DanMcInerney/mma-ai

MMA AI Dataset Artifacts

This dataset contains the database dumps and runtime artifacts needed to reproduce the mma-ai workflow from the companion code repository:

https://github.com/DanMcInerney/mma-ai

Contents

• dumps/mma-ai.postgres-custom - custom-format PostgreSQL dump for the main mma-ai database. This contains the features schema used by DATABASE_URL.
• dumps/odds.postgres-custom - custom-format PostgreSQL dump for the separate odds database. This contains bestfightodds.bfo and is used by ODDS_DATABASE_URL.
• processed/training_data.csv - generated win-model training data, included for convenience.
• processed/training_data_dec.csv - generated decision-model training data, included for convenience.
• processed/prediction_data.csv - generated prediction feature data, included for convenience.
• models/ag-20260304_110750-win-extreme.tar.gz - pretrained AutoGluon win model.
• manifest.json - sizes, SHA256 hashes, DB metadata, and source details.

The dumps were created with PostgreSQL 18.1 custom archive format and gzip compression.

Restore Databases

Create local databases:

createdb -U postgres mma-ai
createdb -U postgres odds

Restore the dumps:

pg_restore --clean --if-exists --no-owner --jobs 4 \
  --dbname "postgresql://postgres@localhost:5432/mma-ai" \
  dumps/mma-ai.postgres-custom

pg_restore --clean --if-exists --no-owner --jobs 4 \
  --dbname "postgresql://postgres@localhost:5432/odds" \
  dumps/odds.postgres-custom

If your local Postgres requires a password or different username, use your own connection string.

Configure The Code Repo

Create .env in the code repo:

DATABASE_URL=postgresql://postgres@localhost:5432/mma-ai
ODDS_DATABASE_URL=postgresql://postgres@localhost:5432/odds
MMA_AI_DATA_DIR=./data
MMA_AI_MODELS_DIR=./AutogluonModels

Install dependencies:

uv sync

Use The Pretrained Model

Extract the model archive into the code repo's AutogluonModels directory:

mkdir -p AutogluonModels
tar -xzf models/ag-20260304_110750-win-extreme.tar.gz -C AutogluonModels

Copy or download the processed prediction data into data/prediction_data.csv, then run:

uv run python predict.py \
  --model-path AutogluonModels/ag-20260304_110750-win-extreme \
  --prediction-data-csv data/prediction_data.csv \
  --training-data-csv data/training_data.csv \
  --no-shap

Rebuild And Retrain Instead

After restoring both databases, you can rebuild generated CSVs from the DB and scrape if needed:

uv run python main.py --reset-db
uv run python -m libs.modeling.train --model-type win

The processed CSVs are included so users can skip the rebuild step for common workflows.