And some power meters will allow you to adjust the slope so you can bring it more into line with the trainer. First job really is to get the outside power meter on the trainer and do a varied workout and record both separately and compare the power outputs of both. If you can adjust the slope of your power meter and bring it to within a reasonable difference of your trainer you can continue using separate bikes (and power meters) indoors and out. @dcrainmaker has a power file comparison tool on his website and quite possibly articles or videos on this whole topic. Hopefully someone else will chime in with more detail on the specifics.
Maybe. This is hard because not many people have this device. With new measurement devices, you get into a chicken/egg situation.
I think thereās lower hanging fruit at first to feed into the model.
Yes, this is what the survey response is for. You give an RPE response of 1-5.
This is really important for us to know. Imagine the changes that could happen if Palisade was a ā5 - All outā vs a ā1 - Easyā.
Yes, this is built into the workout classifier and why we couldnāt just use average interval power.
We looked at a bunch of workouts with short power spikes and had a human classify them to the best of their ability. This taught the ML to recognize ones that is clear someone is hitting the prescribed goal even if they are off by a few seconds (usually due to device lag).
That is very insightful because they are the exact same level 
.
Weāre hoping to get levels out to everyone on Monday, March 15th 2021. This is subject to bugs and could be pushed out.
3 Likes
I agree that very few people have muscle oxygen sensors (humon devices that are out there would also broadcast over ant+ so just as easy to read) but the data is just a simple stream of numbers like HR and FIT files already have a standardized place to store the data. So shouldnāt be that hard to just reuse the code for recording HR data to record to a different location. (this is in comparison to HRV that doesnāt just update once a second so is more complicated to capture) Just collect the data first. Add the ability to pair over Ant+ and if its paired save it. No UI changes to show the data. (at least at first) Just collect it so it can be looked at later
Plus if youāre the first platform to use the data you might get Moxy to push some people to use your platform. The current software that does this isnāt as nice to use.
Does the ML know what trainer is being used? My thinking is take two smart trainers one with high lag and one with low. Same person on both the recorded output would look different so the ML might treat them differently. My thinking is the ML should try to learn the person not their recordings, so that trends that are the same for a grouping of people could be learned and not assume the people are different just cause their recording is different.
We classified a lot of trainers. It should just work and classify correctly no matter what trainer you use.
Great to hear. Thanks, Nate and team!
If you talking about levels. This means the progression level chart and workouts are getting a number right ? 
I cant wait!!
Damn, this ML is good. It has me validating AT without being in beta yetā¦ 
Levels will be out but the progression chart wonāt be out yet.
this is so cool.
Hmmm how does one can āliveā without the other.
If you know the level of an upcoming workout, but not you current level. You still donāt know if this is achievable,productive or stretch right? Or am I missing something else? 
Nate,
This is why it would be great if TR could capture these (plus left / right power where available, Core temperature sensors, etc.): start building up a base of data, so at some point in the future it could be investigated if this info is available. As you indicated, right now TR doesnāt have a lot of this data, so hard to see if it would be useful, and therefore no impetus for more people to buy one.
Iām guessing this is correct, but it doesnāt mean it is entirely useless. You can look at previous workouts in your current block that match the energy system of the upcoming workout. Compare the level of what you completed to the upcoming workout, and also reflect on the performance of that previous workout (was it too hard/too easy). That should give some idea of where you fall in that energy progression system. Then itās just a matter of filtering for workouts of that level in the desired system to pick something that fits what youāre looking for between achievable, productive, stretch and breakthrough.
Definitely takes more legwork than the full experience, but it could help if youāre struggling with certain types of workouts (in my case VO2).
Ok, since I havenāt found the golden ticket yet and canāt do the AI beta, Iāll just have to have someone guess what it would do so I can manually adjust as appropriate.
In week 5 of SSBII. Up to date, with the exception of one ride[1] Iāve completed all the workouts at 100% intensity without having to rest. Today intensity got to the level where I needed a break midway through the 3rd interval. My HR was approaching my max and I just didnāt have 5 more minutes. So I took a short 30 second or so break, then, another after the interval, and another after 6 minutes of the last interval, but I kept the intensity at 100% and was able to complete the workout. Pic for clarity:
So, at this point AI does what?
- lower my workout intensity for Mary Austion -1 on Friday?
- proceed as planned?
- change the next workout?
- Imposible to know what the ML will do??
[1] Last week I had a pretty intense outside ride on an off day that was extra to the training plan. The next dayās workout I did 3 of the intervals at 100% but I was just spent. I did the remaining two at 80% and that was pretty easy to doā¦ probably could have done 90% IDKā¦ but I felt like I had my TSS for the week already.
How will manual mid plan FTP increase/decrease be handled? For example if I feel over-unders or VO2Max workouts are getting too easy and bump my FTP by couple percent.