5. X2 2.0 1.0- t + 1.0 3 4 5 2.0 to 7 x1 (a) Assume...

Transcribed Image Text:

5. X2 2.0 1.0- t + 1.0 3 4 5 2.0 to 7 x1 (a) Assume we want to use Adaboost to classify the training examples S in the 2D plane given in the figure above. The weak learner outputs a function of the form sign(x; -t) fori {1,2} and t {1,2} or its negation. You can assume {+1, -1} labels. If you recall, at the beginning of Adaboost, the weight for each training example is the same, so w = (1, 1,..., 1). What is the best hypothesis for the weak learner to output to minimize the error rate with respect to this initial weighting? What is its error rate? Call it E. (b) Following the Adaboost algorithm, we must now give new weights to each point. What is the new weighting w (recall = )? 1-E Now according to this new weighting of points, what is the best hypothesis for the weak learner to output? 5. X2 2.0 1.0- t + 1.0 3 4 5 2.0 to 7 x1 (a) Assume we want to use Adaboost to classify the training examples S in the 2D plane given in the figure above. The weak learner outputs a function of the form sign(x; -t) fori {1,2} and t {1,2} or its negation. You can assume {+1, -1} labels. If you recall, at the beginning of Adaboost, the weight for each training example is the same, so w = (1, 1,..., 1). What is the best hypothesis for the weak learner to output to minimize the error rate with respect to this initial weighting? What is its error rate? Call it E. (b) Following the Adaboost algorithm, we must now give new weights to each point. What is the new weighting w (recall = )? 1-E Now according to this new weighting of points, what is the best hypothesis for the weak learner to output?

Answer rating: 100% (QA)

a The best hypothesis for the weak learner to output to minimize the error rate with respect to the initial weighting W1 111 is the hypothesis that co... View the full answer