PLOT_RMSE plots RMSE data on the GUI
Description:
Function to plot RMSE data on the GUI, both globally and locally, which
the local version having zoom and averaging parameters determined from the
GUI.
Syntax:
PLOT_RMSE(TrainingParams,TrainingResults);
PLOT_EXTRA(TrainingParams,TrainingResults,TestingResults);
Input Parameters:
o TrainingParams: A structure that contains information for training a
theta neuron network, such as the learning method. This structure is
generated by get_training_params.
o TrainingResults: A structure containing error gradient with respect to
the weights (DEDW) and delays (DEDTau). The structure also includes
the SSE, NonFireFlag to indicate is any neurons are not firing along
with NonFireCount, an array used to keep track of how many times a
neuron has not fired over multiple input patterns. This structure is
generated by calculate_gradient. If passed as an input, certain
results are appended.
o TestingResults: A structure containing results generated on the testing
data in the same format as TrainingResults.
Output Parameters:
o None
Example:
>> %Assumes valid simulation training results are loaded below
>> GUIHandle=start_ThNN;
>> Temp=get(GUIHandle,'UserData');
>> Handles=Temp{2};
>> cd('Results');
>> [File,PathLoad] = uigetfile('*.mat','Load Simulation...');
>> cd('..');
>> load([PathLoad, File]);
>> TrainingParams.Handles=Handles;
>> set_params(Handles,ThNN,TrainingParams);
>> plot_rmse(TrainingParams,TrainingResults);
See also theta_neuron_network0001 function plot_rmse(TrainingParams,TrainingResults,TestingResults) 0002 %PLOT_RMSE plots RMSE data on the GUI 0003 % 0004 %Description: 0005 %Function to plot RMSE data on the GUI, both globally and locally, which 0006 %the local version having zoom and averaging parameters determined from the 0007 %GUI. 0008 % 0009 %Syntax: 0010 %PLOT_RMSE(TrainingParams,TrainingResults); 0011 %PLOT_EXTRA(TrainingParams,TrainingResults,TestingResults); 0012 % 0013 %Input Parameters: 0014 %o TrainingParams: A structure that contains information for training a 0015 % theta neuron network, such as the learning method. This structure is 0016 % generated by get_training_params. 0017 %o TrainingResults: A structure containing error gradient with respect to 0018 % the weights (DEDW) and delays (DEDTau). The structure also includes 0019 % the SSE, NonFireFlag to indicate is any neurons are not firing along 0020 % with NonFireCount, an array used to keep track of how many times a 0021 % neuron has not fired over multiple input patterns. This structure is 0022 % generated by calculate_gradient. If passed as an input, certain 0023 % results are appended. 0024 %o TestingResults: A structure containing results generated on the testing 0025 % data in the same format as TrainingResults. 0026 % 0027 %Output Parameters: 0028 %o None 0029 % 0030 %Example: 0031 %>> %Assumes valid simulation training results are loaded below 0032 %>> GUIHandle=start_ThNN; 0033 %>> Temp=get(GUIHandle,'UserData'); 0034 %>> Handles=Temp{2}; 0035 %>> cd('Results'); 0036 %>> [File,PathLoad] = uigetfile('*.mat','Load Simulation...'); 0037 %>> cd('..'); 0038 %>> load([PathLoad, File]); 0039 %>> TrainingParams.Handles=Handles; 0040 %>> set_params(Handles,ThNN,TrainingParams); 0041 %>> plot_rmse(TrainingParams,TrainingResults); 0042 % 0043 %See also theta_neuron_network 0044 0045 %Copyright (C) 2008 Sam McKennoch <Samuel.McKennoch@loria.fr> 0046 0047 0048 Handles=TrainingParams.Handles; 0049 0050 if nargin==1 0051 Data=get(Handles.axes3,'UserData'); 0052 if isempty(Data) 0053 %No Data Has Been Processed Yet... 0054 return; 0055 end 0056 TrainingResults=Data{1}; 0057 if length(Data)==2 0058 TestingResults=Data{2}; 0059 end 0060 end 0061 0062 RMSE=TrainingResults.RMSE; 0063 if exist('TestingResults') 0064 TestingRMSE=TestingResults.RMSE; 0065 else 0066 TestingRMSE=[]; 0067 end 0068 0069 ZoomWindow=floor(get(Handles.Zoom_Window,'Value')); 0070 if ~isempty(TestingRMSE) 0071 plot(Handles.axes4,0:(length(RMSE)-1),RMSE,... 0072 [0 1 TrainingParams.TestingFrequency*(1:(length(TestingRMSE)-2))],TestingRMSE); 0073 legend('Training','Testing'); 0074 else 0075 plot(Handles.axes4,0:(length(RMSE)-1),RMSE); 0076 end 0077 grid(Handles.axes4,'on'); 0078 Yl=get(Handles.axes4,'YLim'); 0079 try %This will fail if the min is negative, in which case auto-scale is ok 0080 set(Handles.axes4,'YLim',[0, min(Yl(2),3e3)]); 0081 end 0082 title(Handles.axes4,'RMSE over all epochs'); 0083 0084 0085 0086 if get(Handles.Average,'Value')==1 0087 AveWindow=round(str2num(get(Handles.Averaging_Window,'String'))); 0088 RMSETemp=RMSE; 0089 RMSE=conv(RMSETemp,ones(1,AveWindow)); 0090 RMSE=RMSE(1:length(RMSETemp)); 0091 LG=min((AveWindow-1),length(RMSE)); 0092 RMSE(1:LG)=RMSE(1:LG)./(1:LG); 0093 RMSE(AveWindow:end)=RMSE(AveWindow:end)/AveWindow; 0094 end 0095 0096 if ~isempty(TestingRMSE) 0097 plot(Handles.axes3,0:(length(RMSE)-1),RMSE,... 0098 [0 1 TrainingParams.TestingFrequency*(1:(length(TestingRMSE)-2))],TestingRMSE); 0099 legend(Handles.axes3,'Training','Testing'); 0100 else 0101 plot(Handles.axes3,0:(length(RMSE)-1),RMSE); 0102 end 0103 grid(Handles.axes3,'on'); 0104 Yl=get(Handles.axes3,'YLim'); 0105 try 0106 if length(RMSE)>ZoomWindow 0107 set(Handles.axes3,'YLim',[max(min(RMSE((length(RMSE)-ZoomWindow):length(RMSE))),0), min(max(RMSE((length(RMSE)-ZoomWindow):length(RMSE))),4e4)]); 0108 end 0109 end 0110 Xl=get(Handles.axes3,'XLim'); 0111 set(Handles.axes3,'XLim',[max(length(RMSE)-ZoomWindow,0), Xl(2)]); 0112 title(Handles.axes3,'MSE over recent epochs'); 0113 0114 drawnow; 0115 0116 %Save For Possible Action Between Diplay/Testing Frequency 0117 if ~isempty(TestingRMSE) 0118 Data={TrainingResults TestingResults}; 0119 else 0120 Data={TrainingResults}; 0121 end 0122 set(Handles.axes3,'UserData',Data);