Submission

Note: Submission is now closed. All entries were submitted through CodaLab, where you can view the results:

• Track 1

• Track 2

Each competition has a practice phase and 2 main phases, as described below, for 4 possible ways to win!

Two Phases: Cross-semester and Within-semester Prediction

Note: Phase 1 and Phase 2 are now complete.

The following table summarizes the 3 phases of the competition:

See descriptions below:

• Practice: This phase will allow participants to develop their initial models and practice with the CodaLab scoring system, and give time for feedback on the competition before the competition officially begins. Since the Spring 2019 dataset has already been publicly released, we cannot officially judge predictions on Spring 2019.

• Cross-Semester: This represents the most realistic prediction task: we have data from one semester, and we want to build a model to predict on a subsequent semester. For more on differences between the semesters, see Dataset page.

  • Note: For this phase, we will release the Fall 2019 test data, including log data for the test students on early problems for this task. However, this data will not have data or labels for the last 20 problems or final exam grades (these must be predicted).

• Within-Semester: Since cross-semester prediction can be quite difficult, we will also have a simple within-semester prediction phase. However, since this track requires us to release training data for F19, we will hold this phase afterwards.

  • Note: Participants are welcome to use S19 data as well to train their models.

As a result, there will be 4 potential ways to win or place in the CSEDM Data Challenge:

1. Track 1, Cross-Semester

2. Track 1, Within-Semester

3. Track 2, Cross-Semester

4. Track 2, Within-Semester

Code Submission for Winners

For the 1st and 2nd place winners, who are eligible for prizes and publication invitations, you will be asked to submit your code for verification, along with instructions for running the code. The code submission will be fully confidential, and used only to verify how your results were obtained.

Collaboration, Publication and Sharing

A core goal of the CSEDM Data Challenge is to bring the community together to tackle a shared challenge with methods that can be shared and generalized. We particularly want to encourage collaboration across universities. Possible ways to incentivize this include:

• Associating the Data Challenge with a specific conference of journal special issues, where winning results and methods will be published.

  • E.g. JEDM, IJAIED, TOCE, CSE

• Holding a short work-in-progress seminar before the final deadline.

• Inviting winning submissions to contribute their methods as components to LearnSphere, so they can be easily applied to other datasets (again, possible with mini-grant funding).

• Looking for a sponsor (company, university) to offer a monetary award to the winning team.

We are exploring the above options at the 5th CSEDM Workshop and will update this section with more information.

Disclaimer and Bug Reporting

The 2nd CSEDM Data Challenge, including data, example code, and documentation, are provided "as is." The organizers have done their best to ensure their quality, but they cannot make any guarantees.

• Report any issues with the code on the GitHub repository.

• Report any issues with CodaLab or the datasets on the respective forum.