Related

Race predictor makes no sense

6446252

As one of the pieces of feedback on fitness, I keep an eye on the numbers that Garmin's race predictor outputs, and note what causes the step improvements in predictions for future reference, as I do my strict 3 session a week program.

So yesterday (January 1st) when a more speed-focused session had gone well I was please to see the 5k prediction improve by 5 seconds to an equal recent peak (same as two days earlier after a more aerobic session), the best prediction for 6 months (after struggling back from injury)

But then glancing at my watch a few hours later (after midnight, January 2nd), the prediction had slowed by 7 seconds.  How can this make sense?  If the prediction was valid, nothing had invalidated it in those few hours.  And the new prediction was equal worst for the last two weeks, despite the last two sessions supposedly increasing the prediction to its best for six months!

It seems to me this prediction needs to be made more stable. Your expected performances don't improve and get worse in mutiple steps in a week, what they do is gradually improve in weeks where training is going well.   This can be achieved with a better design of the calculation, perhaps even as simply as smoothing the predictions over an appropriate time, but probably something a little better than that.  The result would be that the ups and downs actually mean something substantial, rather than being confusing noise.

  • 0

    I reported a bug in the past regarding to what numbers are displayed in the different places:

    Watch

    Garmin Connect app - Current

    Garmin Connect app - last dot of the 4 week graph

    Garmin Connect web - current

    Garmin Connect web - last dot of the 4 weeks graph

    In my case it's a forerunner 965. Things improved after a while, but it's not unheard of that a bug reappears, or that it was fixed on one device, but not in another.

    So I recommend you post more details (maybe with screenshots where relevant to show the different values) Make sure you sync before, so all 3 places have the latest data.

  • 0

    I'll add, that the comparison only makes sense after sync, and only if you look at the same time (i.e before you record another activity) As I see Garmin update the prediction after all relevant activities, so IMHO in theory it could change multiple times a day. 

    Also it's a totally different discussion whether the predictions are accurate. In my case they seem to be over optimistic all the time. Others reported it's too pessimistic for them, others that it's spot on. I don't know numbers, but I wouldn't be surprised if it's about 1/3 in each group ;)

    Probably because of this what's more interesting is to look at the tendency. This is what many (sometimes even official Garmin documents) say about many numbers your Garmin ecosystem . They are an estimate. Might not be a good estimate, but if you don't try to compare your numbers to someone else's, but rather to your own numbers in the past, then it's useful.

  • 0 in reply to flocsy🤠

    The values I see on my watch (a humble Forerunner 245) are the same I see via the Performance menu on Garmin Connect Web.

  • 0
    6446252 said:
    So yesterday (January 1st) when a more speed-focused session had gone well I was please to see the 5k prediction improve by 5 seconds to an equal recent peak (same as two days earlier after a more aerobic session), the best prediction for 6 months (after struggling back from injury)
    6446252 said:

    But then glancing at my watch a few hours later (after midnight, January 2nd), the prediction had slowed by 7 seconds.  How can this make sense?  If the prediction was valid, nothing had invalidated it in those few hours.

    I think the possible explanation for this is:

    - race predictions are partly based on VO2 Max. (Years ago, they were wholly based on VO2 Max to the extent that any 2 Garmin users with the same VO2 Max would have the exact race predictions. At some point, Garmin improved the algorithm so your mileage is also taken into account.)

    - VO2 Max is some sort of rolling average (this is an educated guess on the fact that it never changes too much from activity to activity)

    - in this specific case, your VO2 Max changed slightly without doing any activities because the oldest VO2 Max-contributing activit(ies) for the Jan 1 predictions were no longer taken into account on Jan 2

  • 0 in reply to flowstate

    You could try to verify what I'm saying by looking at your VO2 Max in the Connect app and seeing whether the fractional value - which you can eyeball on the graph - has decreased slightly.

    Unfortunately there's no way view / export fractional VO2 Max numbers in the Connect app or website with the kind of frequency that you would need to see - e.g. day by day or activity by activity. Garmin never shows fractional VO2 Max numbers in the app, website or watch. You can export VO2 Max in the website with 1 decimal place, but the best frequency is something like 1 value per week or something.

    If you really want to see fractional VO2 Max with a per-activity resolution, you can export activity FIT files and view them in fitfileviewer.com. You will see that there are 2 VO2 Max numbers in the file, with 2 decimal places each:

    - user vo2 max: this is equivalent to your VO2 Max *before* the activity. It's the same as what you see in the Connect app/website and the vo2 max graph on the watch. It does seem to be some sort of rolling average

    - session vo2 max: this is "the VO2 Max of the activity". It's not necessarily exactly equal to user vo2 max (either before or after the activity), but it is usually very close. It does appear to be some sort of rolling average as well, and my guess is that it contributes to the user vo2 max (obviously haha).

    If you want to see your *current* (user) VO2 Max, you would have to record another activity (I guess it could be a dummy activity).

    What I don't think is possible to see is the "raw" session VO2 Max for an activity (i.e. what the value would be if it *wasn't* a rolling average, but if it only applied purely to a single activity).