https://github.com/abhivur/connections-ai---tamu-datathon-2024

Contributors: Meet Gamdha, Gaurav Nimmagadda
https://github.com/abhivur/connections-ai---tamu-datathon-2024

bert python scikit-learn

Last synced: 13 days ago
JSON representation

Contributors: Meet Gamdha, Gaurav Nimmagadda

• Host: GitHub
• URL: https://github.com/abhivur/connections-ai---tamu-datathon-2024
• Owner: abhivur
• Created: 2025-01-06T17:49:20.000Z (about 1 month ago)
• Default Branch: main
• Last Pushed: 2025-01-06T17:51:21.000Z (about 1 month ago)
• Last Synced: 2025-01-25T14:13:32.964Z (13 days ago)
• Topics: bert, python, scikit-learn
• Language: Python
• Homepage:
• Size: 5.86 KB
• Stars: 0
• Watchers: 2
• Forks: 0
• Open Issues: 0
• Metadata Files:
  • Readme: README.md

Awesome Lists containing this project

README

        
# Connections AI ML Model

This project provides a machine learning-based solution for generating optimal guesses in the **Connections AI Game**. The game challenges players to group related words into four sets of four. The model uses **Sentence-BERT** embeddings and **Agglomerative Clustering** to identify relationships between words and propose accurate guesses.

## Features

- **Semantic Understanding**: The model leverages a pre-trained **Sentence-BERT** model (`paraphrase-MiniLM-L6-v2`) to compute embeddings for each word, enabling the detection of subtle relationships and meanings.

  

- **Clustering-Based Grouping**: Using **Agglomerative Clustering** with cosine similarity as the metric, the model groups words into semantically related clusters, providing logical and data-driven guesses.

- **Adaptive Guessing**: The model incorporates game-specific parameters like:

  - Handling cases where the player is "one word away" from a correct group.

  - Tracking previously guessed groups to avoid repetition.

  - Modifying guesses dynamically based on feedback.

- **Robust Fallback Mechanism**: If clustering does not produce a valid guess, the model falls back to a randomized selection of four words, ensuring continuous gameplay.

- **Error Handling**: The system gracefully handles exceptions such as model loading errors and invalid inputs.

## Key Components

1. **Word Preprocessing**: 

   - Converts all words to uppercase for uniformity.

   - Excludes already grouped words from further consideration.

2. **Embedding Generation**: 

   - Computes word embeddings using Sentence-BERT, capturing semantic nuances.

3. **Clustering**: 

   - Groups words into four clusters based on similarity.

   - Uses cosine distance and average linkage for optimal clustering.

4. **Dynamic Guessing**: 

   - Adapts guesses based on the game's state (e.g., previous guesses, correct groups, and strikes).

5. **Game Logic Integration**: 

   - Ensures compliance with game rules, such as selecting exactly four unique words per guess.

   - Ends the turn when strikes reach the limit or all words are grouped.

## Challenges Addressed

- Identifying semantic relationships between words in diverse and ambiguous datasets.

- Balancing logical guess generation with fallback strategies for edge cases.

- Adapting machine learning techniques to a dynamic and rule-based game environment.

## Technologies Used

- **Python**: Programming language for implementation.

- **Sentence-BERT**: Pre-trained transformer model for embedding generation.

- **Scikit-learn**: For clustering and other machine learning utilities.

- **NumPy**: For efficient numerical computations.

- **Random**: For generating fallback guesses when necessary.

## Acknowledgments

- **Sentence-BERT**: Thanks to the creators of the `paraphrase-MiniLM-L6-v2` model for providing robust word embeddings.

- **Scikit-learn**: For providing the tools necessary for clustering and similarity computation.

- Game concept inspired by the **Connections AI Game** mechanics. 

--- 

This project demonstrates the application of machine learning techniques to create a fun and engaging game experience while exploring semantic relationships in natural language data.