imported>Jeremy |
imported>Scott |
| Line 1: |
Line 1: |
| ==Introduction== | | ==Introduction== |
| Eigenevctor Research's [[Function_Reference_Manual|PLS_Toolbox]] and [http://www.siemens.com/ Siemens' SiPAT product] can be used together for deployment of PLS_Toolbox or Solo multivariate analysis models in process control applications. This integration utilizes Mathworks' [http://www.mathworks.com/ Matlab] and functionality built into SiPAT to run custom Matlab functions. These functions can include calls to PLS_Toolbox functions.
| |
|
| |
|
| The following discusses the PLS_Toolbox-specific configuration and provides example m-files which can be used to quickly get SiPAT and PLS_Toolbox integrated. | | The Variable Selection panel contains an interface to several methods for performing variable selection. The goal is to find subsets of variables that improve predictions when compared to using all variables. This interface has several different methods available. Finding the best method and options settings will take some experimentation. Use links below for more information on particular methods. |
| __TOC__
| |
|
| |
|
| ==Installation and Basic Configuration== | | ==Methods== |
| ====SiPAT Configuration====
| |
|
| |
|
| The user is directed to the Siemens documentation for details on configuration of SiPAT for use with Matlab. The discussion below describes some of the PLS_Toolbox-specific considerations of this configuration.
| | * Automatic (VIP or sRatio) |
| | * GA - Genetic Algorithm |
| | * iPLS - Interval PLS |
| | * rPLS - Recursive PLS |
| | * sRatio - Selectivity Ratio |
| | * VIP - Variable Importance in Projection |
|
| |
|
| ====Matlab and PLS_Toolbox Configuration==== | | ==Work Flow== |
|
| |
|
| See [[PLS_Toolbox Unattended Configuration]] for details on configuring PLS_Toolbox in these situations.
| | * <u>Select a Method</u> - Select a method from the drop-down menu. Options for the method will be displayed. If a previous calculation has been done, the results of it will be displayed. |
| | | * <u>Adjust Options</u> - By default, a simplified set of options are displayed. If the "Show All Options" checkbox is selected then all available options will be displayed. Depending on the options set, a particular method can take an extended amount of time to complete. For example, decreasing the window width in GA will increase the amount of time it takes to complete. See documentation for more details on optional settings. |
| ==Specific Model and M-file Configuration==
| | * <u>Run Variable Selection</u> - Clicking the "Execute" button will run the current variable selection method with values specified in the options. A waitbar will be displayed indicating the method is running. Some methods will display a waitbar with a message indicating it can be closed to cancel execution. NOTE: It can take some time for the method to finish a calculation loop and identify the user has canceled. If "Show Plots" is checked then any additional plots will be displayed in separate windows. This is useful for GA as it will show progress of the calculation. |
| | | * <u>View Results</u> - When a calculation is complete the selected variables will be displayed under a plot of the data mean as green bars. |
| The SiPAT interface into Matlab provides for the specification of a data file to load and a Matlab function to execute (along with specific input and output configuration information.)
| |
| | |
| In the examples given here, the data file will be in the Matlab MAT format and will contain the model to apply. The Matlab function will be given in the Matlab m-file format and will contain the specific instructions for applying the model and returning the results to SiPAT. In these examples, it is assumed there is only one model that is being applied per each method (no special calibration transfer or other pre-transformation steps being used).
| |
| | |
| ===Model MAT File Creation===
| |
| | |
| The model you wish to apply to new data should be saved from Matlab or Solo into a MAT file as the one and only item stored in the MAT file. In the [[Analysis_Window:_Layout|Analysis Window]], this is done by selecting the menu item: File > Save Model and using the [[WorkspaceBrowser_ImportingData#To_save_imported_data_to_a_.mat_file save dialog]] to specify a filename and an item name. This will save the model into the given filename with the specified item name. Although the MAT file name can be any standard filename, it will make m-file construction easier if the item name used is always the same in all saved models. In the example here, we will assume that the item is called "model". If a different name is used, the SiPAT-specific configuration comments in the m-file (see below) will need to be modified to indicate to SiPAT the name used for the model.
