机器学习算法模板
包含了KNN，线性回归，逻辑回归，朴素贝叶斯，决策树，支持向量机，随机森林，kmeans，集成算法
各种算法，特征工程，评估方式任你选择！！！
# 导包
from sklearn.neighbors import KNeighborsClassifier
from sklearn.linear_model import LinearRegression
from sklearn.naive_bayes import GaussianNB
from sklearn.naive_bayes import MultinomialNB
from sklearn.linear_model import LogisticRegression
from sklearn.preprocessing import PolynomialFeatures
from sklearn.datasets import make_blobs
from sklearn.cluster import KMeans
from sklearn.svm import SVC
from sklearn.tree import DecisionTreeClassifier
from sklearn.ensemble import RandomForestClassifier
from sklearn.ensemble import BaggingClassifier
from sklearn.ensemble import AdaBoostClassifier
from imblearn.over_sampling import SMOTE
from imblearn.under_sampling import RandomUnderSampler
from sklearn.feature_extraction.text import TfidfVectorizer
from sklearn.metrics import silhouette_score, silhouette_samples
import graphviz
from sklearn.tree import export_graphviz
from sklearn.model_selection import train_test_split
from sklearn.model_selection import cross_val_score
from sklearn.model_selection import GridSearchCV
import sklearn.datasets as datasets
from sklearn.linear_model import Ridge
from sklearn.preprocessing import MinMaxScaler
from sklearn.preprocessing import StandardScaler
from sklearn.metrics import mean_squared_error as MSE
from sklearn.metrics import r2_score as R2
from sklearn.metrics import accuracy_score
from sklearn.metrics import recall_score
from sklearn.metrics import f1_score
from sklearn.metrics import roc_auc_score
import pandas as pd
import numpy as np
import matplotlib.pyplot as plt
# #获取数据
# data = []
# data = datasets.load_iris()
# data = datasets.load_boston() # 小规模数据集
# data = datasets.load_wine()
# data = datasets.load_breast_cancer()
# data = datasets.fetch_california_housing(data_home="./datasets") # 大规模数据集
# data = pd.read_csv("")
# data = pd.read_excel("") # 从本地读取文件
# 提取特征值, 查看特征确保特征是二维
# feature = data["data"]
# target = data["target"]
# print(feature.shape, target.shape)
# 如果不是二维，则增加维度
# feature = feature.reshape(())
# 特征工程：
# 1.特征值化：map映射，onehot编码；
# 1.1 onehot编码
# r1 = pd.get_dummies(df['color']，dtype = int)
# new_df = pd.concat((df,r1),axis=1).drop(targets='color',axis=1) # 行合并，去列
# 1.2 map映射
# new_df['new_size'] = new_df['size'].map({'M':1,'L':2,'XL':3}) # map映射方法后重命名
# new_df = new_df.drop(targets='size',axis=1) # 行合并，去列
# 2.无量纲化：归一化，标准化
# 2.1归一化
# mm = MinMaxScaler() # 类实例化
# feature = mm.fit_transform(feature) # 训练
# 2.2标准化
# ss = StandardScaler() # 类实例化
# feature = ss.fit_transform(feature) # 训练
# 3.特征选择：过滤，降维
# 3.1 特征选择和过滤
# 文件读取 特征值选择
# data = pd.read_csv('cancer.csv',sep='\t')
# fea_col_index = data.columns[data.columns != 'Diagnosis']
# feature = data[fea_col_index].drop(targets='ID',axis=1)
#
# #threshold：方差的阈值 实例化
# v = VarianceThreshold(threshold=0.2)
# result = v.fit_transform(feature)
# print(result.shape)
#
# # 特征方差中位数作为阈值
# import numpy as np
# median_value = np.median(feature.var(axis=0).values)
# v = VarianceThreshold(threshold=median_value)
# result = v.fit_transform(feature)
# print(result.shape)
# 3.2
# PCA 降维
# 将数据分解为较低维度的空间
# n_components可以为小数（保留特征的百分比），整数（减少到的特征数量）
# pca = PCA(n_components=2)
# res = pca.fit_transform([[0,2,4,3],[0,3,7,3],[0,9,6,3]])
# print(res) #将（3，4）变成（3，2）
# 4.