I want to do a sleep analysis for the user in my app.And I think the CoreMotion Framewrok should help.
func queryActivityStartingFromDate(start: NSDate!, toDate end: NSDate!, toQueue queue: NSOperationQueue!, withHandler handler: CMMotionActivityQueryHandler!)
So I use the API above to get the user motion data in the last 7 days.And Now I get a list of CMMotionActivity Object.
And My question is how to calculate the user sleep status from these thousands of CMMotionActivity Object.Is there any algorithms? Or any other way to achieve sleep analysis.
Many Thanks!
CMMotionActivity includes a stationary property which might be useful in your case.
Look for contiguous periods of inactivity, paired with location, timezone & CMDeviceMotion data you should begin to detect patterns in the dataset you have. use statistical variation to define thresholds and fine-tune your results.
Caveat, you will make assumptions which might not be true. Some people sleep in moving vehicles for instance.
I found this pdf useful ftp://ftp.tik.ee.ethz.ch/pub/students/2010-HS/SA-2010-26.pdf
Related
I have a time series of events. The events carry a timestamp and some other metadata. I want to create clusters/buckets of these events based on their temporal proximity in order to discard some redundant events. For example, if I have a cluster of 10 events which are temporally close to each other - let's define temporal proximity as events being no further than x temporal units away from each other -, I can choose one of them as a representative and discard all others. For example, I have the following events
e0-e1-e2-------e3------------------------e5-e6-e7-e8--------------e9-e10
A sample bucketization would be:
(e0,e1,e2)
e3
(e5,e6,e7,e8)
(e9, e10)
I tried to represent the proximity of the events using the dashes above.
I know that there are some elaborate ways to go about this like DBSCAN or k-means but I was hoping that there would be something more straight forward, without any extensive prior knowledge of ML.
I also thought of a sliding window implementation, but the problem becomes quickly very complicated in terms of how to choose the window size and the shifts of the window...
Is there a standard way of doing this?
I am trying to implement an SLM app for iOS using AudioKit. Therefore I need to determine different loudness values to a) display the current loudness (averaged over a second) and b) do further calculations (e.g. to calculate the "Equivalent Continuous Sound Level" over a longer time span). The app should be able to track frequency-weighted decibel values like dB(A) and dB(C).
I do understand that some of the issues im facing are related to my general lack of understanding in the field of signal and audio processing. My question is how one would approach this task with AudioKit. I will describe my current process and would like to get some input:
Create an instance of AKMicrophone and a AKFrequencyTracker on this microphone
Create a Timer instance with some interval (currently 1/48_000.0)
Inside the timer: retrieve the amplitude and frequency. Calculate a decibel value from the amplitude with 20 * log10(amplitude) + calibrationOffset (calibration offset will be determined per device model with the help of a professional SLM). Calculate offsets for the retrieved frequency according to frequency-weighting (A and C) and apply these to the initial dB value. Store dB, dB(A) and dB(C) values in an array.
Calculate the average for arrays over the give timeframe (1 second).
I read somewhere else that using a Timer this is not the best approach. What else is there that I could use for the "sampling"? What exactly is the frequency of AKFrequencyTracker? Will this frequency be sufficient to determine dB(A) and dB(C) values or will I need an AKFFTTap for this? How are values retrieved from the AKFrequencyTracker averaged, i.e. what time frame is used for the RMS?
Possibly related questions: Get dB(a) level from AudioKit in swift, AudioKit FFT conversion to dB?
I have location data from a large number of users (hundreds of thousands). I store the current position and a few historical data points (minute data going back one hour).
How would I go about detecting crowds that gather around natural events like birthday parties etc.? Even smaller crowds (let's say starting from 5 people) should be detected.
The algorithm needs to work in almost real time (or at least once a minute) to detect crowds as they happen.
I have looked into many cluster analysis algorithms, but most of them seem like a bad choice. They either take too long (I have seen O(n^3) and O(2^n)) or need to know how many clusters there are beforehand.
Can someone help me? Thank you!
Let each user be it's own cluster. When she gets within distance R to another user form a new cluster and separate again when the person leaves. You have your event when:
Number of people is greater than N
They are in the same place for the timer greater than T
The party is not moving (might indicate a public transport)
It's not located in public service buildings (hospital, school etc.)
(good number of other conditions)
One minute is plenty of time to get it done even on hundreds of thousands of people. In naive implementation it would be O(n^2), but mind there is no point in comparing location of each individual, only those in close neighbourhood. In first approximation you can divide the "world" into sectors, which also makes it easy to make the task parallel - and in turn easily scale. More users? Just add a few more nodes and downscale.
