AdaBoost $($Adaptive Boosting$)$ is another approach to the ensemble method field.

It always uses the entire data and features $($unlike before$)$ and aims to create $T$ weighted

classifiers $($unlike before, where each classifier had same influence$).$ The new

classification will be decided by linear combination of all the classifiers, by:

$g(x)=$sign$({\displaystyle \underset{t=1}{\overset{T}{}}}{}_t{f}_t(x)),{}_{t}0$

Consider the following dataset in $R^2$ :

The first decision stump is already drawn, the arrow

points in the positive direction. Calculate the

classifier error $({}_1)$ and weight $({}_1).$

Calculate the new weights of the samples $($and

normalize them to get valid distribution$).$

Draw the second decision stump. Reminder: the decision stump $($our classifiers$)$

are parallel to $\frac{x}{y}$ axis.

Without calculations, which classifier's weight is larger, ${}_1$ or ${}_2?$ Explain why.

In the right image, there is the dataset and the weights for each point, after finding

the third decision stump and calculating the new weights. Which of the following

$($green or blue$)$ is the correct third decision stump?

Given ${}_2=1\mathrm{.}1,{}_3=0\mathrm{.}62,$ draw the full

classifier, like in slide $13.$

What is the train accuracy?