Changes between Initial Version and Version 2 of Ticket #1795


Ignore:
Timestamp:
05/21/14 17:48:36 (7 years ago)
Author:
gkronber
Comment:

Legend:

Unmodified
Added
Removed
Modified
  • Ticket #1795

    • Property Priority changed from medium to low
  • Ticket #1795 – Description

    initial v2  
    11It would be nice to support a kind of boosting where multiple models are learned step by step and the weight of observations is adapted based on the residuals of the models learned so far. See [[http://en.wikipedia.org/wiki/AdaBoost|AdaBoost]].
     2
     3AdaBoost with sampling instead of weighting could be implemented for classification problems.
     4
     5Friedmans "Stochasic Gradient Boosting" (1999) could be implemented for regression and classification problems.
     6
     7One important question is where the gradient for our target functions is implemented. In gradient boosting we need to use the correct gradient which is appropriate for the currently selected fitness function.