向数据集加了一堆均值并没有提高多少

准确率为0.8。感觉乱塞特征并没有多大提升。特征加了一堆换了SVM决策树，提升树，随机森林准确率都一般。自己一个人做不上于是不做了。

import matplotlib.pyplot as plt

import numpy as np

import pandas as pd

import torch

import torch.fft as fft

from sklearn import tree

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

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

nmp=df.to_numpy()

feature=nmp[:,:-1]

df=pd.DataFrame(feature)

sk=np.reshape(np.array(df.skew(axis=1)),(-1,1))

Q25=np.reshape(np.array(df.quantile(q=0.25,axis=1)),(-1,1))

Q75=np.reshape(np.array(df.quantile(q=0.75,axis=1)),(-1,1))

dmedia=np.reshape(np.array(df.diff(axis=1).dropna(axis=1).median(axis=1)),(-1,1))

dsk=np.reshape(np.array(df.diff(axis=1).dropna(axis=1).skew(axis=1)),(-1,1))

dQ25=np.reshape(np.array(df.diff(axis=1).dropna(axis=1).quantile(q=0.25,axis=1)),(-1,1))

dQ75=np.reshape(np.array(df.diff(axis=1).dropna(axis=1).quantile(q=0.75,axis=1)),(-1,1))

dk=np.reshape(np.array(df.diff(axis=1).dropna(axis=1).kurtosis(axis=1)),(-1,1))

dmax=np.reshape(np.array(df.diff(axis=1).dropna(axis=1)).max(axis=1),(-1,1))

dmin=np.reshape(np.array(df.diff(axis=1).dropna(axis=1)).min(axis=1),(-1,1))

range_diff1=dmax-dmin

dvar=np.reshape(np.array(df.diff(axis=1).dropna(axis=1)).var(axis=1),(-1,1))

dstd=np.reshape(np.array(df.diff(axis=1).dropna(axis=1)).std(axis=1),(-1,1))

dmean=np.reshape(np.array(df.diff(axis=1).dropna(axis=1)).mean(axis=1),(-1,1))

feature=nmp[:,:-1]

label=nmp[:,-1]#（210，240）

min=np.reshape(nmp[:,:-1].min(-1),(-1,1))

max=np.reshape(nmp[:,:-1].max(-1),(-1,1))

ra=min-max

var=np.reshape(np.var(nmp[:,:-1],axis=1),(-1,1))

std=np.reshape(np.std(nmp[:,:-1],axis=1),(-1,1))

mean=np.reshape(np.mean(nmp[:,:-1],axis=1),(-1,1))

media=np.reshape(np.median(nmp[:,:-1],axis=1),(-1,1))

feature1=torch.fft.fft(torch.Tensor(feature))

feature1=torch.abs(feature1)/240*2

feature1=feature1.detach().numpy()

df=pd.DataFrame(feature1)

sk3=np.reshape(np.array(df.skew(axis=1)),(-1,1))

Q253=np.reshape(np.array(df.quantile(q=0.25,axis=1)),(-1,1))

Q753=np.reshape(np.array(df.quantile(q=0.75,axis=1)),(-1,1))

dmax3=np.reshape(np.array(df.diff(axis=1).dropna(axis=1).max(axis=1)),(-1,1))

dmin3=np.reshape(np.array(df.diff(axis=1).dropna(axis=1).min(axis=1)),(-1,1))

range_diff3=dmax-dmin

dvar3=np.reshape(np.array(df.diff(axis=1).dropna(axis=1).var(axis=1)),(-1,1))

dstd3=np.reshape(np.array(df.diff(axis=1).dropna(axis=1).std(axis=1)),(-1,1))

dmean3=np.reshape(np.array(df.diff(axis=1).dropna(axis=1).mean(axis=1)),(-1,1))

dmedia3=np.reshape(np.array(df.diff(axis=1).dropna(axis=1).median(axis=1)),(-1,1))

dsk3=np.reshape(np.array(df.diff(axis=1).dropna(axis=1).skew(axis=1)),(-1,1))

dQ253=np.reshape(np.array(df.diff(axis=1).dropna(axis=1).quantile(q=0.25,axis=1)),(-1,1))

dQ753=np.reshape(np.array(df.diff(axis=1).dropna(axis=1).quantile(q=0.75,axis=1)),(-1,1))

dk3=np.reshape(np.array(df.diff(axis=1).dropna(axis=1).kurtosis(axis=1)),(-1,1))

min1=np.reshape(feature1.min(-1),(-1,1))

max1=np.reshape(feature1.max(-1),(-1,1))

var1=np.reshape(np.var(feature1,axis=1),(-1,1))

std1=np.reshape(np.std(feature1,axis=1),(-1,1))

mean1=np.reshape(np.mean(feature1,axis=1),(-1,1))

media1=np.reshape(np.median(feature1,axis=1),(-1,1))

import scipy

kur=np.reshape(scipy.stats.kurtosis(feature,axis=1),(-1,1))

ne=np.concatenate((dmin3,range_diff3,dvar3,dstd3,dmean3,dmedia3,dsk3,dQ253,dQ753,dk3,dmax3,sk3,Q253,Q753,dmax,range_diff1,dvar,dmean,dstd,dmin,dmedia,dsk,dQ25,dQ75,dmax,dmin,dvar,dstd,dmean,range_diff1,sk3,Q25,Q75,sk,kur,feature,ra,min,max,var,std,mean,media,feature1,min1,max1,var1,std1,mean1,media1),axis=1)

#ne=np.concatenate((dmedia,dvar,dstd,range_diff1,dmax,dmin,Q25,Q75,sk,kur,feature,ra,min,max,var,std,mean,media,feature1,min1,max1,var1,std1,mean1,media1),axis=1)

from sklearn.model_selection import cross_val_score

from sklearn import svm

import matplotlib.pyplot as plt

from sklearn.ensemble import RandomForestClassifier

from sklearn.ensemble import GradientBoostingClassifier

from sklearn.neighbors import KNeighborsClassifier

for i in range(30):

    #clf = KNeighborsClassifier(n_neighbors=i+1,weights='distance')

    #for j in range(20):

    #clf=RandomForestClassifier(n_estimators=i*50+1,max_depth=3)

    #clf=GradientBoostingClassifier(max_depth=i+1)

    clf=tree.DecisionTreeClassifier(criterion='entropy',random_state=0,max_depth=i+1)

    #clf = svm.SVC(kernel = 'rbf')

    scores = cross_val_score(clf,ne,label,cv=10)

    print(i,scores.mean())