部分依赖图（Partial Dependence Plots）以及实战-疾病引起原因解释

接上篇，特征重要性解释

特征重要性展示了每个特征发挥的作用情况，partial dependence plots可以展示一个特征怎样影响的了预测结果。
前提同样是应用在模型建立完成后进行使用，概述如下：

• 首先选中一个样本数据，此时想观察Ball Possession列对结果的影响。

• 保证其他特征列不变，改变当前观察列的值，例如选择40%,50%,60%（大小）分别进行预测，得到各自的结果。

• 对比结果就能知道当前列(Ball Possession)对结果的影响情况。

• 包： pdpbox

单特征观察

from matplotlib import pyplot as plt
from pdpbox import pdp, get_dataset, info_plots

pdp_goals = pdp.pdp_isolate(model=my_model, dataset=val_X, model_features=feature_names, feature='Goal Scored')

pdp.pdp_plot(pdp_goals, 'Goal Scored')
plt.show()

y轴表示预测结果的变化对比于基本模型，由于在观察时不可能只看一个样本数据，肯定要选择多个样本数据，蓝色区域表示的是置信度.

feature_to_plot = 'Distance Covered (Kms)'

rf_model = RandomForestClassifier(random_state=0).fit(train_X, train_y)

pdp_dist = pdp.pdp_isolate(model=rf_model, dataset=val_X, model_features=feature_names, feature=feature_to_plot)

pdp.pdp_plot(pdp_dist, feature_to_plot)
plt.show()

双特征观察

需要先改一下源码：将Anaconda3\Lib\site-packages\pdpbox 下pdp_plot_utils.py文件中contour_label_fontsize=fontsiz改成fontsize=fontsize

features_to_plot = ['Goal Scored', 'Distance Covered (Kms)']
inter1  =  pdp.pdp_interact(model=rf_model, dataset=val_X, model_features=feature_names, features=features_to_plot)

pdp.pdp_interact_plot(pdp_interact_out=inter1, feature_names=features_to_plot, plot_type='contour')
plt.show()

SHAP VALUES

可以直观的展示每一个特征对结果走势的影响
缺点：只传入一个样本点

row_to_show = 5
data_for_prediction = val_X.iloc[row_to_show] 
data_for_prediction_array = data_for_prediction.values.reshape(1, -1)


my_model.predict_proba(data_for_prediction_array)

导入shap工具包可能出现问题，在Anaconda3/lib/site-packages/numpy/lib/arraypad.py中添加下面两个函数保存，重新加载即可，保存成utf-8

import shap  

#实例化
explainer = shap.TreeExplainer(my_model)

#计算
shap_values = explainer.shap_values(data_for_prediction)

返回的SHAP values中包括了negative和positive两种情况，通常选择一种（positive）即可

shap.initjs()
shap.force_plot(explainer.expected_value[1], shap_values[1], data_for_prediction)

Summary Plots

explainer = shap.TreeExplainer(my_model)

shap_values = explainer.shap_values(val_X)

shap.summary_plot(shap_values[1], val_X)

实战-疾病引起原因模型解释

疾病引起原因模型解释

import numpy as np
import pandas as pd
import matplotlib.pyplot as plt
import seaborn as sns 
from sklearn.ensemble import RandomForestClassifier 
from sklearn.tree import DecisionTreeClassifier
from sklearn.tree import export_graphviz 
from sklearn.metrics import roc_curve, auc 
from sklearn.metrics import classification_report 
from sklearn.metrics import confusion_matrix 
from sklearn.model_selection import train_test_split 
import eli5 
from eli5.sklearn import PermutationImportance
import shap 
from pdpbox import pdp, info_plots
np.random.seed(123) 

pd.options.mode.chained_assignment = None  

dt = pd.read_csv("heart.csv")
dt.head()

• age：该人的年龄
• sex：该人的性别（1 =男性，0 =女性）
• cp：胸痛经历（值1：典型心绞痛，值2：非典型心绞痛，值3：非心绞痛，值4：无症状）
• trestbps：该人的静息血压（入院时为mm Hg）
• chol：人体胆固醇测量单位为mg / dl
• fbs：该人的空腹血糖（> 120 mg / dl，1 = true; 0 = false）
• restecg：静息心电图测量（0 =正常，1 =有ST-T波异常，2 =按Estes标准显示可能或明确的左心室肥厚）
• thalach：达到了该人的最大心率
• exang：运动诱发心绞痛（1 =是; 0 =否）
• oldpeak：运动相对于休息引起的ST段压低（'ST’与ECG图上的位置有关。点击此处查看更多内容）
• slope：峰值运动ST段的斜率（值1：上升，值2：平坦，值3：下降）
• ca：主要数量（0-3）
• thal：称为地中海贫血的血液疾病（3 =正常; 6 =固定缺陷; 7 =可逆缺陷）
• target：心脏病（0 =不，1 =是）

dt.columns = ['age', 'sex', 'chest_pain_type', 'resting_blood_pressure', 'cholesterol', 'fasting_blood_sugar', 'rest_ecg', 'max_heart_rate_achieved',
       'exercise_induced_angina', 'st_depression', 'st_slope', 'num_major_vessels', 'thalassemia', 'target']

dt['sex'][dt['sex'] == 0] = 'female'
dt['sex'][dt['sex'] == 1] = 'male'

dt['chest_pain_type'][dt['chest_pain_type'] == 1] = 'typical angina'
dt['chest_pain_type'][dt['chest_pain_type'] == 2] = 'atypical angina'
dt['chest_pain_type'][dt['chest_pain_type'] == 3] = 'non-anginal pain'
dt['chest_pain_type'][dt['chest_pain_type'] == 4] = 'asymptomatic'

