0
$\begingroup$

I've been reading up on boosting algorithms.

I understand that the main crux of the algorithm is to build weak classifiers that are slightly better than random guessing, and then to add them up so that we end up with a strong classifier. This is done via "boosting rounds".

With different papers using different variables, like k, n, m, etc, it's gotten a bit confusing.

I just want to confirm, if I run a boosting algorithm for, say 150 rounds, is that equivalent to saying that I'm training 150 weak classifiers? I mean, is a weak classifier the output of one boosting round?

Improve this question
$\endgroup$

1 Answer 1

Reset to default
1
$\begingroup$

I went back to the original paper from the literature by Schapire (http://www.cs.utah.edu/~piyush/teaching/brief_intro_boosting.pdf) and it does sound like a weak classifier is the output of a boosting round. Consider how the output of the overall classifier is $$H(x)= \text{sign}\left(\sum_{t=1}^T\alpha_th_t(x)\right)$$ which is expressing a threshold for the sum of $T$ weak classifiers.

Improve this answer
$\endgroup$
1
  • 1
    $\begingroup$ Thanks! Also, looking through the code of scikit-learn's AdaBoost implementation, here: github.com/scikit-learn/scikit-learn/blob/master/sklearn/…, I came across this on line 282, about the n_estimators parameter: n_estimators : integer, optional (default=50). The maximum number of estimators at which boosting is terminated. So n_parameters is clearly the number of rounds, and also the number of estimators (weak classifiers). $\endgroup$
    – user961627
    Commented Apr 10, 2014 at 7:52

Your Answer

By clicking “Post Your Answer”, you agree to our terms of service and acknowledge you have read our privacy policy.

Not the answer you're looking for? Browse other questions tagged or ask your own question.