1.KNN算法

KNN算法

KNN(k近邻算法)，极其简单我用于机器学习入门
选择K个邻居，根据他们的类别投票，分类回归都可以用他

1. knn的使用流程化

1. train_test_split
   我自定义的该方法

import numpy as np
def train_test_split(x_data,y_data,rate=0.8,seed=None):
    '''
    实现自己的train_test_split
    :param x_data: 原生数据集合
    :param y_data:原生数据集合
    :return: X_train, X_test, y_train, y_test
    '''
    if seed:
        np.random.seed(seed)
    index_list=[i for i in range(len(y_data))]
    index_list=np.random.permutation(index_list)
    x_data=x_data[index_list]
    y_data=y_data[index_list]
    rate=int(len(y_data)*rate)
    return x_data[:rate],x_data[rate:],y_data[:rate],y_data[rate:]

sklearn里面的使用方法类似，参数名:train_size random_state
2. 数据归一化，数据预处理（important）

什么时候归一化，不要所有时候都归一化
不恰当的时候的归一化，会降低准确度,当某一个特征比另外的特征的数值在计算时远大时归一化

我们一般使用Standard_Scaler

from sklearn.preprocessing import StandardScaler
standardScaler=StandardScaler()
standardScaler.fit(X_train)
X_train=standardScaler.transform(X_train)
X_test=standardScaler.transform(X_test)

3. 调参(确定超参数):
   KNN的超参 K,P,weight
   weigjt :“uniform” | “distance”
   P:“明可夫斯基距离"
   我们一般使用sklearn的GridSearchCV来进行网格搜索，该方法使用交叉验证的方式进行评估,扩展性广

param_grid=[
    {
        "weights":['uniform'],
        "n_neighbors":[i for i in range(1,11)]
    },
    {
        "weights":['distance'],
        "n_neighbors":[i for i in range(1,11)],
        "p":[i for i in range(1,11)],
    }
]
knns=KNeighborsClassifier()
grid_search=GridSearchCV(knns,param_grid=params_grid,n_jobs=-1,verbose=2)
grid_search.fit(X_train,y_train)
print(grid_search.best_estimator_)
print(grid_search.best_score_)

在KNN里面我们可以自己用准确率来评估，但判断标准比不上交叉验证

weights=['distance','uniform']
best_k=0
best_p=0
best_score=0
best_weights=''
for weight in weights:
    for k in range(1,11):
        for p in range(1,11):
            if weight=='distance':
                knn=KNeighborsClassifier(n_neighbors=k,weights=weight,n_jobs=-1,p=p)
            else:
                knn = KNeighborsClassifier(n_neighbors=k, weights=weight,n_jobs=-1)
            knn.fit(X_train,y_train)
            now_score=knn.score(X_test,y_test)
            if now_score>best_score:
                best_score=now_score
                best_p=p
                best_weights=weight
                best_k=k
print(best_score,best_p,best_k,best_weights)

4. fit模型 ，predict预测