Why should I trust you?

Despite widespread adoption, mission critical industries remain dependent on ‘black-box’ ML approaches. This project will help you investigate methods that can shed some light into the decision making process of classifiers and allow you to experiment with such methods in realistic datasets.

Tasks

  1. Wrire a 4 page tutorial on LIME method as described in this paper. (20 points)

  2. Clone this repository and replicate the results in multi_polatiry_books dataset with random_forest algorithms. (20 points)

  3. Apply a classifier of your choice to the Tubespam dataset that was used in this paper. (20 points)

  4. Apply the LIME explainer on the classifier you implemented in the previous step for the Tubespam dataset and write a 2 page summary with your findings (40 points)

You can use Colab for this project and you must submit a Github repo. Write up tasks can be either embedded in the notebook or in separated markdown and images files.