Overview - Deep Learning in Halite AI competition p.1

TL;DR

This tutorial explores an ML approach using evolutionary models for solving the navigation problem in Halite 3.

Transcript

what is going on everybody and welcome to part 7 of the halite 3 tutorial series as well as the first part to the ml section of halite 3 tutorials I do want to stress I am here mostly just to tinker with m/l I have no promises here it's kind of like the Python plays GTA or the sc2 series I don't know what the outcomes gonna be it's just a realistic... Read More

Key Insights

• ❓ The author emphasizes the preference for evolutionary models over supervised learning for solving the navigation problem in Halite 3.
• 😒 The author highlights the use of convolutional neural networks for processing the game map data.
• 👾 The importance of correct ordering of coordinates in the game map for accurate predictions is emphasized.
• 👤 The author mentions the possibility of using different ML models, such as reinforcement learning with Q-learning, based on user suggestions.
• 🧘 A visual representation of relative positions and their impact on decision-making is shown.
• 👾 Handling varying game sizes by cropping or padding the game map is discussed.
• ⚓ The concept of relative viewpoint for each ship and its significance in model training is explained.

Questions & Answers

Q: What is the author's preferred method for solving the navigation problem in Halite 3?

The author prefers an evolutionary-based method that learns from randomness, providing a more raw and interesting approach.

Q: How does the author suggest handling the varying game sizes in Halite 3?

The author suggests either using padding or cropping the game map to a smallest size, such as 32 by 32, while considering the relative positions of each ship.

Q: What information does the author want to extract from the game map for each ship?

The author wants to extract the amount of halite, presence of ships or shipyards, and whether they are friendly or enemy entities.

Q: How does the author plan to generate data for training the ML model?

The author plans to collect random moves as data and then filter out the best games for training the model iteratively.

Summary & Key Takeaways

• The tutorial focuses on using an evolutionary approach rather than supervised learning to solve the navigation problem in Halite 3.

• The author explores the use of convolutional neural networks and discusses the challenges they face.

• The author discusses the concept of relative viewpoint for each ship and how it affects decision-making.

• The tutorial highlights the importance of correct ordering of coordinates in the game map.