Machine Learning + AWS -- Leveling Up Our Farmers!

Day[9] Code and Eng DK30 Spring 2019 18 22

Description

Eric and I are continuing our Machine Learning explorations with what we feel is the “final stretch” in understanding the technology and math behind modern ML techniques. We currently have a digital farmer that can learn how to plant & tend to a crop field with semi-optimal behavior. OUR NEW GOAL: Have multi-server AWS training functioning properly for our Unity simulations & understand how to tune parameters to optimize our training! By May, we’d like to have enough knowledge that we can plan out future Unity projects with our ML badassery

Recent Updates

6 years ago

IMPORTANT NOTE: Because this is a research heavy task, we primarily have the weekly goals as a guide of what to work on. The important thing is that we’re spending ~15h a week grinding on these tasks :)

Edit Delete

Estimated Timeframe

Apr 5th - May 5th

Week 1 Goal

Set up single server training on AWS servers. Use AWS for grid search on hyperparameters

Week 2 Goal

Deep dive into PPO math to develop intuition for what the algorithm is good and bad at learning and how to tune hyperparameters

Week 3 Goal

Use PPO learnings to do another grid search on current environment. Plan out and begin work on next environment based on knowledge of PPO

Week 4 Goal

Set up networked training on AWS servers so that we can train our next environment more quickly