CSCI 3202: Intro to Artificial Intelligence

I've posted the solutions for the final here:
http://www.piazza.com/class_profile/get_resource/h6awgtogngg3r3/hb64avrw9hi27f

You can also view it on the course page: https://piazza.com/colorado/fall2012/csci3202/resources

The final exam grades and P4 grades are up on D2L now, as well as late submissions for the other projects.

I also added curves to both the final exam (adding 12 "bonus" percentage points) and the quizzes (11 percentage points).

So the "final calculated grade" should be accurate now. Send me email if you see any mistakes.

I appreciate all your hard work! I'm very glad to have had the opportunity to teach the class. Best wishes for the future.

#class-resources

#final #grading #projects

After checking some more I discovered that the actual deadline for csci 3202 grades is not until next Wednesday, due to the weekend and holidays. So I went ahead and extended the P4 deadline until tomorrow 12-20, 5pm, last day of exams, which might help some of you who are still not done.

#p4 #grading

Final Exam in *ECCR 200*, on *Tuesday* 2012-12-18 04:30 PM - 07:00 PM

The final will be cumulative. But there will be more emphasis on the new material following the 2nd quiz. So about 50% will cover the material starting with Reinforcement Learning, and 50% on the earlier material.

It will be closed-book, but you can bring two sheets of paper with notes.

A calculator should not be needed, but you may feel more comfortable bringing a basic (non-programmable) calculator.

The Berkeley CS188 site has a lot of practice exams and a listing of topics to review. Of course we didn't get particle filters, or the Machine Learning section.

• http://www-inst.eecs.berkeley.edu/~cs188/fa11/final.html

The last class will be a review.

#final

See D2L for all the grades to date. I've also made the rest of it a bit more consistent, exposing class averages. Note that the latter include zero scores for no-shows that weren't included in the averages I've shared in class, so they are a bit lower.
Send me an email if something seems amiss.

And remember you can still re-submit previous projects and get 1/2 credit for incremental improvement over your original score.

#grading #p2 #p3

I've posted the official version of Project P4 via the course page: https://piazza.com/colorado/fall2012/csci3202/resources, and set up a dropbox for it. It is due Wed Dec 19th at 17:00. The direct link is:

https://csel.cs.colorado.edu/~nemc6333/cs3202/projects/tracking/busters.html

It is pretty much the same as the version that was up before, with slightly different commentary. But the points have been revised so that you get full credit (20 points) for the first three questions without doing the particle filtering questions. Note: if you really want to learn particle filters on your own, you can get up to 5 points of extra credit.

#p4 #projects

Quiz 2 grades have been posted to D2l. The average was 34.7.

I've posted the Quiz 2 solutions here:

http://www.piazza.com/class_profile/get_resource/h6awgtogngg3r3/h9fu2hrl3cs7jp

You can view it on the course page: https://piazza.com/colorado/fall2012/csci3202/resources

#class-resources #homework

#quiz2

I've posted the official version of Project P3 via the course page: https://piazza.com/colorado/fall2012/csci3202/resources, and set up a dropbox for it. It is due Thu Nov 29th at 17:00. The direct link is:

https://csel.cs.colorado.edu/~nemc6333/cs3202/projects/reinforcement/reinforcement.html

It is pretty much the same as the version that was up before, but has some extra clarifications, and an extra testing option to use a custom "hardcoded" MDP. The example MDP supplied there comes from the second quiz.

#projects

I've found that the easiest way to do value iteration by hand is often to draw a tree of the MDP first. Here is a partially-filled-in tree for the Soccer MDP value iteration exercise we did in class on Oct 24. Use it as a supplement to the solution online at http://www-inst.eecs.berkeley.edu/~cs188/fa11/section_handouts/section5soln.pdf

To the right of each node are the value iteration results for i=0, 1 and 2. The nice thing about the tree form is that you can record the values as you go along, and you just have to look at the immediate children of a node to update the value of that node, using either "max" for the states, or "expected value" for the q-states. E.g. the values for the top node V(1) (not filled in on the blackboard) are simply the max of the corresponding iterations of the q-states below them. 1: 1/6 and 2: 1/4.

Note that the M and G nodes are terminal nodes, and thus always have zero value.

Note that the values from successive iterations are approximations which will converge over time to the true value.

Note that the value of the iteration value "i" decreases as you travel down the tree.



#homework