Homework 2 - Model-free prediction and control

This assignment will cover DS4-DS5, and you will among other things solve the Taxi- and MountainCar-problems from HW0.

Additional packages needed:

As in previous tinkering notebooks and homework, you will need the OpenAI gym package. Additionally you will need the GridWorld environment also used in previous notebooks.

The purpose of the assignment:

1. Work with some fundamental concepts via pen-and-paper exercises.

2. Implement and test methods for learning a policy from experience. 

The assignment:

Download the notebook here. Download Download the notebook here.

The instructions for the assignment is given in the notebook. 

Hand-in:

Hand in the notebook with your solutions, so that the grader can easily run your code. All code and other solutions should also be given in the form of PDF-files. If you prefer, you can hand in the code by exporting your notebook into a PDF, and then write your answers in a different PDF. (It is much easier to give feedback on your solutions if the code is available in a PDF).

Passing requirements:

Each task is awarded a certain amount of points, indicated in the notebook. According to the quality of the answer, each question will receive scores according to 3 levels: 0%, 50%, or 100% of the question's score.

For passing this assignment, you should score at least 60% of the total score of the assignment, that is, you need to obtain least 15 out of 25 points.

The grading will be done through peer-review. Instructions for the peer-review is given here.

Questions:

If you have questions, write in the discussion forum or send an e-mail to per.mattsson@it.uu.se