taabishm2 / cosi

Code Similarity detection for Python files and Jupyter Notebooks

cosi

Locally run code similarity detection for Python files and Jupyter Notebooks

Runs pairwise comparisons between Python code from .py and .ipynb files to determine:

• Code similarity: Compares code contents after substituting variable names and running other normalizations. Uses a windowed, fingerprinting approach to come up with vector representations for the submissions, which are then compared using cosine similarity. Based primarily on http://theory.stanford.edu/~aiken/publications/papers/sigmod03.pdf.
• Output similarity: Compares similarity between .ipynb output cells (again using the cosine similarity metric)
• Variable similarity: Compares similarity between all variable names, including function and class names, using Jaccard similarity calculation.
• Structure similarity: Compares the abstract syntax trees (ASTs) for different submissions using Jaccard similarity.

Example output:

{
  "repoA": "https://github.com/cs544-wisc/project-6-repo-A",
  "repoB": "https://github.com/cs544-wisc/project-6-repo-B",
  "code": 0.0258,
  "variables": 0.357,
  "output": 0.9634,
  "struture": 0.7879,
  "visualize": "python3 display.py 'project-6-repo-A' 'project-6-repo-B'"
}

Note: The code within the files must be syntactically valid since CoSi uses AST parsing to extract code structure.

Checking code similarity

1. Add a new directory (e.g. project-10) within ./files/submissions containing individual student submission directories
2. Update the .env file; set SUBMISSIONS_ROOT_DIR as the folder name (project-10/) and FILES_IN_SUBMISSION as the list of files to be read from each sub-directory of SUBMISSIONS_ROOT_DIR e.g (p10.py or nb/client.py,nb/server.py). Also, set the GITHUB_PREFIX as the URL prefix to add to the directory name (used in the resulting CSV) or leave blank if not applicable
3. Run python3 CoSi.py
4. The result is saved as a CSV in ./files/results identified by the time of generation. A FAILED csv is also included if parsing of any files fails

Viewing code pair diffs

The result csv contains a column to view differences between the code in two submissions. In general, to visualize the text-diff of two submissions, set the SUBMISSIONS_ROOT_DIR and FILES_IN_SUBMISSION fields in .env. Then execute python3 display.py 'repoA' 'repoB', where repoA and repoB are student directories within the SUBMISSIONS_ROOT_DIR. This will open a browser window with a side-by-side comparison of all files from FILES_IN_SUBMISSION in the two repos.

Note: If repoA and repoB are present in SUBMISSIONS_ROOT_DIR, the saved version of the files will be displayed. If any of the repo names passed is not saved locally, it will be cloned from git into SUBMISSIONS_ROOT_DIR first.

Note: The script uses diffcheck.com public APIs to retreive the diff HTML (https://api.diffchecker.com/public/text). Use of diffcheck API must abide by their API terms.

Example

An example submission is provided for which CoSi may be executed immediately.

• To run CoSi for the example, execute python3 CoSi.py
• To view the results, see the csv file generated in results/
• To visualize the diff between the example submission pair, run python3 display.py 'project-6-repo-B' 'project-6-repo-A'