Inspiration
Tried every architecture found most success with CNN and optimized as such
What it does
Ensemble sparse CNN over eight machines for episode dumping to 5x1 output
How we built it
Hyperparameter optimization loop
Challenges we ran into
Challenges with data sanitization and leekage as well as producing valuable synthetic data
Accomplishments that we're proud of
Architecture and training pipeline
What we learned
Better data handling and meta optimization
What's next for House Edge Main Round
Further work into optimizing and speeding up the training pipeline
