Artificial Intelligence

Dear students,
by now, you should have received the results for projects 1-3. Today, an incident occurred where the HFF Bonus mail was mistakenly sent out again. Please disregard this email. If you encounter any other issues, please contact Pascal Lauer via email.

The exam registration is now open. If you wish to participate in the main exam, please follow these steps:

1. Register in LSF by February 20, 2024, 23:59.
2. Register in CMS by February 22, 2024, 23:59.

Please note that you need to have accumulated 30 points in the projects to register in CMS. If you are still short of points for registration and your current daily tests suggest that you will receive them for project 4, you can contact Pascal Lauer via email to register manually. Please be aware that we can only add these points to CMS only after the deadline on Tuesday.

Dear students,
the results for the HMax/HFF project are now public. Please note that we evaluated your project separately for each subtask (HMax and HFF) as they had separate (extended) deadlines. So you will receive two emails, one for each evaluation.

If you encounter any problems regarding the final evaluation, please contact Pascal Lauer via email.

Dear students,
the results for the BlindHeuristic/GoalCount project are now public. We will proceed with a final test run for the other projects. You can expect the results of the HMax Project by the end of the week, followed by the evaluation of the subsequent project(s).

Please note that we encountered issues with sending emails to external servers and thus all emails will now be sent to your student mail. If you encounter any problems with the evaluation results or nightly tests, please contact Pascal Lauer via email.

Hi all,

this is just to draw your attention to small changes in the LAO* pseudo-code (Chapter 14) and the epsilon-greedy policy (Chapter 16):

• The LAO* pseudo-code previously allowed re-insertion of already expanded states into Frontier, which is wrong.
• The epsilon-greedy policy did not correctly handle the case where there is only one applicable action; also, in the random-exploration case, all actions are considered including the one taken by the policy.

Sorry for this inconvenience. I thought it better to correct these definitions posthoc, after the lectures, than to leave the bugs in.

best,

Jörg Hoffmann

Hi all,

my sincere apologies that the online lecture today was abruptly interrupted at the end.

The laptop battery was empty -- even though it was plugged in and showing me the "being loaded" symbol. No idea what went wrong; never happened before, 10min later everything works correctly again.

What got cut off is only the MCTS example run for slide 44. I have uploaded a PDF showing that example run.

Sorry again,

best,

Jörg Hoffmann

Hi all,

I need to take care of relatives in Bavaria for a few days next week (father in law is very sick).

Hence the lecture next Tuesday, Jan 23, will be online only.

In zoom: https://cs-uni-saarland-de.zoom.us/j/98696736217

You'll be invited to ask questions (and give answers to my questions) in the zoom chat.

My apologies for any inconveniences caused. The lecture next Friday will again be in GHH as usual.

best,

Jörg Hoffmann

 

 

Dear students,

we received numerous complaints about missing accuracy reports for your models in the daily tests. This was unintentional and is now fixed. The next daily test run will happen unconditionally, whether you pushed a change or not, and will include one log per task in which the best of your models and its accuracy on the test data will be reported (even if the test is passed). For the next project, we also hope to include some error logs for failing tests which should make it easier for you to find the source of the bugs.

Furthermore, a section regarding reproducibility has been added to the project description (section 4.1). To ensure that the submitted models are indeed yours, we must be able to generate them using the provided generation script. Please make sure that once the project deadline approaches, all your models can be deterministically generated from scratch by calling this script.

Lastly, a subtle but important detail was added regarding the softmax layer that you must append to your final model. Please choose dim = -1 for the softmax layer, which chooses the last dimension as the reference dimension for softmax. This allows us to evaluate your model for both single states and batches of states.

Due to the issues and changes above, we decided to extend the deadline of the project by three additional days, i.e., until January 26th.

Regards, Thorsten

Dear students,

it was brought to our attention that there is an inconsistency between the project specification for hFF and the tests. Specifically, it it checked in your tests whether your delete-relaxed plan applies extra operators after already reaching the goal state, rejecting in this case. The project description did however not mention this as a requirement for the returned plan. With pathological tiebreaking between best supporters and chosen actions during plan extraction, it is in general possible to violate this constraint and still be consistent with the specification.

We have now fixed this issue so that the check matches the specification. This affects the public tests as well, so please merge with the fork repository by running git pull fork main. We are sorry if we wasted your time by forcing you to fix an implementation that should have already worked according to the specification.

Additionally, we received some reports by students that received daily test emails inconsistently. To rule out any disadvantages that may result from this and the previously mentioned testing issue, we decided to extend the deadline for hmax until Friday, December 15th, and the deadline for hFF until Tuesday, December 19th. Note that if your implementation already passes all tests, there is nothing left to do for you.

Regards, Thorsten Klößner

Dear students,

we have three announcements to make regarding the projects.