One idea would be to think in terms of 'mass' and centre of gravity. First of all, do not mark something as event until the mass is not greater than e.g. 15 units. Sure, location is imprecise, but in case of events it should average around centre of the event. If your cluster grows in any direction without adding substantial mass, then most likely it isn't right. Look at methods like DBSCAN (density-based clustering), good inspiration can be also taken from physical systems, even Ising model (here you think in terms of temperature and "flipping" someone to join the crowd)ale at time of limited activity.
How to avoid "single-linkage problem" mentioned by author in comments? One idea would be to think in terms of 'mass' and centre of gravity. First of all, do not mark something as event until the mass is not greater than e.g. 15 units. Sure, location is imprecise, but in case of events it should average around centre of the event. If your cluster grows in any direction without adding substantial mass, then most likely it isn't right. Look at methods like DBSCAN (density-based clustering), good inspiration can be also taken from physical systems, even Ising model (here you think in terms of temperature and "flipping" someone to join the crowd). It is not a novel problem and I am sure there are papers that cover it (partially), e.g. Is There a Crowd? Experiences in Using Density-Based Clustering and Outlier Detection.
There is little use in doing a full clustering.
Just uses good database index.
Keep a database of the current positions.
Whenever you get a new coordinate, query the database with the desired radius, say 50 meters. A good index will do this in O(log n) for a small radius. If you get enough results, this may be an event, or someone joining an ongoing event.
a relatively simple question that I've not been able to find a clear answer to. My app is more complex, but answering this question will suffice.
Suppose you're writing a stopwatch app. When the user taps "start", the app stores the current date and time in startTime:
startTime = [NSDate date];
When the user tapes "stop", the app stores the current date and time in stopTime:
stopTime = [NSDate date];
The duration is calculated by:
duration = [stopTime timeIntervalSinceDate:startTime];
and is displayed with something like:
[durationLabel setText:[NSString stringWithFormat:#"%1.2f", duration]];
The typical durations that my app is timing range from 2 to 50 seconds. I need accuracy to 1/100th of a second (e.g. 2.86 seconds).
I'm assuming that there is some protocol that iOS devices use to keep their clocks accurate (cellular or NTP sources?). My concern is that between starting and stopping the stopwatch, the clock on the iOS device is updated which can result in a shift of the current date/time either ahead or back. If this were to happen, the duration calculated would be inaccurate.
I've seen a few posts relating to timing methods for purposes of improving code efficiency. Some suggest using mach_time.h functions, which I'm not familiar with. It's not obvious to me which is the best approach to use.
Is it possible to disable iOS from updating the date & time? Is mach_absolute_time() unaffected by iOS clock updates?
Many thanks!
Tim
You are correct in thinking that CFAbsoluteTime and its derivatives (NSDate dateand so on) are potentially skewed by network updates on 'real' time. Add that to the fact that NSTimer has an accuracy of 50-100ms and you have a timer that is not suited to the most critical of time-sensitive operations.
The answer to this problem seems to be CACurrentMediaTime.
It is a member of the Core Animation group, but there shouldn't be any problem integrating it into non-animation based applications.
CACurrentMediaTime is a wrapper of mach_absolute_time() and makes sense of the "mach absolute time unit," which from my understanding is no fun to tinker with. mach_absolute_time() is calculated by running a non-network synced timer since the device was last booted.
There is relatively little information on CACurrentMediaTime but here are some sources and further reading:
Apple's sparse documentation of CACurrentMediaTime
Stack Overflow - NSTimer vs CACurrentMediaTime()
http://bendodsonapps.com/weblog/2013/01/29/ca-current-media-time/
http://blog.spacemanlabs.com/2011/09/all-in-the-timing-keeping-track-of-time-passed-on-ios/
http://forum.sparrow-framework.org/topic/accurate-timer
Note: If you do use CACurrentMediaTime, make sure you include and link the QuartzCore.framework
Check out this here. I would say forget about the current time check and use a precision timer since it won't rely on the current time but instead uses an interval.
Most people are limited to about 5 or 6 locations on a daily basis (work, home, school, store, etc). I want to speed up address display by caching a few of these most visited locations. I've been able to get the address info using both google maps GPS and JSON and Locator.reverseGeocode. What would be the best way to cache this information and to check proximity quickly? I found this GPS distance calculation example and have it working. Is there a faster way to check for proximity?
Please see similar question first: Optimization of a distance calculation function
There are several things we can change in distance calculations to improve performance:
Measure device speed and decrease or increase period of proximity test accordingly
Trigonometric calculations takes most of performence, but it may done much faster. First make bold distance calculations using lookup table method, then if distance is less than proximity limit + uncertainty limit, use CORDIC method for more precise calculation.
Use constants for Math.PI/180.0 and 180.0/Math.PI
several links that may be helpful:
Very useful explanations of CORDIC, especially doc from Parallax for dummies
Fast transcendent / trigonometric functions for Java
Cordic.java at Trac by Thomas B. Preusser
Cordic.java at seng440 proj
Sin/Cos look-up table source at processing.org by toxi