作者：Abdullah Alrhmoun
该项目的目标是建立一个模型，该模型可以根据描述疾病的特征组合预测心脏病发生的概率。为了实现这一目标，作者使用了瑞士Cleveland Clinic Foundation收集的数据集。该项目中使用的数据集包含针对心脏病的14个特征。数据集显示不同水平的心脏病存在从1到4和0没有疾病。我们有303行人数据，13个连续观察不同的症状。此项目研究了不同的经典机器学习模型，以及它们在疾病风险中的发现。
导入依赖库
#导入依赖库 import pandas as pnd import numpy as np from sklearn import preprocessing from sklearn import neighbors, datasets from sklearn import cross_validation from sklearn.linear_model import SGDClassifier from sklearn import svm import operator from sklearn.cross_validation import KFold import itertools import numpy as np import matplotlib.pyplot as plt from sklearn import metrics from sklearn.metrics import confusion_matrix from sklearn import tree import seaborn as sns from IPython.display import Image %matplotlib inline
载入数据
# 添加列名 header_row = ['age','sex','chest_pain','blood pressure','serum_cholestoral','fasting_blood_sugar',\ 'electrocardiographic','max_heart_rate','induced_angina','ST_depression','slope','vessels','thal','diagnosis'] # 载入数据 heart = pnd.read_csv('processed.cleveland.data.csv', names=header_row) heart[:5]
#查看数据维度 heart.shape
(303, 14) #303行人的数据，13个连续观察不同症状。
数据探索
# 计算统计值 heart.describe()
计算某个特征的人数
names_descr = dict() categorical_columns = ["sex", "chest_pain", "fasting_blood_sugar", "electrocardiographic", "induced_angina", "slope", "vessels", \ "thal", "diagnosis"] for c in categorical_columns: print (heart.groupby([c])["age"].count())
使用简单的均值插补方法预处理数据，将缺失的数据更改为平均值
for c in heart.columns[:-1]: heart[c] = heart[c].apply(lambda x: heart[heart[c]!='?'][c].astype(float).mean() if x == "?" else x) heart[c] = heart[c].astype(float)
心脏病类型发现
set(heart.loc[:, "diagnosis"].values)
（0 ：没有疾病；1,2,3,4 代表不同疾病类型）
计算1,2,3,4 levels 之间的相似性
vecs_1 = heart[heart["diagnosis"] == 1 ].median().values[:-2] vecs_2 = heart[heart["diagnosis"] == 2 ].median().values[:-2] vecs_3 = heart[heart["diagnosis"] == 3 ].median().values[:-2] vecs_4 = heart[heart["diagnosis"] == 4 ].median().values[:-2]vecs_1 = heart[heart["diagnosis"] == 1 ].median().values[:-2] vecs_2 = heart[heart["diagnosis"] == 2 ].median().values[:-2] vecs_3 = heart[heart["diagnosis"] == 3 ].median().values[:-2] vecs_4 = heart[heart["diagnosis"] == 4 ].median().values[:-2]
print ("Similarity between type 1 and type 2 is ", np.linalg.norm(vecs_1-vecs_2)) print ("Similarity between type 1 and type 3 is ", np.linalg.norm(vecs_1-vecs_3)) print ("Similarity between type 1 and type 4 is ", np.linalg.norm(vecs_1-vecs_4)) print ("Similarity between type 2 and type 3 is ", np.linalg.norm(vecs_2-vecs_3)) print ("Similarity between type 2 and type 4 is ", np.linalg.norm(vecs_2-vecs_4)) print ("Similarity between type 3 and type 4 is ", np.linalg.norm(vecs_3-vecs_4))
sim = {"(1,2)": np.linalg.norm(vecs_1-vecs_2), \ "(1,3)": np.linalg.norm(vecs_1-vecs_3),\ "(1,4)": np.linalg.norm(vecs_1-vecs_4),\ "(2,3)": np.linalg.norm(vecs_2-vecs_3),\ "(2,4)": np.linalg.norm(vecs_2-vecs_4),\ "(3,4)": np.linalg.norm(vecs_3-vecs_4) }
# 根据相近值排序 sorted_sim = sorted(sim.items(), key=operator.itemgetter(1)) sorted_sim
可以分别使用每个特征的值来比较心脏病的类型
heart_d = heart[heart["diagnosis"] >= 1 ] heart_d[:5]
数据预处理
# if "diagnosis" == 0, 没有疾病 # if "diagnosis" >= 1, 有疾病 heart.loc[:, "diag_int"] = heart.loc[:, "diagnosis"].apply(lambda x: 1 if x >= 1 else 0)
#数据标准化 preprocessing.Normalizer().fit_transform(heart)
#划分数据集 heart_train, heart_test, goal_train, goal_test = cross_validation.train_test_split(heart.loc[:,'age':'thal'], \ heart.loc[:,'diag_int'], test_size=0.33, random_state=0)
#计算相关系数 corr = heart.corr() heart.corr()
#绘制热图 cmap = sns.diverging_palette(250, 10, n=3, as_cmap=True) def magnify(): return [dict(selector="th", props=[("font-size", "7pt")]), dict(selector="td", props=[('padding', "0em 0em")]), dict(selector="th:hover", props=[("font-size", "12pt")]), dict(selector="tr:hover td:hover", props=[('max-width', '200px'), ('font-size', '12pt')]) ] corr.style.background_gradient(cmap, axis=1)\ .set_properties(**{'max-width': '80px', 'font-size': '10pt'})\ .set_caption("Hover to magify")\ .set_precision(2)\ .set_table_styles(magnify())
探索可视化
#年龄与血压关系 import matplotlib.pyplot as plt %matplotlib inline plt.xlabel("age") plt.ylabel("blood pressure") # define title plt.title("Relationship between age and blood pressure") # plot plt.scatter(heart['age'], heart['blood pressure']) plt.show()
建立训练模型并评估LSS参数
# add parameters for grid search loss = ["hinge", "log"] penalty = ["l1", "l2"] alpha = [0.1, 0.05, 0.01] n_iter = [500, 1000] # build the models with different parameters and select the best combination for the highest Accuracy best_score = 0 best_param = (0,0,0,0) for l in loss: for p in penalty: for a in alpha: for n in n_iter: print("Parameters for model", (l,p,a,n)) lss = SGDClassifier(loss=l, penalty=p, alpha=a, n_iter=n) lss.fit(heart_train, goal_train) print("Linear regression SGD Cross-Validation scores:") scores = cross_validation.cross_val_score(lss, heart.loc[:,'age':'thal'], heart.loc[:,'diag_int'], cv=10) print (scores) print("Mean Linear regression SGD Cross-Validation score = ", np.mean(scores)) if np.mean(scores) > best_score: best_score = np.mean(scores) best_param = (l,p,a,n) print("The best parameters for model are ", best_param) print("The Cross-Validation score = ", best_score)
lss_best = SGDClassifier(alpha=0.05, fit_intercept=True, loss='log', n_iter=1000, penalty='l1') lss_best.fit(heart_train, goal_train) print("Linear regression SGD Test score:") print(lss_best.score(heart_test, goal_test))
模型验证
# Compute confusion matrix cnf_matrix = confusion_matrix(goal_test, lss_best.predict(heart_test)) np.set_printoptions(precision=2) # Plot non-normalized confusion matrix plt.figure() plot_confusion_matrix(cnf_matrix, classes=["Heart disease", "No heart disease"], title='Confusion matrix, without normalization') plt.show()