多项式回归(欠拟合增加特征数量）
# p2 = PolynomialFeatures(degree = 2, include_bias = False)
# feature_2 = p2.fit_transform(feature)
# 5.岭回归(过拟合减少高次特征权重)
# r = Ridge(alpha = 9).
# r.fit(feature, target)
# print(r.coef_)
# 6.词频统计，中文文本划分
# 6.1多项式词频统计，基于TF-TDF特征值化(英文文本)
# tf = TfidfVectorizer()
# feature = tf.fit_transform(feature)
# #
# x_train, x_test, y_train, y_test = train_test_split(
# feature, target, test_size=0.1, random_state=2021
# )
# 6.2 如果是中文文本，就要先用jieba对中文词语拆分
# text = [
# '因为在自然语言处理中，我们是需要将一段中文文本中相关的词语，成语，形容词......都要进行抽取的',
# '目前CountVectorizer只可以对有标点符号和用分隔符对应的文本进行特征抽取，显然这是满足不了我们日常需求的'
# ]
# new_text = []
# for t in text:
# r = list(jieba.cut(t)) # 返回列表
# s = ' '.join(r) # 返回字符串
# new_text.append(s) # 返回一维列表
# print(result.toarray()) # 返回数组
# 7.分箱
# 7.1等距分箱：cut
# # 连续性变量
# ages = np.random.randint(1, 80, size=(25,))
#
# # 使用分箱对连续性变量进行离散化
# # bins：分箱的个数
# cut_result = pd.cut(ages, bins=5, labels=list("abcde"))
# print(cut_result)
# print(cut_result.value_counts())
#
# # 7.2等频分箱qcut
# # 使用分箱对连续性变量进行离散化
# # q：分箱的个数
# qcut_result = pd.qcut(ages, q=5, labels=list("abcde"))
# print(qcut_result)
# print(qcut_result.value_counts())
# 8.样本类别分布不均衡处理
# 8.1使用smote合成少数过采样技术
# s = SMOTE(k_neighbors=3)
# feature, target = s.fit_resample(df[[0, 1, 2]], df[3])
# 训练完成之后返回特征和标签
# print(feature.shape, target.shape)
# 8.2欠采样技术
# from imblearn.under_sampling import RandomUnderSampler
# 划分数据集
# x_train, x_test, y_train, y_test = train_test_split(
# feature, target, test_size=0.2, random_state=2021
# )
# SVM: 调参优化：
# parameters = {'splitter':('best','random')
# ,'criterion':("gini","entropy")
# ,"max_depth":[*range(1,10)]
# ,'min_samples_leaf':[*range(1,50,5)]
# }
# 模型训练，查看精度
# 0.已保存模型的加载
# 0.1 方式1
# import joblib
# joblib.load('xxx.model') # 模型加载
# joblib.dump('') # 模型保存 写在模型训练完成之后
# 0.2 方式2
# import pickle
# with open('../data/mm.pkl', 'rb') as f:
# mm = pickle.load(f) # 模型加载
# with open('../data/mm.pkl', 'wb') as f:
# pickle.dump(mm, f) # 模型保存 写在模型训练完成之后
# 1.分类模型训练
# 1.1KNN分类模型
# knn = KNeighborsClassifier(n_neighbors=75).fit(x_train, y_train)
# 训练完毕
# 1.2寻找最优k： 学习曲线&交叉验证
# ks = np.arange(3, 50)
# scores = []
# for k in ks:
# knn = KNeighborsClassifier(n_neighbors=k)
# score = cross_val_score(knn, x_train, y_train, cv = 5).mean()
# scores.append(score)
# plt.plot(ks, scores, label="寻找最优K")
# plt.xlabel("k")
# plt.ylabel("score")
# # plt.show()
# best_k = ks[np.argmax(scores)]
# print(f"最优k是：{best_k}; 交叉验证最优精度：{max(scores)}")
# # 使用最优k训练模型
# knn = KNeighborsClassifier(n_neighbors=best_k).fit(x_train, y_train)
# # 1.1.1分类模型精度
# print(f"模型的分类结果{knn.predict(x_test)}")
# print(f"真实的分类结果{y_test}")
#
# # 分类模型评价指标
# y_pred = knn.predict(x_test)
# print(f"分类模型准确率：{knn.score(x_test, y_test)}")
# print(f"分类模型准确率：{accuracy_score(y_test, y_pred)}")
# print(f"分类模型召回率：{recall_score(y_test, y_pred, average = 'micro')}")
# print(f"分类模型精确率：{precision_score(y_test, y_pred, average = 'macro')}")
# print(f"分类模型精确率和召回率的调和平均数：{f1_score(y_test, y_pred, average = 'weighted')}")
# y_score = lre3.predict_proba(x_test)
# print(f"分类模型AUC曲线：{roc_auc_score(y_test, y_score, multi_class = 'ovr')}")
# # 1.2.1 逻辑回归（分类模型）
# # 1.2.1逻辑回归模型
# # 加入惩罚项（正则化方式），正则化力度（C），和求解器求解方式（坐标下降法）
# lre1 = LogisticRegression(penalty="l1", C=0.7, solver="liblinear").fit(x_train, y_train)
# lre2 = LogisticRegression(penalty="l2", C=0.7, solver="liblinear").fit(x_train, y_train)
# lre3 = LogisticRegression().fit(x_train, y_train)
# # 训练完毕
# y_pred = lre3.predict(x_test)
# print(f"分类模型准确率：{lre3.score(x_test, y_test)}")
# print(f"分类模型准确率：{accuracy_score(y_test, y_pred)}")
# print(f"分类模型召回率：{recall_score(y_test, y_pred, average = 'micro')}")
# print(f"分类模型精确率：{precision_score(y_test, y_pred, average = 'macro')}")
# print(f"分类模型精确率和召回率的调和平均数：{f1_score(y_test, y_pred, average = 'weighted')}")
# y_score = lre3.predict_proba(x_test)
# print(f"分类模型AUC曲线：{roc_auc_score(y_test, y_score, multi_class = 'ovr')}")
# 1.3.0 分类回归模型 : SVM
# 选择不同的核函数，计算最优超平面
# s1 = SVC(kernel='rbf').fit(x_train, y_train)
# s2 = SVC(kernel='poly').fit(x_train, y_train)
# s3 = SVC(kernel='linear').fit(x_train, y_train)
# s4 = SVC(kernel='sigmoid').fit(x_train, y_train)
# print('rbf:', f1_score(y_test, s1.predict(x_test)))
# print('ploy:', f1_score(y_test, s2.predict(x_test)))
# print('liner:', f1_score(y_test, s3.predict(x_test)))
# print('sigmoid:', f1_score(y_test, s4.predict(x_test)))
# # 1.4.0 分类模型： 决策树
# tree = DecisionTreeClassifier(random_state=2021).fit(x_train, y_train)
# GS = GridSearchCV(tree, parameters, cv=5)
# GS.fit(x_train, y_train)
# # 训练完毕
#
# # 模型评估
# print(tree.score(x_test, y_test))
# print(f"参数调优的最优参数：{GS.best_params_}")
# print(f"参数调优的最优精度：{GS.best_score_}")
#
#
# # 使用graphviz画出决策树，决策树可视化：
# feature_name = ['酒精', '苹果酸', '灰', '灰的碱性', '镁', '总酚', '类黄酮', '非黄烷类酚类', '花青素', '颜色强度', '色调', 'od280/od315稀释葡萄酒', '脯氨酸']
# dot_data = export_graphviz(tree, out_file=None # 图片保存路径
# , feature_names=feature_name, class_names=["琴酒", "雪莉", "贝尔摩德"], filled=True, fontname='SimSun' # 使用颜色表示分类结果
# )
# graph = graphviz.Source(dot_data, format="png")
# graph.view()
#
# print(tree.feature_importances_, tree) # 每一个维度特征的重要性
#
# # 返回特征的重要性
# feature_name = ['酒精', '苹果酸', '灰', '灰的碱性', '镁', '总酚', '类黄酮', '非黄烷类酚类', '花青素', '颜 色强度', '色调', 'od280/od315稀释葡萄酒', '脯氨酸']
# print("特征的重要程度")
# print([*zip(feature_name, tree.feature_importances_)])
# # 1.5 分类模型： 随机森林集成算法/加入决策树算法比较
# rfc = RandomForestClassifier(n_estimators = 10, random_state=2023).fit(x_train, y_train)
# dtc = DecisionTreeClassifier(random_state=2023).fit(x_train, y_train)
# # rfc = RandomForestClassifier().fit(x_train, y_train)
# # 训练结束
#
# # 交叉验证
#
# # 模型精度评估
# print(f"随机森林模型精度：{rfc.score(x_test, y_test)}")
# print(f"随机森林模型精度：{dtc.score(x_test, y_test)}")
# 集成算法：
# Bagging（装袋）： 通过构建多个相互独立的基本模型，
# 然后将它们的预测结果进行平均或投票来提高整体模型的性能。
# Bagging 适用于高方差的模型，如决策树等，能够降低过拟合的风险。
#
# Boosting（提升）： 通过串行地构建多个基本模型，
# 每个模型都在前一个模型的误差上进行学习，逐步提高整体模型的性能。
# Boosting 适用于低偏差的模型，如弱分类器或弱回归器，能够提升模型的预测能力。
#
# Stacking（堆叠）： 通过将不同基本模型的预测结果作为特征，
# 再训练一个元模型来进行最终的预测。
# Stacking 适用于需要更高级别整合多个模型时，能够获得更好的泛化能力。
