RoboTutor-Analysis

Official implementation of the experiments in Deep Reinforcement Learning to Simulate, Train, and Evaluate Instructional Sequencing Policies published at EDM Workshop on RL4ED (RL for Education) 2021.

Installation and Setup

1. This repo works with hotDINA to fit model parameters which has scripts to automatically extract RoboTutor-Analysis/Data (which has to be setup). After fitting, this repo obtains these parameters to update on logged data from usage of RoboTutor. The first step would be to create a project folder: mkdir project

2. Clone this repo and the hotDINA repo inside the project folder.
   

   cd project
   git clone https://github.com/jithendaraa/hotDINA
   git clone https://github.com/jithendaraa/RoboTutor-Analysis
   

3. Setting up all the required RoboTutor data in RoboTutor-Analysis/Data:

   • Check out step 3 in README for hotDINA to set up activity tables, CTA and Code Drop 2 Matrices.
   • For getting transactions tables per village, see step 2 in the same README.
   • At the end of the data setup your Data directory must look something like this:
   

   Don't worry if you do not have the .pkl and CTA_22.xlsx files, the various scripts will still work!

Instructions and Usage

1. Coming soon!

Todo:

• Complete this README.md

Student Models

• ItemBKT
• ActivityBKT
• hotDINA_skill
• hotDINA_full

RL Agents and integrations with student models (primary usage is meant to be with PPO since it is SOTA)

• Type 1 (Action: Finding optimal performance thresholds t1, t2, t3 per student)
  • ItemBKT
  • ActivityBKT
  • hotDINA_skill
  • hotDINA_full
• Type 2 (Action: Finding optimal performance thresholds t1, t2, t3 after each attempt)
  • ItemBKT
  • ActivityBKT
  • hotDINA_skill
  • hotDINA_full
• Type 3 (Action: Decide promotion/demotion; choose one out of prev, same, next, next-next)
  • ItemBKT
  • ActivityBKT
  • hotDINA_skill
  • hotDINA_full
• Type 4 (Action: Same as type 3 but transition constraint is no longer 4 (prev, same, next, next-next))
  • ItemBKT
  • ActivityBKT
  • hotDINA_skill
  • hotDINA_full
• Type 5 (Action: Same as 3 but with removed transition constraint and area rotation constraint "Lit-Num-Lit-Sto")
  • ItemBKT
  • ActivityBKT
  • hotDINA_skill
  • hotDINA_full

For meaning of "types of RL agent", please refer this doc