/aied2024-xlogo-tasksyn

Primary LanguagePython

Task Synthesis for Elementary Visual Programming in XLogoOnline Environment

This repository contains the implementation for the paper Task Synthesis for Elementary Visual Programming in XLogoOnline Environment.

Project Structure

The project is organized as follows:

  • scripts: Contains scripts for task synthesis using various methods.
  • src
    • xlogomini: Utility functions for the XLogoOnline-Mini platform.
    • xlogominidatagen: Source code for task synthesis.

Installation

To install the required dependencies, run:

pip install -r requirements.txt

Getting Started

Loading Reference Tasks

To load the reference task and code JSON files:

from src.xlogomini.utils.load_data import load_task_json, load_code_json

task_id = '87'
task_json = load_task_json(task_id)
code_json = load_code_json(task_id)

Visualizing a Task

To visualize a task:

from src.xlogomini.utils.image_conversions import task2image

task_id = '87'
task_json = load_task_json(task_id)
task2image(task_json, show=True, save=False, show_desc=True)

task87.png

Displaying Code

To display the code for a task:

from src.xlogomini.utils.load_data import load_code_json
from src.xlogomini.components.code.xlogo_code import Code

task_id = '87'
code_json = load_code_json(task_id)
print(Code(code_json))
# Example output:
# fd
# rt
# repeat(3){
#   fd
# }
# rt
# fd

Executing Code

To execute the code within a task:

import src.xlogomini.emulator.executor as executor

task_id = '87'
task_json = load_task_json(task_id)
code_json = load_code_json(task_id)
executor.execute(task_json, code_json)
# Example output: {'crashed': False, 'crashed_msg': None, 'goal_ok': True, 'cons_ok': True, 'err_msg': 'OK'}

Synthesizing Tasks

We provide two techniques for task synthesis: XLogoSyn and RotateFlip.

XLogoSyn

To synthesize tasks using XLogoSyn, run:

bash scripts/xlogosyn.sh

The following directories will be created:

  • ./results/datagen/code: Generated code JSON files.
  • ./results/datagen/goal: Generated goal JSON files.
  • ./results/datagen/task: Generated task JSON files.
  • ./results/datagen/params: Parameters used for data generation.
  • ./results/datagen/image: Images of the generated tasks.

In the script, you can specify the --task_id and the --diff parameters. There are three difficulty levels available for tasks: easy, medium, and hard.

Note: Running the above script may take some time, ranging from a few seconds to several hours, depending on the specific tasks and parameters set in the script. The exact time and parameters used will be saved in a JSON file in the ./results/datagen/params folder.

RotateFlip

To generate tasks using RotateFlip, run:

bash scripts/rotateflip.sh

The generated data will be stored in the ./results/datagen/image folder.

Citation

If you use this code in your research, please cite the following paper:

@inproceedings{wen2024xlogosyn,
  author    = {Chao Wen and Ahana Ghosh and Jacqueline Staub and Adish Singla},
  title     = {Task Synthesis for Elementary Visual Programming in XLogoOnline Environment},
  booktitle = {AIED Track on Late Breaking Results},
  year      = {2024}
}