dt['fasting_blood_sugar'][dt['fasting_blood_sugar'] == 0] = 'lower than 120mg/ml'
dt['fasting_blood_sugar'][dt['fasting_blood_sugar'] == 1] = 'greater than 120mg/ml'

dt['rest_ecg'][dt['rest_ecg'] == 0] = 'normal'
dt['rest_ecg'][dt['rest_ecg'] == 1] = 'ST-T wave abnormality'
dt['rest_ecg'][dt['rest_ecg'] == 2] = 'left ventricular hypertrophy'

dt['exercise_induced_angina'][dt['exercise_induced_angina'] == 0] = 'no'
dt['exercise_induced_angina'][dt['exercise_induced_angina'] == 1] = 'yes'

dt['st_slope'][dt['st_slope'] == 1] = 'upsloping'
dt['st_slope'][dt['st_slope'] == 2] = 'flat'
dt['st_slope'][dt['st_slope'] == 3] = 'downsloping'

dt['thalassemia'][dt['thalassemia'] == 1] = 'normal'
dt['thalassemia'][dt['thalassemia'] == 2] = 'fixed defect'
dt['thalassemia'][dt['thalassemia'] == 3] = 'reversable defect'

dt.dtypes

dt['sex'] = dt['sex'].astype('object')
dt['chest_pain_type'] = dt['chest_pain_type'].astype('object')
dt['fasting_blood_sugar'] = dt['fasting_blood_sugar'].astype('object')
dt['rest_ecg'] = dt['rest_ecg'].astype('object')
dt['exercise_induced_angina'] = dt['exercise_induced_angina'].astype('object')
dt['st_slope'] = dt['st_slope'].astype('object')
dt['thalassemia'] = dt['thalassemia'].astype('object')

dt = pd.get_dummies(dt, drop_first=True)
dt.head()

X_train, X_test, y_train, y_test = train_test_split(dt.drop('target', 1), dt['target'], test_size = .2, random_state=10) 

model = RandomForestClassifier(max_depth=5)
model.fit(X_train, y_train)

estimator = model.estimators_[1]
feature_names = [i for i in X_train.columns]

y_train_str = y_train.astype('str')
y_train_str[y_train_str == '0'] = 'no disease'
y_train_str[y_train_str == '1'] = 'disease'
y_train_str = y_train_str.values

export_graphviz(estimator, out_file='tree.dot', 
                feature_names = feature_names,
                class_names = y_train_str,
                rounded = True, proportion = True, 
                label='root',
                precision = 2, filled = True)


from subprocess import call
call(['dot', '-Tpng', 'tree.dot', '-o', 'tree.png', '-Gdpi=600'])

from IPython.display import Image
Image(filename = 'tree.png')

y_predict = model.predict(X_test)
y_pred_quant = model.predict_proba(X_test)[:, 1]
y_pred_bin = model.predict(X_test)

confusion_matrix = confusion_matrix(y_test, y_pred_bin)
confusion_matrix

fpr, tpr, thresholds = roc_curve(y_test, y_pred_quant)

fig, ax = plt.subplots()
ax.plot(fpr, tpr)
ax.plot([0, 1], [0, 1], transform=ax.transAxes, ls="--", c=".3")
plt.xlim([0.0, 1.0])
plt.ylim([0.0, 1.0])
plt.rcParams['font.size'] = 12
plt.title('ROC curve for diabetes classifier')
plt.xlabel('False Positive Rate (1 - Specificity)')
plt.ylabel('True Positive Rate (Sensitivity)')
plt.grid(True)
plt.show()

auc(fpr, tpr)

perm = PermutationImportance(model, random_state=1).fit(X_test, y_test)
eli5.show_weights(perm, feature_names = X_test.columns.tolist())

base_features = dt.columns.values.tolist()
base_features.remove('target')

feat_name = 'num_major_vessels'
pdp_dist = pdp.pdp_isolate(model=model, dataset=X_test, model_features=base_features, feature=feat_name)

pdp.pdp_plot(pdp_dist, feat_name)
plt.show()

feat_name = 'age'
pdp_dist = pdp.pdp_isolate(model=model, dataset=X_test, model_features=base_features, feature=feat_name)

pdp.pdp_plot(pdp_dist, feat_name)
plt.show()

inter1  =  pdp.pdp_interact(model=model, dataset=X_test, model_features=base_features, features=['st_slope_upsloping', 'st_depression'])

pdp.pdp_interact_plot(pdp_interact_out=inter1, feature_names=['st_slope_upsloping', 'st_depression'], plot_type='contour')
plt.show()

inter1  =  pdp.pdp_interact(model=model, dataset=X_test, model_features=base_features, features=['st_slope_flat', 'st_depression'])

pdp.pdp_interact_plot(pdp_interact_out=inter1, feature_names=['st_slope_flat', 'st_depression'], plot_type='contour')
plt.show()

explainer = shap.TreeExplainer(model)
shap_values = explainer.shap_values(X_test)
shap.summary_plot(shap_values[1], X_test)

def heart_disease_risk_factors(model, patient):

    explainer = shap.TreeExplainer(model)
    shap_values = explainer.shap_values(patient)
    shap.initjs()
    return shap.force_plot(explainer.expected_value[1], shap_values[1], patient)

data_for_prediction = X_test.iloc[1,:].astype(float)
heart_disease_risk_factors(model, data_for_prediction)