Via

This repository contains the source code and the parameters for the construction of various datasets and projection models, as part of the Via academic pathways project, sponsored by the Carta lab at Stanford University. The below sections are usage instructions. We are actively modifying this code, so if you have any questions please feel free to contact Geoffrey Angus ([email protected]) or Richard Diehl Martinez ([email protected]).

Overview

This repository makes use of params.json files in order to maximize reproducibility in experiments. There are (at time of writing) four commands you can currently execute: build_dataset, build_projection, run_metrics and enrich_projection.

0. Setup and Installation

In order to use this repository, you must have a csv of enrollment data for your school. We suggest you store this .csv file, by creating a folder under /data called 'raw'. Then you can save your .csv file to the /data/raw folder.

The general format and column names of your data must be consistent with the following example table:

student_id	course_id	quarter_id	quarter_name	dropped	curr_major	final_major
23Hkad2	CS106A	2130	2000/2001, Winter	0	MATH-BS	CS-BS
12Lazc8	CS106B	2132	2000/2001, Spring	0	MATH-BS	SYMBO-BS
...

The columns of your dataset must include exactly the following values:

• student_id (hash string): a unique hash-id corresponding to an enrolled student
• course_id (string): the id of a course that a student specified by student_id enrolled in during the quarter specified by quarter_id
• quarter_id (integer): the id of the quarter during which a student specified by student_id took a certain course specified by course_id
• quarter_name (string): a human legible string of the quarter corresponding to quarter_id
• dropped (0/1): a boolean to indicate whether the class specified by course_id was dropped by the student specified by student_id.
• curr_major (string): the current major of the student specified by student_id
• final_major (string): the final major of the student specified by student_id

Our model was trained using data kindly provided by the Carta lab at Stanford University. Unfortunately, we are unable to redistribute this data.

Please run the following command from within the repository upon cloning the repository:

pip install -r requirements.txt
python setup.py develop

1. Create a Sequence Matrix Dataset

The code implementing the pathways network requires a matrix M of shape (num_students, num_classes), where M[i][j] = t means that student i took course j at timestep t (t=1 is the student's first quarter at university, t=2 is the student's second quarter, etc.). Zero-valued entries signify non-enrollment. We can generate this matrix by running the following command:

build_dataset <params_dir>

This command assumes that there is a params.json file within the directory provided that specifies the type of sequence dataset you would like to generate. The sequence matrix will be saved at <params_dir>/sequences.npy and each entry of the dataset will be mapped to course ids in <params_dir>/course_indices.json. Example params.json files can be found in the data/pathways_datasets directory.

2. Create a Pathways Network

Once you have generated a sequence matrix, one can generate a pathways network. Run the following command:

build_projection <params_dir>

Sample params.json files can be found in the experiments directory.

3. Prepare Network for Visualization

We have been using Cytoscape in order to create the visualizations for our pathways networks. Run the following command in order to create an enriched graph text file:

enrich_projection <params_dir>

The enriched text file will isolate the departments in which each of the course pairings reside, as well as creating an edge attribute differentiate between intra- and inter- departmental pathways. This is beneficial when styling the visualizations later on.

4. Evaluating the Pathways Network (Under Construction)

One can also run metrics on the pathways networks. Run the following command:

run_metrics <params_dir>

Note: you MUST create a metrics directory within the original experiment directory otherwise the program will not run. A successful run_metrics command will consist of the following file structure:

experiments/
  # some number of sub-directories
  ...
    test_experiment/
      metrics/
        params.json
      params.json
      projection.txt

The <params_dir> argument thus refers to the params.json found in the metrics folder.

---

That's it for now. Again, please email Geoffrey Angus ([email protected]) or Richard Diehl Martinez ([email protected]) if you have any questions.