Sleep Tracking #1918
Replies: 8 comments 28 replies
-
|
Shameless plug: https://github.com/thiswillbeyourgithub/sleep_tracker_pinetime_wasp-os |
Beta Was this translation helpful? Give feedback.
-
|
Ok but it's for wasp-os, adding this functionality in infinitime would be more practical in my opinion. |
Beta Was this translation helpful? Give feedback.
-
|
I am wondering, I found nothing on Github about this topic before this post, has this idea never come up before? |
Beta Was this translation helpful? Give feedback.
-
|
It seems to me that the subject has already been broached, but to no avail. |
Beta Was this translation helpful? Give feedback.
-
|
That would be fantastic! I am already in the middle of developing a small prototype of this functionality, but of course it would be good to know how to best measure all this stuff 😅 . |
Beta Was this translation helpful? Give feedback.
-
|
Ah, I actually found an issue about this: #307 |
Beta Was this translation helpful? Give feedback.
-
|
indeed, they were concerned about the reliability of sensor support. Fortunately there has been progress although I find that it is still a little unstable |
Beta Was this translation helpful? Give feedback.
-
|
Just adding my 2 cents: There's a nice paper on the subject by Olivia Walch, specifically on the topic of sleep tracking with wearables using "raw" sensor data, intended to get some sleep tracking algorithms out in the free open source world: https://academic.oup.com/sleep/article/42/12/zsz180/5549536 Quick summary: they subjected a bunch of people to polysomnography (the gold standard for sleep tracking, in her words) while wearing an apple watch, which they managed to pull raw accelerometer and heart rate data from. They then run a bunch of different classifiers on the data collected with the watch, comparing their performance using the data from the polysomnogram as the "ground truth". They put a lot of work into comparing how well the classifiers work depending on how many data sources they use (accelerometer, heart rate, circadian clock estimate). As far as I recall, the accelerometer data holds the most information and actually goes a long way by itself. Heart rate alone is not worth much, but combining the two is a good improvement over accelerometer alone. There's a slight improvement in also including an estimate of the users' circadian clock, for instance through an activity sensor such as the step counter. I hacked together a small Python script using bleak to collect data from my PineTime. I had to buy an external bluetooth adapter and a long usb extension cable to get a reliable enough connection between my computer and watch throughout the night, my laptop internal adapter was simply too poor. It then worked without a hitch, though. The logging here is on the extreme side, since I'm getting every single update on both the accelerometer and the heart rate:
Walch has been kind enough to make both the data they collected and her code publicly available: https://github.com/ojwalch/sleep_classifiers I tried plotting a few of her datasets, and they do resemble the plot above. I think the PineTime heart rate could use a bit of filtering, though (or perhaps it's just because my sample rate is a bit high). Her code is mainly focused on generating the figures for the paper, and I had to spend some time getting the right versions of dependencies and such. My idea was to dig out the actual preprocessing and the parameters of their model after being trained, then try repeating that processing on my data and running the classifier. Of course I don't have any ground truth to compare with, but I think I could borrow a Garmin watch to compare with if nothing else :) A couple more papers I noted as candidates for future reading: |
Beta Was this translation helpful? Give feedback.
-
|
Sounds good! If you need it, here's the bluetooth adapter I had success with: https://www.amazon.com/Bluetooth-Adapter-Wireless-Transmitter-Receiver/dp/B09KG7QQ5V |
Beta Was this translation helpful? Give feedback.
-
|
thank you, but my pc already has bluetooth. |
Beta Was this translation helpful? Give feedback.
-
|
This PR implements that feature but is not merged yet: #1718 |
Beta Was this translation helpful? Give feedback.
-
I had some range issues initially, and I was not too keen on sleeping on top of my laptop :) In case you run into the same trouble, I had luck with that adapter.
Yes, as mentioned in #1918 (reply in thread) and linked by @Schweini07, I used #1718 to get continuous background measurements. |
Beta Was this translation helpful? Give feedback.
