Coding Assistant for PADS and UKY Statistics

Step 1. Create an OpenAI account and obtain an API key and an organization ID

• Finding your OpenAI organization ID
• Generating and retrieving your OpenAI API Token
  
  

Step 2-1. CLI Users : Clone the git repository and install packages

• To clone the repository and install packages, open the terminal or PowerShell for both Mac/Linux and Windows users.
• Note: For Windows users, it is strongly recommended to use the WSL.
• Run the following command in the terminal:

git clone https://github.com/leeparkuky/pads-coding-assistant.git
cd pads-coding-assistant
pip install -r requirements.txt
cd pads_gpt_assistant

Step 2-2. Python Users: install package

pip install pads_gpt_assistant

Step 3-1. [CLI User] Copy and Paste the organization ID and API Token into pads_gpt_assistant/openai_config file

Note : Remember to save the openai_config file.

Step 3-2. [Python User] Execute the following python lines

from pads_gpt_assistant.config import create_openai_config_file
organization_key = "<paste the organization key>"
api_key = "<paste your api key>"
create_openai_config_file(organization_key, api_key)

Step 4. Use either the coding_gpt.py script or coding_assistant (from pads_gpt_assistant python package)

• Option 1: Using CLI (Using Mac/Linux terminal)

  • Quick start:

    • First, copy the code you want to evaluate or get a feedback on; specifically, the codes to be copied in the clipboard.
    • The --language argument tells ChatGPT to be an instructor for a specific programming language. The default language is python
    • If the code is running fine and is error-proof, ChatGPT may explain each component in the code.

    python coding_gpt.py --language "R"

  • When you have an error message:

    python coding_gpt.py --error_message "<error message>" --language "SAS"

  • Other options:

    • If you are not sure whether the codes are saved in the clipboard but want to paste them in the CLI command, use the --user_message argument:

    python coding_gpt.py --user_message "<paste your wrong codes>" --error_message "<error message>" --language SQL --num_result 2

    • If you want to check other responses, set the --num_result argument greater than 1

    python coding_gpt.py --error_message "<error message>" --num_result 2

    • If you want to save the response(s), set the --save_result argument true. It will create a file history.txt

    python coding_gpt.py --error_message "<error message>" --num_result 2 --save_result true

    • If you have an error message, and just want to ask a question, set the --only_question argument true

    python coding_gpt.py --user_message "<your question regarding the error message>" --error_message "<error message>" --save_result true --only_question true

    • If you just want to ask a question, leave --user_message empty and set --only_question true

    python coding_gpt.py --user_message "<your question>" --only_question true

• Option 2: Using the module installed by pip install pads_gpt_assistant

• Quick start:
  • First, create an assistant object:

    from pads_gpt_assistant.coding_gpt import *
    assistant = coding_assistant(record_history = True, continue_conversation = True) 

  • Setting the record_history argument true will create the history.txt file and the continue_conversation argument with true will enable the use of past conversations.
  • Then, for the first conversation, use the get_chat_response method function:

    assistant.get_chat_response()

    In the example above, wrong codes are in the clipboard and you don't want to or have an error message. If you want to paste the wrong codes with the error message, do following:

    assistant.get_chat_response(codes = "<wrong codes>" , error_message = "<error message>")

  • If you liked the response, and wish the GPT perform as well in the future, let the assistant know you liked the previous response. It will remember how it responded to the previous question and will try to perform as good or better than before.

    assistant.like_it()

  • For the subsequent queries, use get_chat_response again. But suppose you want to have multiple responses, for example 3, and want to check the two best responses. Then set samples and top_k as 3 and 2 respectively.

    assistant.get_chat_response(codes = "<wrong codes>" , error_message = "<error message>",
                                samples = 3, top_k = 2) # generate 3 responses and show 2 best responses