| |
| | |
| If saving the model from the Matlab command line, the following command can be used (assuming the model to use is currently named "model" in your Matlab workspace) :
| |
| save myfile.mat model
| |
| | |
| ===Function M-file Creation===
| |
| | |
| The second part to the SiPAT/PLS_Toolbox interface is a Matlab m-file which contains a function definition (Note: The contents of this m-file must actually be a Matlab function, meaning it must contain a function header line as shown in the scripts below. It cannot be a "script" in the strict Matlab definition of that term which implies code that is not wrapped inside a function definition.)
| |
| | |
| Three example m-files are given below: one for use with any of the regression model types (PLS, PCR, MLR, CLS, SVM, LWR, etc), one for use with classification model types (PLSDA, KNN, SIMCA, SVMDA/SVM-C), and one for use with principal component analysis (PCA) models. Each of these m-files assumes that the input is a single matrix (passed by SiPAT) and each returns two or three values which correspond to the predictions from the model.
| |
| | |
| These functions all also assume that the input data will be two columns where the first column contains axis scale information for the variables (such as wavelength, m/z, time, etc) and the second column is the actual measured data. If this does not fit the type of data being passed by SiPAT (e.g. no axisscale information), the initial lines:
| |
| | |
| <pre>
| |
| %convert second column of input data into a dataset (if not appropriate column, change next line)
| |
| x = data(:,2);
| |
| x = dataset(double(x'));
| |
| | |
| %Assume first column is axisscale information. If not true, comment out the next line
| |
| x.axisscale{2} = double(data(:,1));
| |
| </pre>
| |
| | |
| should be converted as necessary to handle the input data. For example, if only a single column of values is being passed, the following can be substituted in for the above code:
| |
| | |
| <pre>
| |
| %convert column of input data into a dataset
| |
| x = dataset(double(data'));
| |
| </pre>
| |
| | |
| '''Errors:''' These functions make use of try/catch statements to trap errors and save pertinent information into a text file in the root C: directory. The location of this file can be changed as desired. This code is added to help diagnose configuration problems. It is recommended that, in a final deployment, specific error codes be used such as setting all outputs to "inf" to trigger an alarm in SiPAT.
| |
| | |
| ====Regression Model Predictions====
| |
| | |
| The following m-file contents are appropriate for use with regression model types:
| |
| | |
| <pre>
| |
| %START SIPAT
| |
| %<CONFIG>
| |
| % <MODEL>REGRESSION</MODEL>
| |
| % <PREFIX></PREFIX>
| |
| % <SUFFIX></SUFFIX>
| |
| % <INPUTS>
| |
| % <INPUT Name="data" XDataType="MultiValue" YDataType="Single" />
| |
| % </INPUTS>
| |
| % <OUTPUTS>
| |
| % <OUTPUT Name="y" XDataType="SingleValue" YDataType="Double" />
| |
| % <OUTPUT Name="q_x" XDataType="SingleValue" YDataType="Double" />
| |
| % <OUTPUT Name="H_x" XDataType="SingleValue" YDataType="Double" />
| |
| % </OUTPUTS>
| |
| % <FUNCTION>[y,q_x,H_x]=regpred(data,model)</FUNCTION>
| |
| %</CONFIG>
| |
| %END SIPAT
| |
| %%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%
| |
| % Regression using PLS_toobox from Eigenvector
| |
| %%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%
| |
| %This function calculates the responses y and the Q-residuals
| |
| %(q_x) and Hotellings T2 (H_x) corresponding to the input variables data
| |
| %(data) using a standard model structure.
| |
| %
| |
| %Before using this function, make sure you loaded the following data into
| |
| %the workspace:
| |
| % model: Model structure used in PLS_Toolbox/Solo including all pretreatment.
