Augmentation_with_GPT

What I did: Augmenting Concept ARC Problem

How to augment

1. Guessing input data by some characteristics in ARC Problem

2. Augmentation Process

Prompt(역변환 방법 프롬프트) is written for each Task (Concept ARC has 16 Tasks)

• dotted input(pink and green one) is augmentation target.

• Chat GPT makes other inputs which is suitable to given output

  i. Make prompt file(Prompt.json) with GPT_prompt.py

  ii. Use GPT API (GPT3.5_prompt.py / GPT4.0_prompt.py) to make Augmented DATA

  iii. I have chosen appropriate input by hand_filter.py

If you want to use this method with other prompt, you may change prompt in GPT_prompt.py

3. Code File(GPT_DATA) to augment Concept ARC

   i. GPT3.5_prompt.py

   This file is using GPT-3.5 API to make ARC Demonstration

   ii. GPT4.0_prompt.py

   This file is using GPT-4.0 API to make ARC Demonstration

   iii. GPT_prompt.py

   This code is making prompt json file

   • You should change this file if you would like to change prompt.

   iv. Prompt.json

   This json file is that I have used to augment concept.

   This json file is composed of input, output, task(Concept ARC Task e.g. AboveBelow, Center,...), result, test_input and test_output.

   1. input: input data from concept ARC(train)
   2. output: output data from concept ARC(train)
   3. task: task name from concept ARC
   4. result: this array is for complement from chat-GPT, which means it is okay to be empty.
   5. test_input: this array is just for deliver to Result file(to adapt ARC interface) <- this array is not so important
   6. test_output: this array is just for deliver to Result file(to adapt ARC interface) <- this array is not so important

Result Table

Problem Category	Total available	The number of generated data	The number of valid augmentated data	Ratio(valid/generated)
Above Below	58	158	34	21.52%
Center	65	236	35	14.83%
Clean Up	106	183	83	45.36%
Complete Shape	58	147	37	25.17%
Copy	27	153	4	2.61%
Count	56	202	29	14.36%
Extend To Boundary	37	167	8	4.79%
Extract Objects	44	176	21	11.93%
Filled Not Filled	58	203	29	14.29%
Horizontal Vertical	32	114	7	6.14%
Inside Outside	52	191	24	12.57%
Move To Boundary	36	165	12	7.27%
Order	47	162	26	16.05%
Same Different	107	246	76	30.89%
Top Bottom 2D	92	255	59	23.14%
Top Bottom 3D	55	215	25	11.63%
Total	930	2913	509	17.12%

v. hand_filter.py

this python file is that I've used to filter inadequate data.