Formatting input data for linear regression model in leave-out-one validation testing

Question

Isabelle Museck 2024-7-25，14:07

0
链接

此问题的直接链接

https://ww2.mathworks.cn/matlabcentral/answers/2140171-formatting-input-data-for-linear-regression-model-in-leave-out-one-validation-testing

评论： Umar about 12 hours 前

Hello there I have data from 10 trials stored in a 10x1 cell (Predictors) and the corespoding respose vairables stored in a 10x1 cell (Response). I am trying to trian a simple linear regression model and make predictions by leaving one trial out and using the other 9 trials to train the linear regression model and the one to predict/test the model by producing RMSE values. I am unsure of how to format my input within the "fitlm" function as I keep getting the follwing error:

% Train the network
for i = 1:length(Predictors) %iterate over all data points
    validationdataX = Predictors(i);
    validationdataY = Response(i);
%Exclude the current index (i) for training
    trainingIndices = setdiff(1:length(Predictors),i);
    traningdataX = Predictors(trainingIndices)
    trainingdataY = Response(trainingIndices)
net = fitlm(traningdataX,trainingdataY)
ypred = predict(net,validationdataX);
TrueVal = validationdataY;
TrueValue = cell2mat(TrueVal);
Predvalue = {Predval};
PredictedValue = cell2mat(Predvalue);
RMSE = rmse(PredictedValue,TrueValue)
end
Error using classreg.regr.TermsRegression/handleDataArgs (line 589)
Predictor variables must be numeric vectors, numeric matrices, or categorical vectors.
Error in LinearModel.fit (line 1000)
            [X,y,haveDataset,otherArgs] = LinearModel.handleDataArgs(X,paramNames,varargin{:});
Error in fitlm (line 134)
model = LinearModel.fit(X,varargin{:});

Any suggestions on how to fix this and to get the model to work correcly and make predictions using leave out one validation approach would be greatly appreciated!

5 个评论
显示 3更早的评论隐藏 3更早的评论

Umar 2024-7-25，16:24

Hi Isabelle,

Sounds like interesting project. In your code, you are passing cell arrays as predictors, which is causing the error.To resolve this issue, you need to convert your cell arrays to numeric arrays before fitting the linear model. However, I did update the code including leave out one validation approach. Here is updated code snippet example,

% Define and populate sample data for 'data' and 'responseData'

data = {1, 2, 3, 4, 5}; % Sample predictor data

responseData = {10, 20, 30, 40, 50}; % Sample response data

% Define and populate the 'Predictors' variable with sample data

Predictors = cell(1, length(data));

for i = 1:length(data)

    Predictors{i} = data{i};

end

% Define and populate the 'Response' variable with sample data

Response = cell(1, length(responseData));

for i = 1:length(responseData) Response{i} = responseData{i}; end

% Train the linear regression model with leave-one-out cross-validation

for i = 1:length(Predictors)

    % Extract validation data for the current iteration

    validationdataX = cell2mat(Predictors(i));

    validationdataY = cell2mat(Response(i));

    % Exclude the current index (i) for training

    trainingIndices = setdiff(1:length(Predictors), i);

    trainingdataX = cell2mat(Predictors(trainingIndices));

    trainingdataY = cell2mat(Response(trainingIndices));

    % Train the linear regression model

    mdl = fitlm(trainingdataX, trainingdataY);

    % Make predictions on the validation data

    ypred = predict(mdl, validationdataX);

    % Calculate RMSE for the current iteration

    RMSE = sqrt(mean((ypred - validationdataY).^2));

    % Display RMSE for each iteration

    disp(['RMSE for iteration ', num2str(i), ': ', num2str(RMSE)]);

 Hope, this is what you are looking for. Please see attached results.

Please let me know if you have any further questions.

Umar 2024-7-25，22:49

Hi Isabelle,

This is a common issue when working with regression models. In your case, the training data matrix trainingdataX is of size 567x541, while trainingdataY is of size 9x541. This discrepancy in dimensions is causing the error as the number of observations (rows) in the predictor matrix should match the number of observations in the response matrix. To resolve this issue, you need to ensure that the predictor and response matrices have the same number of observations. One way to address this is by transposing the trainingdataY matrix to match the number of observations in trainingdataX. Here is an example illustrating this:

% Example to transpose trainingdataY to match the number of observations in trainingdataX

trainingdataX = randn(567, 541); % Example predictor matrix

trainingdataY = randn(9, 541); % Example response matrix

% Transpose trainingdataY to match the number of observations in trainingdataX

trainingdataY = trainingdataY'; % Transpose the matrix

% Verify the dimensions after transposing

size(trainingdataX)

size(trainingdataY)

By transposing the trainingdataY matrix, you align the number of observations with trainingdataX, resolving the mismatch in dimensions and addressing the error you encountered. Please let me know if this helps resolving your problem.

Isabelle Museck 2024-7-26，11:57

Hi Umar,

Thank you so much for your response. Unfortunaltey that did not fix the isse however I think it is due to the fact that there are 63 input features that I am trying to use to predict the one contunuous variable as you can see here in the predictors and response cells:

This may be causing the differnces in dimensions seen here in the training X and Y data:

Is there a way to account for this in the code and use the 63 featuresx541 time steps for each trial to properly train and predict the continous variable in the response cell and obtain ans rmse value for each trial?

Umar about 12 hours 前

Hi Isabelle,

Try considering implementing dimensionality reduction techniques such as Principal Component Analysis (PCA) or feature selection methods to reduce the number of input features while retaining relevant information. This can help in aligning the dimensions of your input data with the response variable, potentially improving the accuracy of your predictions. Additionally, you can reshape your input data to match the desired format of 63 features x 541 time steps for each trial to ensure that the model receives the correct input dimensions during training and prediction. After making these adjustments, you can train your model using the modified data and evaluate its performance by calculating the root mean squared error (RMSE) for each trial. The RMSE value will provide insight into how well your model is predicting the continuous variable based on the input features. Hope this answers your question. Please let me know if you have any further questions.

请先登录，再进行评论。

请先登录，再回答此问题。