Scribd
Upload
7 views6 pages

PLSR Using MATLAB

The document provides a detailed guide on performing Partial Least-Squares Regression (PLSR) using MATLAB R2018a, including steps for importing data, normalizing it, and executing the regression. It outlines the necessary code snippets for data manipulation, regression analysis, and visualization of results, such as percent variance explained and scores plots. Additionally, it covers cross-validation techniques to assess the model's predictive capability.

Uploaded by

madadi morad
Copyright
© © All Rights Reserved
We take content rights seriously. If you suspect this is your content, claim it here.
Available Formats
Download as PDF, TXT or read online on Scribd
7 views6 pages

PLSR Using MATLAB

The document provides a detailed guide on performing Partial Least-Squares Regression (PLSR) using MATLAB R2018a, including steps for importing data, normalizing it, and executing the regression. It outlines the necessary code snippets for data manipulation, regression analysis, and visualization of results, such as percent variance explained and scores plots. Additionally, it covers cross-validation techniques to assess the model's predictive capability.

Uploaded by

madadi morad
Copyright
© © All Rights Reserved
We take content rights seriously. If you suspect this is your content, claim it here.
Available Formats
Download as PDF, TXT or read online on Scribd
You are on page 1/ 6

Lazzara Lab

Samantha Clayton
April 14, 2019

Partial Least-Squares Regression (PLSR) in MATLAB R2018a

Importing Data into MATLAB

1. Click on the Home tab in Matlab. Press the “Import Data” button and select the dataset you would like to use.

2. The dataset will open onto a screen. Select the data you would like to use then press the “Import Selection” button.
The output type options are table, column vectors, numeric matrix, cell array, or string array depending on what you
are importing. I suggest using a string array for this process to retain the column labels (if your dataset has them) and
to more easily manipulate them. If you select something other than a table, make sure that all the rows and columns
you wish to import are selected, as MATLAB may not select them by default.

a. You can also click on the arrow on the Import Selection button to get a drop-down menu. You can click on
generate script or function to auto-import this if you need to run it again.
b. Or you could save the workspace after importing all of the files you need and load the workspace rather than
having to import the data again if you need to run this a separate time.

save(plsr_workspace) % saves workspace as a .mat file with the given filename


load(plsr_workspace.mat) % loads workspace into MATLAB
Lazzara Lab
Samantha Clayton
April 14, 2019

3. The dataset will be imported into MATLAB as the data output type you selected with the same name as the original
file. If the data is in two separate files, repeat the previous steps for the second file. The remaining steps and sample
code outlined below are written assuming that the data is imported using string arrays. Separate the data and labels:
X = str2double(Xmatrix(2:end, 2:end));
Y = str2double(Ymatrix(2:end, 2:end));
X_col_labels = cellstr(Xmatrix(1,2:end));
X_row_labels = cellstr(Xmatrix(2:end,1));
Y_col_labels = cellstr(Ymatrix(1,2:end));
Y_row_labels = cellstr(Ymatrix(2:end,1));

4. If x and y data are in the same table, allocate the x and y portions of the array into two separate arrays:
Data = str2double(data_matrix(2:end,2:end));
X = Data(:,1:last_x_col)
Y = Data(:,first_y_col:end) %last_x_col and first_y_col need to be defined based on your input
X_col_labels = str2cell(Xmatrix(1,2:end));
Y_col_labels = X_row_labels;
X_row_labels(data_matrix(2:last_x_col,1)
Y_row_labels(data_matrix(first_y_col:end,1)

Performing the Regression

5. The data needs to be normalized before using PLSR. You can use the zscore function to do this.
z_x = zscore(X);
z_y = zscore(Y);

6. Code for PSLR function:


ncomp = 5; %use only 5 components → change this to desired number of components
[x_loadings, y_loadings, x_scores, y_scores, beta, pctvar, mse, stats] = plsregress(z_x, z_y, ncomp);
Lazzara Lab
Samantha Clayton
April 14, 2019
- If there are any missing values in the matrix, they must be removed before proceeding to this step.
- If ncomp is omitted, its default value is min(size(X,1)-1,size(X,2)).
- You can also specify parameter name/value pairs for cross-validation, ‘cv’ or ‘mcreps’ (See step 10 for sample
code used to perform cross-validation).

From the MathWorks documentation for plsregress:


