Day #165 - Adaboost

Key Summary

• Combine all ML methods to make a strong classifier
• 1NN, KNN, SVM, Neural Nets, Naive Based Classification, ID Trees - Weak or Imperfect classifiers
• Combine them all together or some set of them (Ensemble or Aggregate Classifier)
• More voting power when it classifies correctly
• Negative voting power when it classifies wrongly
• Pick the best weak classifier, Assign it voting power
• Best - Classifier that makes fewest errors
• Calculate Error rate for each of weak classifiers
• Update weights to emphasise on points misclassified
• Take misclassified points increase weights in the next round
• Adding new weights so that it becomes one half (For misclassified points)
• Classifiers may make overlapping errors

Summary

• Initialize weights
• Calculate error for each h
• Pick the best h with smaller error rate
• Calculate voting power for best weak classifier
• Update weights to emphasize on points misclassified

This 165 post beats my all-time record of posts in a single year :). I hope to break a lot more useful learning in AI / ML / Research papers before the end of the year.

Knowledge is wealth, Patience is the character, Perseverance is Strength. Keep Learning and Keep Going!!!


Happy Mastering DL!!!