ChatGPT-GameRec

A GPT-powered chatbot that suggests games you'll love

Setup & Install

Install

python -m venv env
source env/bin/activate
pip install --upgrade pip
pip install -r requirements.txt

On Windows cmd or PowerShell, instead of using the source command above to activate the virtual environment, use:

env\Scripts\activate.bat

Docker install on Ubuntu

Install docker using the repository

VS Code & WSL with Ubuntu

Prerequisites:

• Windows 11 or Windows 10 version 2004 or higher to support WSL2: How to Install Linux on Windows with WSL
• VS Code with WSL, Remote Development, and Python extensions

Optional:

• Windows Terminal
• GitHub Desktop for Windows

Mind the End of Your Line (link): GitHub Desktop for Windows may change end of line characters from LF to CRLF to match Windows defaults. Scripts in this repo must use LF to run correctly in Docker - in particular run.sh will fail with ^M if CRLF eol characters are used. In the file C:\Users\[username]\.gitconfig section labeled [core] add autocrlf = false to preserve line endings in the repo. Install the VS Code extension eol for a visual indicator of LF and CRLF in the editor. A .gitattributes file in this repo prevents CRLF eol characters from being checked into .sh files.

1. From a PowerShell prompt, run wsl --install from PowerShell and restart Windows
2. Install Ubuntu from the Microsoft Store and setup a root user (a prompt should launch automatically after installing Ubuntu or by manually launching Ubuntu from Windows Terminal)
3. Clone the repository in the Linux filesystem for improved performance, either in Ubuntu shell or with Windows GitHub Desktop app referencing folder \\wsl.localhost\Ubuntu\<path to repo>.
4. Launch VS Code from the repository location in WSL: code . and observe VS Code connected to WSL via green status bar icon at lower-right: ><WSL: Ubuntu
   1. Or: Launch VS Code from Windows, choose "Connect to WSL" in the command palette (Ctrl+Shift+P). Then open the repo folder in VS Code Explorer.
5. Ensure python, pip, and venv are installed on Ubuntu. In VS Code Terminal or Windows Terminal (Ubuntu): apt install python3 python3-pip python3.10-venv
   1. Note: Python 3.10 was the default version installed in Ubuntu 22 LTS as of this writing 9/4/2023.
6. In the VS Code command palette, run Python: Create Environment. This will create a new venv in the folder .venv by default.
7. After VS Code creates the environment, navigate to the VS Code Terminal window and install dependencies:
   1. Note that the terminal prompt is prefixed by (.venv) and you are working in the virtual environment created in VS Code
   2. Run pip install --upgrade pip
   3. Run pip install -r requirements.txt
8. VS Code setup is complete, main.py can be run in the debugger with F5.

Config

cp config.template.json config.json

Replace values from the template with your actual values:

• YouTube Data API v3 Key
• channel ID(s) (if desired)
• OpenAI API Key
• [Optional] LangSmith API Key (https://smith.langchain.com/settings)

Note: channel_id can be a single string for one channel ID, or a list of strings if you want to extract video transcripts from multiple channels.

Streamlit config

1. Create file: .streamlit\secrets.toml
2. Add OpenAI API key to secrets.toml: OPENAI_API_KEY = "your_api_key"

Execution

Run python main.py -h for more info and options.

End-to-End

python main.py -m end-to-end

This extracts transcripts to a .json file, creates and saves FAISS embeddings to the faiss_index directory, then starts an interactive chat session with the extracted documents.

Piecemeal

Extract Transcripts

python main.py -m extract-transcipts

Extracts transcripts to a .json file; defaults to transcripts.json, specify with -tf argument.

Create Embeddings

python main.py -m create-embeddings

A .json file with extracted transcripts must exist to create embeddings; specify with -tf, defaults to transcripts.json.

Summary Demo

Get an LLM-generated summary of one random video.

python main.py -m summary-demo

A .json file with extracted transcripts must exist to run the summary demo; specify with -tf, defaults to transcripts.json.

Chat Demo

python main.py -m chat-demo

FAISS embeddings must have already been extracted and placed in the faiss_index directory.

Streamlit

Command Line

streamlit run Chat.py

Docker

docker build -t gamerec:v1 .
docker run --rm -p 8880:8501 gamerec:v1

Navigate to http://localhost:8880 in your browser to view the application.