A=bagging(A,H) returns a bagging object initialized with algorithm A and
hyperparameters H.
Bagging trains BAGS classifiers on resamplings with replacement of the data
of size M. The final classifier is the averaged classifier.
For classifiers that output a real-valued output in pattern recognition this
gives you the choice of performing the average before or after taking the
sign.
You can also use the bagging object to average the results of already
trained classifiers, simply define BAGGING(A) where A is a GROUP of
classifiers
Hyperparameters (with defaults)
child=svm -- the algorithm to bag
bags=10 -- the number of times to bag
m=500 -- the number of points to sample for each bag
(can also be a fraction of the training data)
Example:
a1=svm; a1.C=10;
a=bagging(a1); a.bags=10; a.m=20;
[r a]=train(a,toy2d);
loss(test(a,toy2d))
| |
Reference : Bagging Predictors |