Seasons Of Code

Adaboost    • Riddhish Bhalodia   

Adaboost

Adaboost

Boosting is a well known machine learning technique, we use simple weak classifiers in cascade fashion to form a strong classifier. It’s extremely effective, facebook uses some version of this algorithm for detecting faces (99.9% accurate). Implement basic adaboost on simulated data, then for digit recognition.

Again implementation in python using opencv and must follow blockwise execution and tutorial format.

Week 0

  • Read up on the Adaboost algorithm.
  • Study mathematical theory.
  • Start implementation of basic Adaboost algorithm in Python.

Week 1

  • Finish code for Adaboost in Python.
  • Vectorise the code using numpy. Speed up of about 10x.
  • Read up on digit recognition via Adaboost.
  • Begin writing code for digit recognition.
  • Write Ipython notebook for Adaboost.

Week 2

  • Acquire training and testing data(images and labels).
  • Complete digit classification code.
  • Test the digit classifier on data.
  • Debug and optimise output and successfully implement the digit classifier.
  • Add digit classification to the Ipython notebook.

The project was completed in a little less than three weeks. The repo for the project can be found here. The rendered iPython notebook can be found at the official iPython Wiki or can be directly accessed here.