chess_player_similarity

Note

• The models for this repository (trained weights) live in LFS using hugging face repository at https://huggingface.co/spitfire1970/chess_player_similarity. GitHub only contains the code.

---

datasets:

• Lichess/standard-chess-games pipeline_tag: image-feature-extraction

---

• this inference endpoint has three sub endpoints defined by the dictionary in the class constructor.
• all requests sent should be of the form {"inputs": data} where the data is another JSON/dictionary/js object.
• the data must be of the form {"num_endpoint": int} where int is: <br>    0 (for testing endpoint) <br>    1 for creating user <br>    2 for the model's move
• to create user embeddings, specify additional fields in data: <br>    "username" : chess.com username <br>    "pgn_content": a sinlge pgn consisting of multiple games played by the user <br>    "games_per_player": specify how many games in the pgn file
• to get an ai_move, specify additional fields in data: <br>    "pgn_string" : pgn of game so far <br>    "color": "white" if model is playing with white pieces else "black" <br>    "player_centroid": player_centroid as python list generated by endpoint 1
• the pytorch model definitions are in the encoder folder, and the model weights are in the *.pt file(s).