This repository contains the complete configuration files necessary for the replication of Fei and Yeung (2015), "Temporal Models for Predicting Student Dropout in Massive Open Online Courses" using the MOOC Replication Framework (MORF). The complete results of this replication are described in Gardner, Yang, Baker, and Brooks (2018), "Enabling End-To-End Machine Learning Replicability: A Case Study in Educational Data Mining."
docker
: contains dockerfile and necessary scripts to build the docker image. This image can also be pulled directly from docker cloud by running the following in a terminal (note Docker must be installed):
docker pull themorf/morf-public:fy2015-replication
config
: contains two subdirectories, holdout
and cv
, with configuration files to reproduce the experiment using the holdout and cross-validation architectures, respectively. Note that weeks are zero-indexed (so week_0
actually uses one week of features, and week_4
uses weeks one through five, utilizing the method described in the original Fei and Yeung paper).
To execute one of the trials described here (where a trial is a specific model evaluated with features up to a specific week number), use the MORF API functions:
from morf.utils.submit import easy_submit
easy_submit(client_config_url="https://raw.githubusercontent.com/educational-technology-collective/fy2015-replication/master/config/holdout/week_4/svm/controller.py", email_to="[email protected]")
Note that the complete extraction-training-testing pipeline may take several hours. Also note that if you are using a job which utilizes fork_features()
, the job it is forking from must be executed first.
Each experiment also includes a persistent Digital Object Identifier which contains links to the client.config
and controller
scripts, which, along with the Docker image described above (which is common to all of the trials), fully reproduces every trial of the experiment.