The Treebagger give different results in 2012a and 2013a

I used

B = TreeBagger(NTrees(j),train_feats',train_labels');

and

 Y = predict(B,test_feats');

to classification, in train_feats each column is a sample,the train_labels total of 30 categories,label from 0 to 29.

when I run the code in matlab 2012a the accuracy almost 90%,but when i update to 2013a the accuracy is less than 1%. The data and the code are intact, why the result is so different?

 Does anyone have an explanation?

Answer 1

[Tom Lane]

This might be the explanation, and it includes a suggestion of how to avoid the problem:

http://www.mathworks.com/support/bugreports/927692

[xq]

Thank you so much for your answer and advice. The answer is exactly the problem. I change the labels from ordinals to chars:

NewLabels{i} = num2str(labels(i));

problem solved!

Thank you again for your help.