Team Members: Aarav Sharma, Benjamin Hinchliff, Nathan Ip, Brian Bivinetto

Acknowledgement:

• Cal Poly CSC 480 Project
• Professor/Instructor: Rodrigo Canaan

External Sources:

• K-means Constrained:
  • https://joshlk.github.io/k-means-constrained/
  • https://github.com/joshlk/k-means-constrained
• GLoVE
  • https://nlp.stanford.edu/projects/glove/
  • https://github.com/stanfordnlp/GloVe
• Word2Vec
  • https://radimrehurek.com/gensim/auto_examples/tutorials/run_word2vec.html
  • https://towardsdatascience.com/introduction-to-word-embedding-and-word2vec-652d0c2060fa

Dependencies:

• Installation: pip3 install -r requirements.txt
• Primary Dependencies
  • numpy
  • pandas
  • sklearn
  • k_means_constrained
  • gensim
  • matplotlib
  • jupyter

Running Code:

• To choose and play one game: python3 connections_agent.py
  • Agent plays the Connections game and either wins or loses
  • Agent usually wins when playing Connections Game #41
  • Agent usually loses when playing Connections Game #1
• To play all games: python3 connections_agent.py all
  • Agent plays all the Connections games
  • Results: Total number of games played and win rate is shown at the end

Notebooks:

• glove.ipynb
  • Uses GloVe for word embeddings
• word2vec.ipynb
  • Uses Word2Vec for word embeddings

Results/Visualizations

• Running the corresponding notebook will create these visualizations/reports
• Images in the visualization folder
• word2vec.ipynb
  • word2vec-cluster-visualization.png
  • word2vec-group-accuracies.png
  • word2vec-silhouette-scores.png
  • Word2Vec Summary Statistics Output
• glove.ipynb
  • glove-cluster-visualization.png
  • glove-group-accuracies.png
  • glove-silhouette-scores.png
  • GloVe Summary Statistics Output
• Running the agent (in the running code section) will allow you to play the Connections games

Files Needed:

• Connections Game Files:
  • connection-archive.txt (Raw text format of all Connections games)
  • connections.csv (CSV format of all Connections games)
• Pre-trained Embedding Files (Not in GitHub Repo - Too Large)
  • Word2Vec File
    • GoogleNews-vectors-negative300.bin (pre-trained Google News corpus embedding model containing 3 million 300-dimension English word vectors)
    • Link: https://drive.google.com/file/d/0B7XkCwpI5KDYNlNUTTlSS21pQmM/edit?resourcekey=0-wjGZdNAUop6WykTtMip30g
  • GloVe File
    • ./get-glove.sh
    • glove.840B.300d.txt
    • Link: https://huggingface.co/stanfordnlp/glove/resolve/main/glove.840B.300d.zip