This standalone application shows two panels: a control surface with buttons, and a cluster canvas where courses are shown as colored dots. The colors indicate schools; a legend is at the bottom of the cluster canvas.

The check boxes at the top of the control surface are school selectors. They allow choice of which school(s) to include when recomputing a clustering. The more schools are selected, the slower the recomputation.

The draft checkbox allows limiting the total number of courses that are included in the cluster panel. Checking this box results in an incomplete view, but all operations are faster.

Dot size indicates that several courses are combined into the one dot.

• Hover over a dot and see one of the courses represented by that dot.

• Left-click on a dot and see the comprised courses in the course board, which is the tall, white area on the right of the control surface.

• Right-click on a white area of the cluster canvas, then left click to lasso several dots. When the lasso is to be closed, right-click again. The result is display of all lassoed courses in the control surface's course list panel.

  In addition, the reported difficulties and enrollment history of the lassoed courses is shown in two separate popup windows. Those windows are on top of each other; so move the top one out of the way with the mouse.

• Entering a course number, such as "cs 106a" in the search box below the draft mode selector, and clicking the Where Is button shows where in the cluster canvas the course is represented.

• With a course number in the search box, click Top 10 Like Above Course to find bundles of related courses. Even courses not in the model are included in these lists.

• The Save Visualization saves the current cluster canvas and associated metadata in a file in /src/cache. An appropriate name is contructed, which includes the parameters of the display, such as which schools where included, and whether draft or high quality mode.

• The Load Visualization displays shows the files in the src/cache directory, and loads the user-selected file.

• The Recompute button refits a Tsne model, honoring the currently selected choices: draft/full mode, selected schools. In full quality mode the process takes several minutes, up to 20 minutes when all schools are included.

The main code consists of three files: courseSim.py is the main entry point. Starting it without command line arguments starts everything. Python 3 is required.

Clone this repo. Then:

python setup.py install
python setup.py test

File courseSim.py is the entry point into the system. File control_surface_process.py generates, and responds to the buttons and text boxes in the control surface. The UI is constructed using QtDesigner. The generated .ui file is in the qtui/courseTsne.ui file. This file drives the Python code for the control surface. The interaction is managed using PyQt5.

File course_tsne_visualization.py creates the cluster panel, and recomputes. All three files are in their own threads, and communicate via command queues.

The courseSim.py is a mediator between the cluster panel application and the control surface. Message from the control panel to the cluster canvas travel through courseSim.py.

Creating an entirely new vector space model, as opposed to just recomputing a Tsne projection, involves creating input sentences into the word2vec program. Sentence format is:

   *major, course1, course2, course3,...*

where course_n comprise one student's courses in partial order between Freshman and Senior years, inclusive. The README.txt in the data subdirectory contains the necessary queries and other procedures.