| |
| | |
| function [y,q_x,H_x] = regpred(data,model)
| |
| | |
| %defaults if something goes wrong or we can't get Q or T^2 from this model type
| |
| q_x = inf;
| |
| H_x = inf;
| |
| | |
| try
| |
| %convert second column of input data into a dataset (if not appropriate column, change next line)
| |
| x = data(:,2);
| |
| x = dataset(double(x'));
| |
| | |
| %Assume first column is axisscale information. If not true, comment out the next line
| |
| x.axisscale{2} = double(data(:,1));
| |
| | |
| %make a prediction
| |
| opts =[];
| |
| opts.plots='none';
| |
| opts.display='off';
| |
| pred_x = feval(lower(model.modeltype),x,model,opts);
| |
| | |
| %return prediction in y
| |
| y = double(pred_x.pred{2});
| |
| if isfield(pred_x,'ssqresiduals')
| |
| q_x = double(pred_x.ssqresiduals{1,1}./pred_x.detail.reslim{1});
| |
| H_x = double(pred_x.tsqs{1}./pred_x.detail.tsqlim{1});
| |
| end
| |
| | |
| catch
| |
| %errors are saved to the following file (change location as desired)
| |
| fid=fopen('C:\sipat_error_sout.txt','w');
| |
| fwrite(fid,encode(lasterror));
| |
| fwrite(fid,encode(evalin('base','whos')));
| |
| fclose(fid);
| |
| end
| |
| </pre>
| |
| | |
| ====Classification Model Predictions====
| |
| | |
| The following m-file contents are appropriate for use with classification model types:
| |
| | |
| <pre>
| |
| %START SIPAT
| |
| %<CONFIG>
| |
| % <MODEL>CLASSIFICATION</MODEL>
| |
| % <PREFIX></PREFIX>
| |
| % <SUFFIX></SUFFIX>
| |
| % <INPUTS>
| |
| % <INPUT Name="data" XDataType="MultiValue" YDataType="Single" />
| |
| % </INPUTS>
| |
| % <OUTPUTS>
| |
| % <OUTPUT Name="y" XDataType="SingleValue" YDataType="Double" />
| |
| % <OUTPUT Name="q_x" XDataType="SingleValue" YDataType="Double" />
| |
| % <OUTPUT Name="H_x" XDataType="SingleValue" YDataType="Double" />
| |
| % </OUTPUTS>
| |
| % <FUNCTION>[y,q_x,H_x]=classpred(data,model)</FUNCTION>
| |
| %</CONFIG>
| |
| %END SIPAT
| |
| %%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%
| |
| % Classification using PLS_toobox from Eigenvector
| |
| %%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%
| |
| %This function calculates the numerical class assignment, the Q-residuals
| |
| %(q_x) and Hotellings T2 (H_x) corresponding to the input variables data
| |
| %(data) using a standard model structure.
| |
| %
| |
| %Before using this function, make sure you loaded the following data into
| |
| %the workspace:
| |
| % model: Model structure used in PLS_Toolbox/Solo including all pretreatment.
