Last week was a memorable one for PTP hosting the first AWS Partner led DeepRacer event!Our workshop was set to start at 11:30 AM with 11:00 AM arrivals at The Hive in Norwood, MA (home to PTP’s HQ). With a line out the door already at 10:45 AM, it was clear this was an event in which DevOps engineers in New England were interested (or perhaps geeked-out about!). The AWS DeepRacer League is growing fast, and this was a unique opportunity to get instruction from AWS experts as well as work together in teams focusing on #machinelearning and #reinforcementlearning with the opportunity to measure results with racing the DeepRacer car on the track. This event is far different than a typical high-level company presentation and then breaking to eat/drink. This was a full day of instruction, learning, teaming, and measuring. PTP was thrilled to be able to host.
Once the attendees got settled and a brief introduction from yours truly, the event was kicked off by Jillian Forde, a Solutions Architect with AWS. An expert in the DeepRacer, she walked the group through the curriculum inclusive of building learning modules in AWS SageMaker and instructions on building out models for the DeepRacer cars. After about an hour of instruction from Jillian, teams of 4 were established and the modeling began.
The teams had about 3 hours to work together on their learning models. As they made progress, they were able to test their models on the track before the “real” racing began. We were extremely fortunate to have the AWS Pit Crew full of other expert Solutions Architects available to transfer the models to the cars, provide instructions for managing the speed of the cars from a tablet and to be the sherpa for the cars on the track. Special thanks to Jon Myer, Nick Wislett and Neal Mcfee for carving out the time and making the trip to Norwood to lend their support the entire day. As teams would test their cars navigating around the track, they’d head back to their work areas in search of improvements and perhaps some additional consultation with Jillian for some pointers!
3:30 PM was the hard stop for the model development and the #reinforcementtraining. It was now time to officially test what they had learned and what they were able to accomplish in a relatively short three-hour period of time. Each team had 4 minutes of race time whereby we would record their single fastest lap time. If a car went off track more than 3 times in a lap, the AWS Pit Crew member would bring it back to the start.
We had some really creative team names, such as “S-Car-Go”, and some not-so-creative like “Team Cloud” and “Team I Don’t Know”. Perhaps we should have offered time bonuses for creative names, but alas, that will be for our next event! Each team had a driver who controlled the speed of the car from the tablet, helping guide it around the track. From my observation, the right speed at the right time was definitely a key to managing the car to navigate around the track successfully. It was an exciting time for the team with the car on the track as well as the other teams interested to see how well others were doing. As it turned, out, the first team to hit the track, Team Cloud, delivered a lap time that was never able to be matched. S-Car-Go got within .5 seconds, but they were on top of the leaderboard for the afternoon.
We wrapped our session with medals to the top three teams and some brief discussion on lessons learned. Consistent feedback was that the more time they had to let their models learn, the better success they would have. Perhaps we will test that sometime with a longer day of learning.
We’ve had many organizations come to us asking if we can bring this enriching event to their office and conduct the workshop for their teams. Yes! But be aware there are some logistics involved and space is required (approximately 30′ x 20′) for the track so the right facility is needed. We will absolutely deliver this workshop again, please let us know if you’re interested.
Lastly, thanks to our AWS Partner team who worked with us to make this possible, we look forward to future events to help develop DevOps talent in an effort to improve our industry.
#ReinforcementLearning #MachineLearning #AWS #DeepRacer #DevOps #Cloud #SageMaker #Modeling