First off, thanks for the TR team for launching this. And congrats @ambermalika.
As we have all probably experienced, sometimes our data collection devices fail us (power meter, HR and Cadence Sensors, etc). Literally, my HR monitor battery died during today’s VO2 max workout, and was stuck at 120bpm for half the workout. These kind of issues, while frustrating, never informed future workouts the way adaptive training might.
In this case I would love to be able to delete out the HR data from the ride, as is is completely worthless data. This will be more important as I move to outdoor rides, and will lose the power meter data collection. As they say garbage in, garbage out.
Maybe this isn’t so important with version 1.0 of adaptive training, but I would hate for the system to think I’m a Watt God being able to do VO2 max repeats with an Endurance level bpm.
Thanks again for releasing a product that is geared for me. I just started a new job, and have been missing more workouts then I would like. It’s hard missing a workout, knowing the next one just got harder as a result.
While I don’t know for sure, I’m quite certain this has to have been built in. Noisy/bad data is an issue you tackle in any large data set.
It doesn’t have to be bad data either, you could just upgrade power meters. That’ll be an issue for me, I’ll be replacing my current road bike with a 4iiii singled-sided power meter with a new road bike that has a Quark DZero on it. So I will have to re-test my FTP, etc. If you use a smart trainer to measure power indoors and a power meter to measure power outdoors, you’ll run into the same problems. Or if you have more than one bike.
Don’t worry - we’ve already accounted for this! It’s pretty mind-blowing how well our models can detect what’s happening in these cases. The model can also ‘understand’ more than enough from the power data alone to determine how to adapt your plan. So, in this instance for example, the model would automatically rely on the power data to inform future workouts. We’ve done a lot of internal testing, and this was a shared concern that our ML team carefully addressed. The coolest part is that the model will only get better at this with time!
In one of the screenshots Nate shared during the podcast there was an option to denote equipment failure, which would omit the ride data from your AT model.