做特征工程

准确率并没提升太多。

import matplotlib.pyplot as plt

import numpy as np

import pandas as pd

import seaborn as sns

ep=1e-5

func_dict={'add':lambda x,y:x+y,'mins':lambda x,y:x-y,'div':lambda x,y:x/(y+ep),'mul':lambda x,y:x*y}

df = pd.read_csv('train.csv')

df=df.drop(['ID'],axis=1)

prfe=df.iloc[:,:-1]

me=np.reshape(prfe.var(axis=1).to_numpy(),(-1,1))

me1=np.reshape(prfe.std(axis=1).to_numpy(),(-1,1))

df=df.to_numpy()

feature=np.abs(np.fft.fft(df[:,:-1]))

df1 = pd.read_csv('test.csv')

df1=df1.drop(['ID'],axis=1)

df1=df1.to_numpy()

yunt_feature=np.abs(np.fft.fft(df1[:,:]))

feature=pd.DataFrame(feature)

yunt_feature=pd.DataFrame(yunt_feature)

def auto_fea(fe,dunc_dict,col_list):

    for col_i in col_list:

        for col_j in col_list:

            if col_i!=col_j:

                for func_name,func in dunc_dict.items():

                    func_feature=func(fe[col_i],fe[col_j])

                    col_name='-'.join([str(col_i),func_name,str(col_j)])

                    fe[col_name]=func_feature

feature=np.concatenate((feature,np.reshape(df[:,-1],(-1,1))),axis=1)

train=pd.DataFrame(feature)

heat=train.corr()

fe=heat.index[abs(heat[feature.shape[1]-1])>0.3]

train=train.to_numpy()

feature=train[:,fe]

feature=feature[:,:-1]

feature=pd.DataFrame(feature)

yunt_feature=yunt_feature.to_numpy()

yunt_feature=yunt_feature[:,fe[:-1]]

yunt_feature=pd.DataFrame(yunt_feature)

auto_fea(feature,func_dict,feature.columns)

auto_fea(yunt_feature,func_dict,yunt_feature.columns)

yunt_feature=yunt_feature.to_numpy()

feature=feature.to_numpy()

feature=np.concatenate((feature,np.reshape(df[:,-1],(-1,1))),axis=1)

train=pd.DataFrame(feature)

heat=train.corr()

fe1=heat.index[abs(heat[feature.shape[1]-1])>0.42]

print(len(fe1))

train=train.to_numpy()

train=train[:,fe1]

yunt_feature=yunt_feature[:,fe1[:-1]]

from sklearn.model_selection import train_test_split

from sklearn.metrics import accuracy_score

from sklearn import tree

from sklearn.model_selection import cross_val_score

from sklearn.model_selection import KFold

kf=KFold(n_splits=5,shuffle=False)

for k in range(10):

    sum=0

    sum1=0

    i=0

    for train_index,test_index in kf.split(train):

        i=i+1

        tfeature=train[train_index,:-1]

        label=train[train_index,-1]

        clf=tree.DecisionTreeClassifier(criterion='gini',random_state=0,max_depth=k+1)    

        clf.fit(tfeature,label)

        l=clf.predict(tfeature)

        ttest=train[test_index,:-1]

        testlabel=train[test_index,-1]

        l1=clf.predict(ttest)

        pr=accuracy_score(label, l)

        pr1=accuracy_score(testlabel, l1)

        sum=sum+pr

        sum1=sum1+pr1

    clf1=tree.DecisionTreeClassifier(criterion='gini',random_state=0,max_depth=k+1)

    scores = cross_val_score(clf1, train[:,:-1], train[:,-1], cv=5)

    print(k,sum/i,sum1/i,scores.mean())

   

clf1=tree.DecisionTreeClassifier(criterion='gini',random_state=0,max_depth=4+1)    

clf1.fit(train[:,:-1],train[:,-1])

out=clf1.predict(yunt_feature)

out=pd.DataFrame(out)

out.columns = ['CLASS']

w=[]

for k in range(out.shape[0]):

    w.append(k+210)

out['ID']=np.reshape(w,(-1,1))

out[['ID','CLASS']].to_csv('out.csv',index=False)