Add basic instructions for Radar

[markkuriekkinen]

The A+ manual has only a short chapter about Radar (similarity/plagiarism detection) that explains how to add the Radar option to the submit directive. However, we need short basic instructions on how to use Radar as a teacher.

• exercise setup in Radar (the automatic setup is preferred and it requires that the tokenizers are configured in the A+ course, i.e., exercise submit directive or the config.yaml)
• how to efficiently check the results? Where to look and what to look for?
• how to quickly deduce how many submissions look suspicious, per exercise?
• when one clicks on an exercise, there are rows and columns with student numbers. How to read these?
• Is there a possibility to get this data in some kind of csv?

What the teacher wants to know: In how many exercises are solutions from student x the same as from student y?

[markkuriekkinen]

Quick comments about what can be included in short instructions

You can open the "Graph view" in the top of the Radar course page. This view reveals student pairs that have several matches with each other's submissions across multiple exercises.

In the exercise page, in the matrix with many rows and columns, one cell represents a comparison between two submissions with a high similarity. You can click on the cell to see the submission source codes side-by-side as well as links back to the A+ submission pages. The first submission is the same for all cells in the same row, i.e., the row compares one student to other students. It looks like the first two cells in the row show the same submissions, but the cells show the student ids of the students. The same student pair may be shown on two different rows, but with the pair swapped (A-B to B-A). Basically, the matrix may alert multiple times about the same student pair.

Note that one student may submit multiple times in an exercise and Radar compares all submissions with each other. One student may have multiple submissions that are similar to another student's submission.

Unfortunately, I don't think it is now possible to download the comparison data in CSV. There is no export feature.

Remember that high similarity can be incidental and not every case involves plagiarism.

[markkuriekkinen]

These are important too:

• explain the comparison logic:
  • static (no dynamic program execution and run-time comparison)
  • tokenizing the code so that renaming variables does not fool the system
  • how long identical token sequences are alerted (default 15 and longer)
  • can the system be fooled by breaking identical sequences with no-op operations like "if False: pass"
  • how is the similarity percentage 0-100% defined and the default threshold for alerting (70%)

[markkuriekkinen]

The commit https://github.com/apluslms/aplus-manual/commit/6ce0da1c49506af5173aeef3bf5f0932197c600b added some new Radar instructions.