【算法对比图】回归、时序预测的多算法对比图

硬件电子工程师 · 发表于 2023-12-21 22:21:00

rx3akpfe0mf64015154517.png

点击上方蓝字关注我们

jkiz3myaftd64015154617.png

注明：此推文来自公众号Lvy的口袋，欢迎大家关注Lvy小姐姐公众号~ 多种算法对比图是常用的科研绘图，你知道几种合适的绘图样式呢？

g4kxpj422m064015154717.png

) b4 R& g! i; A* I7 J

1.真实值和预测值展示图

la2n3mfhzg264015154817.png

Tips：数据比较多、算法多的适合比较难看出实际的效果
数据就是各个算法预测值和真实值数据（工具箱直接导出）

data_pre_all=[]; %记录预测数据load(' 多元线性回归 17_Dec_11_34_33 train_result_train_vaild_test.mat')data1=data_Oriny_prey.y_test_predict;data_pre_all=[data_pre_all,data1];data_true=data_Oriny_prey.test_y;load('SSA麻雀搜索算法随机森林回归 17_Dec_11_35_55 train_result_train_vaild_test.mat')data2=data_Oriny_prey.y_test_predict;data_pre_all=[data_pre_all,data2];load(' SVM-RF回归 17_Dec_11_37_18 train_result_train_vaild_test.mat')data3=data_Oriny_prey.y_test_predict;data_pre_all=[data_pre_all,data3];load(' MLP回归 17_Dec_11_38_31 train_result_train_vaild_test.mat')data4=data_Oriny_prey.y_test_predict;data_pre_all=[data_pre_all,data4];load(' LSTM回归 17_Dec_11_40_29 train_result_train_vaild_test.mat')data5=data_Oriny_prey.y_test_predict;data_pre_all=[data_pre_all,data5];str={'真实值','多元线性回归','SSA麻雀搜索算法随机森林回归','SVM-RF回归' ,'MLP回归','LSTM回归'};figure('Units', 'pixels', ... 'Position', [300 300 860 375]);plot(data_true,'--*') hold onfor i=1:size(data_pre_all,2) plot(data_pre_all(:,i)) hold on endlegend(str)set (gca,"FontSize",12,'LineWidth',1.2)box offlegend Box off

2.误差柱状对比图

wab51yehzkx64015154917.png

Tips：建议选取量纲差别不大的误差衡量指标，不然可能会有点丑

Test_all=[];for j=1:size(data_pre_all,2) y_test_predict=data_pre_all(:,j); test_y=data_true; test_MAE=sum(abs(y_test_predict-test_y))/length(test_y) ; test_MAPE=sum(abs((y_test_predict-test_y)./test_y))/length(test_y); test_MSE=(sum(((y_test_predict-test_y)).^2)/length(test_y)); test_RMSE=sqrt(sum(((y_test_predict-test_y)).^2)/length(test_y)); test_R2= 1 - (norm(test_y - y_test_predict)^2 / norm(test_y - mean(test_y))^2); Test_all=[Test_all;test_MAE test_MAPE test_MSE test_RMSE test_R2];end%%str={'真实值','多元线性回归','SSA麻雀搜索算法随机森林回归','SVM-RF回归' ,'MLP回归','LSTM回归'};str1=str(2:end);str2={'MAE','MAPE','MSE','RMSE','R2'};data_out=array2table(Test_all);data_out.Properties.VariableNames=str2;data_out.Properties.RowNames=str1;disp(data_out)%% 柱状图 MAE MAPE RMSE 柱状图适合量纲差别不大的color= [0.1569 0.4706 0.7098 0.6039 0.7882 0.8588 0.9725 0.6745 0.5490 0.8549 0.9373 0.8275 0.7451 0.7216 0.8627 0.7843 0.1412 0.1373 1.0000 0.5333 0.5176 0.5569 0.8118 0.7882 1.0000 0.5333 0.5176];figure('Units', 'pixels', ... 'Position', [300 300 660 375]);plot_data_t=Test_all(:,[1,2,4])';b=bar(plot_data_t,0.8);hold on
for i = 1 : size(plot_data_t,2) x_data(:, i) = b(i).XEndPoints'; end
for i =1:size(plot_data_t,2)b(i).FaceColor = color(i,:);b(i).EdgeColor=[0.6353 0.6314 0.6431];b(i).LineWidth=1.2;end
for i = 1 : size(plot_data_t,1)-1 xilnk=(x_data(i, end)+ x_data(i+1, 1))/2; b1=xline(xilnk,'--','LineWidth',1.2); hold onend
ax=gca;legend(b,str1,'Location','best')ax.XTickLabels ={'MAE', 'MAPE', 'RMSE'};set(gca,"FontSize",12,"LineWidth",2)box offlegend box off

