Ensemble Learners Flashcards
Ensemble Learners
Combined group of learning algorithms
Ensemble Process
Classification: Have each learner “vote” on answer
Regression: Take mean of result from each learner
Ensemble Pros
Lower Error
Less Over fitting
The bias of each learner “fights” (counteracts) bias of others
Bootstrap Aggregating (“Bagging”)
Use same training algorithm, but train each learner on different data set
Bag Creation: Random with Replacement
Randomly select samples from training set, samples can repeat
Boosting
Focus on areas where system is not performing well
AdaBoost
increase selection chance of points for next bag that had high error in first bag
Test error on ensemble of previous learners to re-weight picks for next bag
AdaBoost Overfit
As # bags increases, AdaBoost assigns more specific data points to learners, leading to more overfit