# # bagging (RandomForest)
# rfc_bagging = RandomForestClassifier(n_estimators = 10, random_state = 2023).fit(x_train, y_train)
# print(f"bagging装袋法模型精度：{rfc_bagging.score(x_test, y_test)}")
#
# # boosting(adaboost)
# base_classifier = DecisionTreeClassifier(max_depth=100, random_state=2023)
# ada_boosting = AdaBoostClassifier(base_estimator=base_classifier, n_estimators=10, random_state=2023).fit(x_train, y_train)
# print(f"AdaBoost提升法模型精度：{ada_boosting.score(x_test, y_test)}")
#
# # Stacking
# model1 = RandomForestClassifier(n_estimators=10, random_state=42).fit(x_train, y_train)
# model2 = DecisionTreeClassifier(max_depth=10, random_state=42).fit(x_train, y_train)
# pred1 = model1.predict(x_test)
# pred2 = model2.predict(x_test)
# x_stacked = np.column_stack((pred1, pred2))
# stacked_model = LogisticRegression().fit(x_stacked, y_test)
# print(f"Stacking 模型的准确率为：{stacked_model.score(x_stacked, y_test)}")
# 2回归模型训练
# 2.1线性回归模型
# linner = LinearRegression().fit(x_train, y_train)
# 训练完毕
# 2.1.1回归模型精度
# print(f"模型精度score：{linner.score(x_test,y_test)}")
#
# y_pred_test = linner.predict(x_test)
# print(f"模型精度R**2: {R2(y_test, y_pred_test)}")
# R2： 1- 模型没有捕获到的信息量占真实标签中所带的信息量的比例
# R2越接近1越好
# y_pred_test = linner.predict(x_test)
# print(f"模型精度MSE：{MSE(y_test, y_pred_test)}")
# MSE：预测值和真实值之间的差异
# # 交叉验证得R2
# x1 = cross_val_score(linner, x_train, y_train, cv=5, scoring="r2").mean()
# x2 = cross_val_score(linner, x_test, y_test, cv=5, scoring="r2").mean()
# print(x1, x2)
# # 交叉验证得MSE
# x3 = cross_val_score(linner, x_train, y_train, cv=5, scoring='neg_mean_squared_error').mean()
# x4 = cross_val_score(linner, x_test, y_test, cv=5, scoring='neg_mean_squared_error').mean()
# print(x3, x4)
# 3. 概率模型训练
# 3.1 贝叶斯模型训练
# (多用于文档分类，先进行特征工程词频统计TfidfVectorizer)
# 3.1.1高斯模型 ： 连续性特征
# g = GaussianNB().fit(x_train, y_train)
# 训练完毕
# # 贝叶斯模型预测类别和真实类别,最大概率分配到的标签类别
# print(f"概率模型预测标签类别{g.predict(x_test)}")
# print(f"概率模型真实标签类别{y_test}")
# # 3.1.1.1贝叶斯模型评分
# print(f"高斯贝叶斯模型的精度：{g.score(x_test, y_test)}")
#
# # 查看样本分到不同类别的概率
# print(g.predict_proba(x_test[20].reshape((1, -1))))
# # 概率的对数转化
# print(g.predict_log_proba(x_test[20].reshape((1, -1))))
#
#
# # 3.1.2 多项式模型： 主要适用于离散特征的概率计算
# # 3.1.2.1多项式项式模型训练和精度)
# m = MultinomialNB().fit(x_train, y_train)
# print(f"多项式贝叶斯模型的精度：{m.score(x_test, y_test)}")
# 10.1.1 无监督模型训练kmeans
# # 生成模拟数据
# X, _ = make_blobs(n_samples=300, centers=4, cluster_std=0.6, random_state=0)
#
# # 创建 KMeans 模型并拟合数据
# kmeans = KMeans(n_clusters=4)
# kmeans.fit(X)
#
# # 预测簇标签
# labels = kmeans.labels_
#
# # 获取簇中心点
# centers = kmeans.cluster_centers_
# # 绘制数据点和簇中心点
# plt.figure(figsize=(8, 6))
# plt.scatter(X[:, 0], X[:, 1], c=labels, cmap='viridis', s=50, alpha=0.7, edgecolors='k')
# plt.scatter(centers[:, 0], centers[:, 1], c='red', marker='o', s=200, label='Cluster Centers')
# plt.title('KMeans Clustering')
# plt.xlabel('Feature 1')
# plt.ylabel('Feature 2')
# plt.legend()
# plt.show()
#
# # 计算整个数据集的轮廓系数
# score = silhouette_score(X, labels)
# print("整个数据集的轮廓系数:", score)
#
# # 计算每个样本的轮廓系数
# sample_silhouette_values = silhouette_samples(X, labels)
# print("每个样本的轮廓系数:", sample_silhouette_values)
311
被折叠的 条评论
为什么被折叠?
到【灌水乐园】发言