1 m2 [4 g) N" U) R. j) L2 K

3.误差散点对比图

wa5kjssogz264015155017.png

Tips：可以任意选择两个误差衡量维度

figureplot_data_t1=Test_all(:,[1,5])';MarkerType={'s','o','pentagram','^','v'};for i = 1 : size(plot_data_t1,2) scatter(plot_data_t1(1,i),plot_data_t1(2,i),120,MarkerType{i},"filled") hold onendset(gca,"FontSize",12,"LineWidth",2)box offlegend box offlegend(str1,'Location','best')xlabel('MAE')ylabel('R2')grid on* S5 k& L# R2 a% p2 V4 s' s

4.误差密度散点图

ged0r4oybv564015155117.png

figure('Units', 'pixels', ... 'Position', [150 150 920 500]);for i=1:5 subplot(2,3,i) n=50; X=double(data_true); Y=double(data_pre_all(:,i)); M=polyfit(X,Y,1); Y1=polyval(M,X); XList=linspace(min(X),max(X),n); YList=linspace(min(Y),max(Y),n); [XMesh,YMesh]=meshgrid(XList,YList); F=ksdensity([X,Y],[XMesh(:),YMesh(:)]); ZMesh=reshape(F,size(XMesh)); H=interp2(double(XMesh),double(YMesh),double(ZMesh),X,Y); scatter(data_true,data_pre_all(:,i),35,'filled','CData',H,'MarkerFaceAlpha',.5); hold on plot(X(1:10:end),Y1(1:10:end),'--','LineWidth',1.2) hold on str_label=[str1{1,i},' ','R2=',num2str(Test_all(i,end))]; title(str_label) set(gca,"FontSize",10,"LineWidth",1.5) xlabel('true') ylabel('predict')end

5.误差雷达图

jlvj1itw1pj64015155218.png

Tips：为了让图片更美观将多个维度评价指标进行归一化处理了

figure('Units', 'pixels', ... 'Position', [150 150 520 500]);Test_all1=Test_all./sum(Test_all); %把各个指标归一化到一个量纲Test_all1(:,end)=1-Test_all(:,end);RC=radarChart(Test_all1);str3={'A-MAE','A-MAPE','A-MSE','A-RMSE','1-R2'};RC.PropName=str3;RC.ClassName=str1;RC=RC.draw(); RC.legend();colorList=[78 101 155; 138 140 191; 184 168 207; 231 188 198; 253 207 158; 239 164 132; 182 118 108]./255;for n=1:RC.ClassNum RC.setPatchN(n,'Color',colorList(n,:),'MarkerFaceColor',colorList(n,:))end
本图参考了公众号：slandarer随笔
https://mp.weixin.qq.com/s/8Lu7yBs3cLlZk9bPStdgUA