| |
| | |
| function [y,q_x,H_x] = classpred(data,model)
| |
| | |
| %defaults if something goes wrong or we can't get Q or T^2 from this model type
| |
| q_x = inf;
| |
| H_x = inf;
| |
| | |
| try
| |
| %convert second column of input data into a dataset (if not appropriate column, change next line)
| |
| x = data(:,2);
| |
| x = dataset(double(x'));
| |
| | |
| %Assume first column is axisscale information. If not true, comment out the next line
| |
| x.axisscale{2} = double(data(:,1));
| |
| | |
| %make a prediction
| |
| opts =[];
| |
| opts.plots='none';
| |
| opts.display='off';
| |
| pred_x = feval(lower(model.modeltype),x,model,opts);
| |
| | |
| %return prediction in y
| |
| y = double(pred_x.classification.mostprobable);
| |
| if isfield(pred_x,'ssqresiduals')
| |
| q_x = double(pred_x.ssqresiduals{1,1}./pred_x.detail.reslim{1});
| |
| H_x = double(pred_x.tsqs{1}./pred_x.detail.tsqlim{1});
| |
| end
| |
| | |
| catch
| |
| %errors are saved to the following file (change location as desired)
| |
| fid=fopen('C:\sipat_error_sout.txt','w');
| |
| fwrite(fid,encode(lasterror));
| |
| fwrite(fid,encode(evalin('base','whos')));
| |
| fclose(fid);
| |
| end
| |
| </pre>
| |
| | |
| | |
| ====PCA Model Predictions====
| |
| | |
| The following m-file contents are appropriate for use with a PCA model type. As written, it returns only TWO values, the Q and T2 values indicating if the sample belongs in the PCA model or not. This construction can be used when using a PCA model to detect process anomalies.
| |
| | |
| <pre>
| |
| %START SIPAT
| |
| %<CONFIG>
| |
| % <MODEL>PCA</MODEL>
| |
| % <PREFIX></PREFIX>
| |
| % <SUFFIX></SUFFIX>
| |
| % <INPUTS>
| |
| % <INPUT Name="data" XDataType="MultiValue" YDataType="Single" />
| |
| % </INPUTS>
| |
| % <OUTPUTS>
| |
| % <OUTPUT Name="q_x" XDataType="SingleValue" YDataType="Double" />
| |
| % <OUTPUT Name="H_x" XDataType="SingleValue" YDataType="Double" />
| |
| % </OUTPUTS>
| |
| % <FUNCTION>[q_x,H_x]=pcapred(data,model)</FUNCTION>
| |
| %</CONFIG>
| |
| %END SIPAT
| |
| %%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%
| |
| % PCA Projection using PLS_toobox from Eigenvector
| |
| %%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%
| |
| %This function calculates the Q-residuals (q_x) and Hotellings T2 (H_x)
| |
| %corresponding to the input variables data (data) using a standard model
| |
| %structure.
| |
| %
| |
| %Before using this function, make sure you loaded the following data into
| |
| %the workspace:
| |
| % model: Model structure used in PLS_Toolbox/Solo including all pretreatment.
| |
| | |
| function [q_x,H_x] = pcapred(data,model)
| |
| | |
| %defaults if something goes wrong or we can't get Q or T^2 from this model type
| |
| q_x = inf;
| |
| H_x = inf;
| |
| | |
| try
| |
| %convert second column of input data into a dataset (if not appropriate column, change next line)
| |
| x = data(:,2);
| |
| x = dataset(double(x'));
| |
| | |
| %Assume first column is axisscale information. If not true, comment out the next line
| |
| x.axisscale{2} = double(data(:,1));
| |
| | |
| %make a prediction
| |
| opts =[];
| |
| opts.plots='none';
| |
| opts.display='off';
| |
| pred_x = feval(lower(model.modeltype),x,model,opts);
| |
| | |
| %return prediction of Q and T^2
| |
| if isfield(pred_x,'ssqresiduals')
| |
| q_x = double(pred_x.ssqresiduals{1,1}./pred_x.detail.reslim{1});
| |
| H_x = double(pred_x.tsqs{1}./pred_x.detail.tsqlim{1});
| |
| end
| |
| | |
| %NOTE: if output of scores is desired, add a "y" output to
| |
| %the function and SiPAT definition and use the next line:
| |
| % y = pred_x.loads{1,1};
| |
| %which would return all the scores (as a row vector) for the given data
| |
| | |
| catch
| |
| %errors are saved to the following file (change location as desired)
| |
| fid=fopen('C:\sipat_error_sout.txt','w');
| |
| fwrite(fid,encode(lasterror));
| |
| fwrite(fid,encode(evalin('base','whos')));
| |
| fclose(fid);
| |
| end
| |
| </pre>
| |