SCML Newsletter (May 16th 2022)

This is the first SCML newsletter for the 2022 iteration of the competition.

SCML is one of the leagues of ANAC 2022 one of the official competitions of IJCAI 2021.

Important Request 🙏🙏🙏: Please upload an early version of your agent to the online competition website AS SOON AS POSSIBLE (hopefully before May 30th). This will be of tremendous help for us (in estimating the computational capacity we need) and yourself (by providing you with feedback about any issues in your agent as early as possible). From our past experience, early and repeated submissions to the online competition website helps tremendously in finding bugs that you many not notice otherwise.

There are no upper limits on the number of times you can upload your agent. It is OK if you update it with every change you make.

  1. Due to a problem with our servers, agents uploaded before May 14th were NOT TESTED. Please re-upload your agent if your last upload was before that date.

  2. We released new versions of NegMAS [0.9.4] and SCML [0.5.4]. The new versions contain bug-fixes and quality-of-life improvements. All participants must upgrade to the newest version. Please upgrade your installation by running:

    pip install -U negmas scml

  3. If you are using our visualizer, please upgrade to the latest version as well. The visualizer provides a simple means to understand what happens in a specific world simulation or a tournament. If you are not using the visualizer, consider starting to use it. Once installed, it should be able to visualize existing simulations as well as new ones. You can do that using:

    pip install -U scml-vis

  4. Remember that you have access to all agents from previous iterations of the competition and can use any of them while training/debugging your agent. These agents are contained in the scml-agents package which you should also update to the latest version if you plan to use:

    pip install -U scml-agents

  5. Related to debugging agents, please make sure to check our tutorials, youtube tutorials and specially the tutorial about running controller experiments. Using the techniques in this tutorial (combined with the visualizer), you can test your agent in highly controller experiments to understand how to improve its performance.

  6. There is a small change in the timing of negotiations in the OneShot track compared with last year which is explained in details in Section 3.1 (page 4) of the description document. In summary, negotiations now use the standard alternating offers protocol with no modifications and the first offer will be coming from agents on the same level at every simulations step. Most likely this has no effect on your agent (and all past agents work perfectly with it) but it makes debugging and thinking about negotiations simpler and ensures that agents based on OneShotSyncAgent are receiving all offers in every call to counter_all() which was not guaranteed earlier. It also speeds up the simulation.

  7. Please make sure to periodically check our announcements on the official competition website and be sure to add our email to your contacts/spam filter to avoid missing important announcements.

  8. We added better type annotations to the skeletons for oneshot and std-monolithic. The changes are only in type annotations and are of no consequences to your code but they make your type checker happier and provide better auto-completion if you are using an IDE.

  9. Some participants reported seeing several warnings when running std/collusion agents. These warnings can be safely ignored. To run a python script while ignoring warnings you can use:

    python -W ignore myagent.py

  10. Our community is growing:

We thank our sponsors for supporting SCML:

We also thank Brown University for hosting our website (https://scml.cs.brown.edu)

On Behalf of the SCML Organizing Committee

yasser