调用函数

classdef radarChart% @Author : slandarer% 公众号  : slandarer随笔% 知乎 : hikari
properties       ax;arginList={'ClassName','PropName','Type'}       XData;RTick=[];RLim=[];SepList=[1,1.2,1.5,2,2.5,3,4,5,6,8]       Type='Line';       PropNum;ClassNum       ClassName={};       PropName={};
      BC=[198,199,201;  38, 74, 96; 209, 80, 51; 241,174, 44; 12,13,15;          102,194,165; 252,140, 98; 142,160,204; 231,138,195;          166,217, 83; 255,217, 48; 229,196,148; 179,179,179]./255;
      % 句柄       ThetaTickHdl;RTickHdl;RLabelHdl;LgdHdl;PatchHdl;PropLabelHdl;BkgHdl end
methods       function obj=radarChart(varargin)          if isa(varargin{1},'matlab.graphics.axis.Axes')             obj.ax=varargin{1};varargin(1)=[];          else             obj.ax=gca;          end          % 获取版本信息          tver=version('-release');          verMatlab=str2double(tver(1:4))+(abs(tver(5))-abs('a'))/2;          if verMatlab             hold on          else             hold(obj.ax,'on')          end
         obj.XData=varargin{1};varargin(1)=[];          obj.PropNum=size(obj.XData,2);          obj.ClassNum=size(obj.XData,1);          obj.RLim=[0,max(obj.XData,[],[1,2])];
         % 获取其他信息          for i=1:2:(length(varargin)-1)             tid=ismember(obj.arginList,varargin{i});             if any(tid)             obj.(obj.arginList{tid})=varargin{i+1};             end          end          if isempty(obj.ClassName)             for i=1:obj.ClassNum                   obj.ClassName{i}=['class ',num2str(i)];             end          end          if isempty(obj.PropName)             for i=1:obj.PropNum                   obj.PropName{i}=['prop ',num2str(i)];             end          end          help radarChart       end
      function obj=draw(obj)          obj.ax.XLim=[-1,1];          obj.ax.YLim=[-1,1];          obj.ax.XTick=[];          obj.ax.YTick=[];          obj.ax.XColor='none';          obj.ax.YColor='none';          obj.ax.PlotBoxAspectRatio=[1,1,1];          % 绘制背景圆形          tt=linspace(0,2*pi,200);          obj.BkgHdl=fill(cos(tt),sin(tt),[252,252,252]./255,'EdgeColor',[200,200,200]./255,'LineWidth',1);          % 绘制Theta刻度线          tn=linspace(0,2*pi,obj.PropNum+1);tn=tn(1:end-1);          XTheta=[cos(tn);zeros([1,obj.PropNum]);nan([1,obj.PropNum])];          YTheta=[sin(tn);zeros([1,obj.PropNum]);nan([1,obj.PropNum])];          obj.ThetaTickHdl=plot(XTheta(:),YTheta(:),'Color',[200,200,200]./255,'LineWidth',1);          % 绘制R刻度线          if isempty(obj.RTick)             dr=diff(obj.RLim);             sepR=dr./3;             multiE=ceil(log(sepR)/log(10));             sepR=sepR.*10^(1-multiE);             sepR=obj.SepList(find(sepR
            sepNum=floor(dr./sepR);             obj.RTick=obj.RLim(1)+(0:sepNum).*sepR;             if obj.RTick(end)~=obj.RLim(2)                   obj.RTick=[obj.RTick,obj.RLim];             end          end          obj.RLim(obj.RLim          obj.RLim(obj.RLim>obj.RLim(2))=[];
         XR=cos(tt').*(obj.RTick-obj.RLim(1))./diff(obj.RLim);XR=[XR;nan([1,length(obj.RTick)])];          YR=sin(tt').*(obj.RTick-obj.RLim(1))./diff(obj.RLim);YR=[YR;nan([1,length(obj.RTick)])];          obj.RTickHdl=plot(XR(:),YR(:),'Color',[200,200,200]./255,'LineWidth',1.1,'LineStyle','--');
         % 绘制雷达图          for i=1:size(obj.XData,1)             XP=cos(tn).*(obj.XData(i,:)-obj.RLim(1))./diff(obj.RLim);             YP=sin(tn).*(obj.XData(i,:)-obj.RLim(1))./diff(obj.RLim);             switch obj.Type                   case 'Line'                      obj.PatchHdl(i)=plot([XP,XP(1)],[YP,YP(1)],...                         'Color',obj.BC(mod(i-1,size(obj.BC,1))+1,:),'Marker','o',...                         'LineWidth',1.8,'MarkerFaceColor',obj.BC(mod(i-1,size(obj.BC,1))+1,:));                   case 'Patch'                      obj.PatchHdl(i)=patch(XP,YP,obj.BC(mod(i-1,size(obj.BC,1))+1,:),...                         'EdgeColor',obj.BC(mod(i-1,size(obj.BC,1))+1,:),'FaceAlpha',.2,...                         'LineWidth',1.8);
            end          end
         % 绘制R标签文本          tnr=(tn(1)+tn(2))/2;          for i=1:length(obj.RTick)             obj.RLabelHdl(i)=text(cos(tnr).*(obj.RTick(i)-obj.RLim(1))./diff(obj.RLim),...                                     sin(tnr).*(obj.RTick(i)-obj.RLim(1))./diff(obj.RLim),...                                     sprintf('%.2f',obj.RTick(i)),'FontName','Arial','FontSize',11);          end
         % 绘制属性标签          for i=1:obj.PropNum             obj.PropLabelHdl(i)=text(cos(tn(i)).*1.1,sin(tn(i)).*1.1,obj.PropName{i},...                   'FontSize',12,'HorizontalAlignment','center');          end
      end% =========================================================================       function obj=setBkg(obj,varargin)          set(obj.BkgHdl,varargin{:})       end
      % 绘制图例       function obj=legend(obj)          obj.LgdHdl=legend([obj.PatchHdl],obj.ClassName,'FontSize',12,'Location','best');       end       % 设置图例属性       function obj=setLegend(obj,varargin)          set(obj.LgdHdl,varargin{:})       end
      % 设置标签       function obj=setPropLabel(obj,varargin)          for i=1:obj.PropNum             set(obj.PropLabelHdl(i),varargin{:})          end       end       function obj=setRLabel(obj,varargin)          for i=1:length(obj.RLabelHdl)             set(obj.RLabelHdl(i),varargin{:})          end       end
      % 设置轴       function obj=setRTick(obj,varargin)          set(obj.RTickHdl,varargin{:})       end       function obj=setThetaTick(obj,varargin)          set(obj.ThetaTickHdl,varargin{:})       end
      % 设置patch属性       function obj=setPatchN(obj,N,varargin)          set(obj.PatchHdl(N),varargin{:})       end end% @author : slandarer% 公众号  : slandarer随笔% 知乎 : hikariend
; w. ^2 w/ E1 G# w5 c