-
|
thank you, I will try to compile the version which is not yet merged but if I cannot do it I will try with the adapter |
Beta Was this translation helpful? Give feedback.
-
|
I created a repo to gather all the information, measurements, codes etc. https://github.com/MelvinRedondoTanis/SleepTracking-PineTime I haven't put much into it yet |
Beta Was this translation helpful? Give feedback.
-
|
In the end I didn't manage to compile. I could not yet do a test over a long period. for the moment I have not yet bought the adapter but in the meantime I tried to do a little function to detect a significant movement as simply as possible |
Beta Was this translation helpful? Give feedback.
-
|
Hi! |
Beta Was this translation helpful? Give feedback.
-
|
Thank you so much |
Beta Was this translation helpful? Give feedback.
-
|
Did you manage to log some data, @Schweini07? If so, did it resemble those I plotted earlier? I'm curious whether the noisy heart rate measurements in my plot were due to operator error :) |
Beta Was this translation helpful? Give feedback.
-
|
Here is an example of a night about two weeks ago. I sadly forgot to document what kind of interval I was using for the heart tracking each night, but I assume this was with the 10s interval. Although I think Continuous and 10s don't have much of a difference. This is the overall plot of the night, where one thing is quite obvious. The data only sometimes seems to be taken in a regular interval, other times there may be a difference of even an hour, before the next value is plotted. Here, we can see this in more detail. I am quite unsure to be honest, what might be the reason of this, especially because I feel the Gadgetbridge graph is far more accurate when using heart tracking during the day. |
Beta Was this translation helpful? Give feedback.
-
|
Thanks, interesting. This is logged using gadgetbridge? |
Beta Was this translation helpful? Give feedback.
-
|
Yes, it is logged through Gadgetbridge. It would be pretty weird though if the problem occurs because of a bluetooth connection issue, as my phone was always laying just half a meter next to the watch. |
Beta Was this translation helpful? Give feedback.
-
|
I ultimately think that the best thing is to work on an application dedicated to pinetime, even if simplified, sleep monitoring is too demanding for the power of pinetime |
Beta Was this translation helpful? Give feedback.
-
|
Though if we'd iron out these sensor bugs, we should be able to provide Gadgetbridge, or any other companion app, with the right data for sleep tracking, no? And I'd think the calculations shouldn't be much of a drain factor for the watch. With the sensor bug you mean that for 1-2 hours a night Gadgetbridge only shows the same Value, right? Because that is the problem I am having at least. |
Beta Was this translation helpful? Give feedback.
-
|
I think a PC application that takes PineTime sensor data in and generates sleep quality estimates out would be a good place to start. This would let us test the impact of sensor update interval on the sleep quality estimate and get a feel for the necessary computation and storage. It could also serve as a platform for choosing and tweaking sleep tracking algorithms. If at all possible, though, I agree with @Schweini07 that we should strive for a solution that can run on the PineTime.
This sounds like something we should dig into. Are there any open issues mentioning this? Is it possible the watch did not have good skin contact during the periods with poor readings? I did not see anything similar during my tests. |
Beta Was this translation helpful? Give feedback.
-
|
making a mini app to develop algorithms is a good solution for the moment. how could we communicate, a discord or something else? |
Beta Was this translation helpful? Give feedback.
-
|
A platform for easier communication would be quite nice. Discord should be fine, but if there is a better alternative that would be great too. I always feel like on Discord information gets so easily lost 😅 |
Beta Was this translation helpful? Give feedback.
-
|
I haven't found anything else, in the meantime let's use discord |
Beta Was this translation helpful? Give feedback.
-
Do you think it could be possible to add this functionality according to the following points ? :
score depends on :
low heart rate + weak movements = deep sleep
down
high heart rate + high movement = light sleep
example formula:
bpm * (movement/2)
Beta Was this translation helpful? Give feedback.
All reactions