", {'label1': lr1_df, 'label2': lr2_df})

write("SVM.xlsx", {'label1': svc1_df, 'label2': svc2_df})

plot((20,10), [lr1_df, lr2_df], ['label1', 'label2'], 'lr')

plot((20,10), [svc1_df, svc2_df], ['label1', 'label2'], 'svm')