6.误差罗盘图

1yz2olsrjyw64015155318.png

figure('Units', 'pixels', ... 'Position', [150 150 920 600]);t = tiledlayout('flow','TileSpacing','compact');for i=1:length(Test_all(:,1))nexttileth1 = linspace(2*pi/length(Test_all(:,1))/2,2*pi-2*pi/length(Test_all(:,1))/2,length(Test_all(:,1)));r1 = Test_all(:,i)';[u1,v1] = pol2cart(th1,r1);M=compass(u1,v1);for j=1:length(Test_all(:,1)) M(j).LineWidth = 2; M(j).Color = colorList(j,:);
end title(str2{i})set(gca,"FontSize",10,"LineWidth",1)end legend(M,str1,"FontSize",10,"LineWidth",1,'Box','off','Location','southoutside')
时序的和回归的算法比较也是类似的，【领取数据和代码方式】，在公众号【Lvy的口袋】（下方链接直接进行公众号）后台回复关键词【算法对比图】领取，还有什么比较合适的对比图可以私发小编看能不能复现奥~

ps.合适的绘图之后可能会更新到工具箱中,全家桶大力更新中~早上车早实惠

全家桶系列
一键打包公众号过去和未来所有的作品~持续更新中【获取方式】扫码获取或者点击链接
https://mbd.pub/o/bread/mbd-ZJabmJ9v

01efpmj0crv64015155418.png

hk1fqkwdzgy64015155518.png

END

m2j1p5lkx0y64015155618.png

5gobmhmv0we64015155718.jpg