Disabled Daily Tests in Project 1

We received numerous reports targetting three particular daily tests in project 1 in which operators with an empty precondition occur. While the relevant algorithms shown in the lecture technically work properly in presence of empty preconditions, their handling is very subtle. Because of this subtely, and especially because there are no public tests covering this edge case, we decided to disable these daily tests and convert to (secret) bonus tests instead, each worth 1 bonus point if passed (you still need to pass the public tests). The following daily tests are affected:

HMaxDailyTest.HeuristicEvalTest.check_heuristic_estimate.2
HMaxDailyTest.SearchTest.test_astar_statistics.2
HFFDailyTest.RelaxedSolvableTest.check_relaxed_plan_and_estimate.0

 

Deadline Extension for Project 2

Secondly, we decided to extend the deadline for project 2 from December 22nd to January 9th. This means that the deadline for project 2 will coincide with the beginning of project 3.

 

CMS Results for Project 0 and following

Entering the points for the projects into the CMS will unfortunately still take some time and we should be able to do so in the next couple of weeks. However, you can refer to the final daily test email you received for the project to calculate the amount of points you will be awarded yourself. To remind you, if a project task has X points and you pass Y% of all daily tests for this task, then will you obtain X * Y% points for this task, under the condition that you pass all public tests for the respective task.

 

Regards, Thorsten Klößner

Dear students,

since we have witnessed unintended use of the State interface within some project implementations, we have refined this interface a bit by restricting access to those functions not intended to be used. Please update your project by running git pull --no-edit fork main. The change potentially makes your code no longer compile if you use one of the functions, which is intended because you should not use them and may cause unintended results.

Regards, Thorsten Klößner

Dear students,

the description for project 1 has been released. You can find it in the Materials section. The project will go on for 2 weeks. Please start as early as possible.

Note that the deadline for project 0 is today, 23:59. Make sure that you push your implementation to the main branch of the repository until said deadline. We will not evaluate other revisions.

Regards, Thorsten

 

Dear students,

we fixed some issues reported to us with the project build for Mac users with Apple Silicon processor and an inconvenience appearing for Windows users. Please pull the latest changes by running 'git pull --no-edit fork main', if you belong to the aforementioned groups, otherwise feel free to skip this.

In particular, the following issues were fixed:

• Windows: Incompatibility between build script and VSCode settings. This caused a build triggered by the build button in VSCode to fail after initially building with the build script build.py, and vice versa. After pulling the changes and removing the builds folder, both methods of building the project should now be interchangable.
• MacOS with Apple Silicon processor (in particular the M1 series): Errors during linking due to incompatible processor architecture should no longer occur. A workaround was provided in a forum post, but is now no longer needed, everything should work out-of-the-box. If you are affected by this, remove the builds folder and re-build the project after pulling the changes.

Regards, Thorsten Klößner

Dear students,
we will release the setup instruction for the project software after the lecture today in the Materials section. On Tuesday, there will be a short demonstration of the setup, followed by an introduction to the projects. Thorsten will show you how to work on the projects in Visual Studio Code, how to build them, how to test and debug your implementations and give an overview over the software. Please read the description before the demonstration and set up your system following the provided instructions as best as you can. ( The description is neither supposed to be a lot nor supposed to be complicated. Doing this on Monday gives you plenty of time. Enjoy your weekend. :) )

 

An important note is that we use the Saarland Informatics Campus (SIC) Gitlab to grant you access to the repositories. If you've never used this Gitlab before, please sign in to https://gitlab.cs.uni-saarland.de/ to activate your account. Your username is the beginning of your student email address. For example, if your name is Konrad Klug and you were assigned the email kogl00001@stud.uni-saarland.de, your username is "kogl00001". Your password is the one you've set for your SIC account. If you're unsure or confused, you can visit the password reset page: https://sam.sic.saarland/accounts/password/reset/ A quick note: These services are quite useful. Every student, for instance, can create unlimited repositories on the aforementioned Gitlab or use Nextcloud. More information can be found here: https://sam.sic.saarland/

If you have already logged in once into the Gitlab you should have just received access to your repository. If you did not, access should be granted after a few hours of your initial sign in. This is a scheduled job running on our server. If it does not work immediately, you should stay patient. If after a day you still do not have access, please contact lauer@cs.uni-saarland.de . For further trouble shooting there will be office hours starting next week on Thursday, ofcourse you can also always use the forum and remember to make use of Thorten's demonstration.

In case you are curious, the initial state of your project repository can be found here: https://gitlab.cs.uni-saarland.de/fai-public/teaching/ai/project-skeleton

 

Good luck with the projects and happy dancing on the weekend. 💃🪩🕺

Dear students,
we would like to clarify that tutorials will start next week Monday, November 13th. If you wish to switch your tutorial slot, you have the option to do so by finding a fellow student who is willing to make the switch. Both you and your partner should send send an email to lauer@cs.uni-saarland.de to initiate the process.

For a list of available tutorial slots, please visit the team page at: https://cms.sic.saarland/ai_23/tutors/

Additionally, there is a dedicated forum post where you can find potential partners for tutorial slot exchanges: https://fai-forum.cs.uni-saarland.de/c/artificial-intelligence-winter-2023-2024/tutorial-exchange/69

 

 

Dear students,

tutorial sheets for chapters 3 and 4 are now available in CMS. Both of these sheets will be discussed in tutorials next week. The chapter 4 on heuristic search will be finished in Friday's lecture, but the slides for the whole chapter 4 can already be found in CMS. The tutorial slots will be published tomorrow as you have still time to set your preferences until midnight.

Best,

   Dan