1

The picture of the algorithm

The paper about SAMME.R algorithm

Firstly,in the 2a step, if fits a classifier T(x) to the training data using weights,but I don't know how the algorithm uses the classifier T(x) in the following part.

Secondly, in the 2b step, i don't know how to obtain the weighted class probability estimates. It just says we can use decision tree to estimate the probability, but I don't know how to do it.

Thanks in advance. My English is poor and my question may be vague. I am really sorry for this. If you can't understand my question, just comment it and I will try my best to expound my question clearly! Thank you very much!!

Song Wang
  • 11
  • 3
  • 1
    "I can not understand" is really way to vague. Aswell did you even _once_ re-read your question? "enter image description here" - really? I don't even want to edit your question. – Xatenev Dec 11 '17 at 10:56
  • I am sorry for this. I just edit my question! sorry! – Song Wang Dec 11 '17 at 11:45
  • I added the tag [machine-learning] to your question. Might attract more potential answerers since [adaboost] isn't a heavy used tag on this page. – Xatenev Dec 11 '17 at 11:55

1 Answers1

0

I also found myself contemplating on the same problem not too long ago. For whatever it's worth, here is my opinion on the matter:

Step 2a

Train a DecisionTree or any other classifier that can supply probability estimates. You can find an interesting article on estimating probabilities with DecisionTrees here. This classifier will be used in step 2b.

Step 2b

One way to look at this if you expand the formula:

formula

In words, to compute the weighted probability for some label (i) you multiply the probability estimated at the previous step for label i with the sum of weights of the samples that have the label i. In practice, the classifier at step 2a may use the weights in some other way and at the end only supply the weighted probability estimates. A nice post on this for Decision Trees is here.

I hope you find this answer helpful!

Vali Rosca
  • 479
  • 2
  • 4
  • 18