This repository contains source files for Gradescope's autograding facilities, described at https://gradescope-autograders.readthedocs.io/en/latest/getting_started/. By default we are assuming submissions in Python 3.
There are two parts to a Gradescope autograder: (1) the setup script, which installs the necessary software, and (2) the test runner and feedback mechanism, which runs a script over the submission and produces a results.json output.
Nbgrader, https://nbgrader.readthedocs.io/en/stable/, consists of several parts:
- A series of scripts that take a "homework template" (with solutions and test cases) and produce a "homework release" (eliding solutions and protected tests).
- A grading database, with scores and per-question breakdowns.
- A set of user tools.
We expect that Step 3 is set up outside of Gradescope, eg on the local machine, with statically named cells in the notebook for the key parts the user is submitting.
From that step, we take the homework template (under sources/ps1). The script will create a local copy of the nbgrader database for each submission, and will essentially assume there is one student ("Alyssa Hacker") and one homework ("ps1"). It will run nbgrader for this, then export the results. Those results are converted back into a JSON file fed to Gradescope.
The main components are:
-
setup.sh: This installs (onto the Gradescope Ubuntu 18.04 image) common libraries and the nbgrader package, including:pandasandnumpytensorflowpysparknetworkxmatplotlibsqlalchemykeras-preprocessing
-
nbgrader_config.py- configuration for nbgrader. Assumes one student (with ID 'hacker', aka Alyssa Hacker) and one problem set,ps1. -
requirements.txt- this is a Python dependencies file for pip. Include versioned dependencies here. -
run_autograder- this is a bash script to run nbgrader (with feedback) and to generate theresults.jsonfile Gradescope expects.
We have also included sample notebooks, based on those from nbgrader, in the source directory. See the LICENSE file for information on the license.