fitcecoc and fitclinear are suitable for categorical data ?

1 次查看（过去 30 天）

显示更早的评论

Gorkem Akgul 2021-4-23

0
链接

此问题的直接链接

https://ww2.mathworks.cn/matlabcentral/answers/810530-fitcecoc-and-fitclinear-are-suitable-for-categorical-data

评论： Jon Cherrie 2021-4-23

Hi,

I'm a newbie in Machine Learning and i'm trying to create a model on a kaggle dataset below.

Stroke Dataset

The thing is i'd like to use k-fold crossvalidation on my model but i cannot use fifcecoc and fitclinear functions because they only accept numerical values whereas i also have some categorical ones.

Is there a way to use them with categorical data or i need to use cvpartition for k-fold crossvalidation due to the fact that my dataset not completely numerical ?

Also can i use regularization with categorical variables ?

1 个评论
显示 -1更早的评论隐藏 -1更早的评论

Jon Cherrie 2021-4-23

在 MATLAB Online 中打开

You should be able to use fitcecoc with categorical predictors and k-fold cross-validation. Here is an example using the "carbig" data and k=5 folds:

load carbig.mat
t = table(Acceleration,Cylinders,Displacement,Horsepower,Mfg,Model_Year,MPG,Origin,Weight);
t.Mfg = categorical(cellstr(t.Mfg));
t.Cylinders = categorical(t.Cylinders);
t.Origin = categorical(cellstr(t.Origin));
m = fitcecoc(t,"Cylinders","KFold",5)

In this cases I get warning about unique class values (which I think means I have more folds than this data will tolerate), but I can see the categorical predictors in the output:

>> m
m = 
  ClassificationPartitionedECOC
      CrossValidatedModel: 'ECOC'
           PredictorNames: {'Acceleration'  'Displacement'  'Horsepower'  'Mfg'  'Model_Year'  'MPG'  'Origin'  'Weight'}
    CategoricalPredictors: [4 7]
             ResponseName: 'Cylinders'
          NumObservations: 406
                    KFold: 5
                Partition: [1×1 cvpartition]
               ClassNames: [3    4    5    6    8]
           ScoreTransform: 'none'