● Plsregress computes a partial least-squares (PLS) regression of Y on X, using ncomp PLS components, and
returns the predictor and response loadings in XL and YL, respectively.
● X is an n-by-p matrix of predictor variables, with rows corresponding to observations and columns to variables.
● Y is an n-by-m response matrix.
● XL is a p-by-ncomp matrix of predictor loadings, where each row contains coefficients that define a linear
combination of PLS components that approximate the original predictor variables.
● YL is an m-by-ncomp matrix of response loadings, where each row contains coefficients that define a linear
combination of PLS components that approximate the original response variables.
● XS represents the predictor scores, that is, the PLS components that are linear combinations of the variables in X.
XS is an n-by-ncomp orthonormal matrix with rows corresponding to observations and columns to components.
● YS represents the linear combinations of the responses with which the PLS components XS have maximum
covariance. YS is an n-by-ncomp matrix with rows corresponding to observations and columns to components.
YS is neither orthogonal nor normalized.
● PCTVAR is a 2-by-ncomp matrix containing the percentage of variance explained by the model. The first row of
PCTVAR contains the percentage of variance explained in X by each PLS component, and the second row
contains the percentage of variance explained in Y. (These values are R2Y values.)
● MSE is a 2-by-(ncomp+1) matrix containing estimated mean-squared errors for PLS models with 0:ncomp
components. The first row of MSE contains mean-squared errors for the predictor variables in X, and the second
row contains mean-squared errors for the response variables in Y.
● Stats returns a structure with fields: W (a p-by-ncomp matrix of PLS weights, XS = X0*W), T2 (the T2 statistic
for each point in XS), Xresiduals (the predictor residuals, that is, X0 – XS*XL), and Yresiduals (the response
residuals, that is, Y0 – XS*YL)

Displaying outputs:

7. Follow the example code provided below to complete the percent variance in y explained by each component (R2Y):

figure;
plot(1:ncomp,cumsum(100*pctvar(2,:)),'-bo');
xlabel('Components')
ylabel('Percent Variance Explained in Y')
Lazzara Lab
Samantha Clayton
April 14, 2019

8. Follow the example code below to create a scores plot (done below for first two components):

figure;
xlabel('Component 1')
ylabel('Component 2')
hold on
x = x_scores(:,1);
y = x_scores(:,2);
scatter(x,y);
xlim([-1 1]);
ylim([-1 1]);
box on
hax = gca;
line([0 0],get(hax,'YLim'),'Color','k','LineStyle','--')
hline = refline([0 0]); hline.Color = 'k'; hline.LineStyle = '--';
labels = X_row_labels;
yline = 0;
xline = 0;
dx = 0.02; dy = 0.02; % displacement so the text does not overlay the data points
text(x+dx, y+dy, labels, 'Fontsize', 10, 'Interpreter', 'none'); % labeling points
Lazzara Lab
Samantha Clayton
April 14, 2019

9. Follow the example code provided below to create a loadings plot:

This portion deals with duplicate values in the data before labeling. It is not necessary if there are none.

matches = zeros(max(size(x_loadings(:,2))),max(size(x_loadings(:,2)))); % allocate matrix


if size(unique(x_loadings(:,1:2))) ~= size(x_loadings(:,1:2)) % check if there all values are unique
[C,ia,ic] = unique(x_loadings(:,1:2),'rows'); % getting all unique values
end
duplicate_ind = setdiff(1:size(x_loadings, 1), ia); % get indices of values that were repeated
duplicate_value = x_loadings(duplicate_ind, 1:2); % storing duplicate values

%% Plotting loading values


figure;
x1 = C(:,1); %if portion above is not used, replace C with x_loadings
y1 = C(:,2);
x_dup = x_loadings(duplicate_ind,1);
y_dup = x_loadings(duplicate_ind,2);
scatter(x1,y1,'o')
xlabel('Loading 1')
ylabel('Loading 2')
xlim([-3 3]); ylim([-3 3]); %adjust these limits as necessary to include all data points on your plot
box on
hax = gca;
line([0 0],get(hax,'YLim'),'Color','k','LineStyle','--')
hline = refline([0 0]); hline.Color = 'k'; hline.LineStyle = '--';
cell_col = X_col_labels(ia); %if portion above is not used remove ia and just leave X_row_labels
Lazzara Lab
Samantha Clayton
April 14, 2019
yline = 0; xline = 0;
dx = max(x_loadings(:,1))/50;
dy = max(x_loadings(:,2))/50; % displacement so the text does not overlay the points
text(x1+dx, y1+dy, cell_col, 'Fontsize', 10, 'Interpreter', 'none','Color','k');

The following portion shifts the labels on points with the same value. If there are no duplicate values this portion is
again not necessary.

for i = 1:size(duplicate_value,1) - 1
text(x_dup(i+1)+dx, y_dup(i+1)+dy-2.5*i*dy, X_col_labels(duplicate_ind(i+1)), 'Fontsize', 10, 'Interpreter',
'none','Color','k')
end

Cross-validation

10. To perform a PLSR calculation with cross-validation, modify the plsregress command accordingly:

ncomp = 5; %use only 5 components → change this to desired number of components


fcv = 5; %perform PLSR with 5-fold cross-validation -- this is an integer value, up to the total number of
%rows in the input data sets (for this sample data set, can be up to 8)
mcreps = 1; %number of Monte-Carlo repetitions used during cross-validation

[x_loadings, y_loadings, x_scores, y_scores, beta, pctvar, mse, stats] = plsregress(z_x, z_y, ncomp, 'CV',
ncv,’mcreps’,mcreps);

%Calculate Q2Y values for each model component using mean squared errors from cross-validation. For %this particular
data set, the Q2Y values will be negative, indicating that the PLSR model developed here %is not a predictive one.
Q2Y = 1- mse(2,2:end)/sum(sum((z_y-mean(z_y)).^2)./size(z_y,1));

You might also like