MATLAB: Replacement for ‘eval’ function

classificationtreeclassregtreeerrorevalfitctreepredictpruneremovedStatistics and Machine Learning Toolboxsubtrees

I used to use 'eval' in Matlab 2017, but this triggers an error in 2018. I have looked in the documentation and user group, but found no solution. Can you help ?

Copy from command window:

Error using eval
Too many input arguments.

Best Answer

That "eval" function was a method of the "classregtree" class which was removed in MATLAB R2018a. I noticed that you are now using the "fitctree" method instead of the removed "classregtree".

The trees from "fitctree" are part of the "ClassificationTree" class. A similar method to "eval" is in that class. It is called- "predict" and it predicts labels using classification tree. The third input in "predict" is slightly different than "eval", so you would need to change the inputs slightly.

Instead of:

>> eval(T, X, s)

The "predict" function should be used with the 'Subtrees' Name-Value pair.

>> predict(T, X, 'Subtrees', s);

For more information on the "ClassificationTree" class and the "predict" function, see the documentation links below.

https://www.mathworks.com/help/stats/classificationtree-class.html

https://www.mathworks.com/help/stats/compactclassificationtree.predict.html

Related Solutions

MATLAB: Is the Score Transform lost when using “saveCompactModel”

One of the limitations of using "saveCompactModel" to save an SVM model with capabilities to predict posterior probabilities is that "loadCompactModel" cannot restore the 'ScoreTransform' property to the MATLAB Workspace.

However, while preforming code generation, "loadCompactModel" is able to load the model along with the 'ScoreTransform' property at compile time within an entry function, and hence should function properly.

For more information on such limitations on various model objects while using "loadCompactModel", please refer to the documentation link below:

https://www.mathworks.com/help/stats/loadcompactmodel.html#bvcl05n-1-CompactMdl

MATLAB: How to deploy a model created by Classification learner as an standalone application

To deploy the model, you would have to save the model from Classification Learners App. Please follow the steps below:

1. Once you have trained the model in the app, save that model using "Export Model" function from the app. This should generate a struct variable for the model in MATLAB workspace. Let it be "trainedModel"

2. Save the "trainedModel" struct into a "MAT" file. Lets say, "model.mat"

3. Write a "M" file to be deployed as a standalone application which should perform the following steps:

Load "model.mat"
Convert the input data from excel into "table" as model requires table datatype to execute.
Extract "ClassificationTree" from the model.
Perform a call to "predict" function passing "ClassificationTree" and the data from excel as Input parameters.
The call to predict function would return a prediction based on input data and the trained model. Return this value.

NOTE:

You cannot use "predictFcn" which is inside the "trainedModel" in a deployed target.
Since there are more than one predict function in all the MATLAB toolboxes, you would have to use "function pragma" to let the dependency analyzer know which "predict" function to use. In this case we have to use "predict" function from "Classification Tree" method.
The input data requirements may differ based on the model being used. In this case, the model requires data to be in form of table.

Even after including the ClassificationTree, the following warning message was received:

“Warning: Could not find appropriate function on path loading function handle <C:\Program Files\MATLAB\R2017a\toolbox\stats\mlearnapp\+mlearnapp\+internal\+model\DatasetSpecification.m>@(x)exportableModel.predictFcn(predictorExtractionFcn(x))>”

The error message is not clear enough to include the DataSetSpecification function. An enhancement request is raised for the same.

% This function takes "ValidationData" as an input from Excel and
% returns "yfit" which a prediction based on the "trainedModel" from "model.mat" 
function yfit = runModelFromMATLAB(ValidationData)
%# function CompactClassificationTree
%# function ClassificationTree
%# function mlearnapp.internal.model.DatasetSpecification % declares which "predict" function to use
load model.mat; % trained model from Classification Learners App
ModTableData = ValidationData;
% convert data to table
ModTableData = cell2table(ModTableData, 'VariableNames',...
  {'status','trial_devcount','errorcnts','trial_avgviewedmins',...
   'trialviewedmins','trial_avgbitrate'});
% get classification tree from the model and call predict using this tree
ModelClassTree = trainedModel.ClassificationTree;
yfit = predict(ModelClassTree, ModTableData);

You can refer the following documentation.

・Deploy Predictions Using MATLAB Compiler

https://www.mathworks.com/help/stats/export-classification-model-for-use-with-new-data.html#mw_34971269-0a8e-4ea6-a27c-77ad81b8c82e

Best Answer

Related Solutions

MATLAB: Is the Score Transform lost when using “saveCompactModel”

MATLAB: How to deploy a model created by Classification learner as an standalone application

Related Question