Non-equidistance intervals

roqua / physiqual

Ruby Engine for merging various datasources with diary questionnaire data

MIT License

1 stars 1 forks source link

Non-equidistance intervals #16

Closed frbl closed 8 years ago

frbl commented 9 years ago

Currently physiqual only offers support for protocols where the measurements are a given distance apart. We could also introduce either:

Random measurement moments, where the scheduling moments are given using a json file (for example)
Non equidistance measurement points, that is, one at 10, one at 11, one at 5 and one at 12 (every day the same)

emerencia commented 9 years ago

You can't use non-equidistant measurement points in any time series analysis of the data (and I don't think there's many other analyses that can be done with repeated measurements). So necessarily, when you measure at random intervals, or at distinct intervals during the day, there will be a later step in the process where you resample those values to equidistant intervals. If people use Physiqual to add physical data to their existing measurement points, they might as well skip the step where they match the data up with possibly random intervals and ask of physiqual directly the equidistant intervals. Doing it the other way (asking physical to return the data in nonequidistant time intervals and then resampling those to equidistant time intervals) would lead to more loss in precision.

frbl commented 9 years ago

Well, they do not need to use a time series analysis technique. What about 3mpca or multilevel modelling? I think both work with non-equidistant measurement points. Physiqual focuses on ESM/EMA data, not on time-series analysis I guess.

emerencia commented 9 years ago

If those techniques don't go through the same resampling process then sure. I would say add it if it's not too much work.

frbl commented 8 years ago

Although we are able to generate a schedule which is not equidistant, it is not possible to use these buckets in the current clustering method. This method should be improved so it only depends on the data provided to it, instead of assuming equidistance.

emerencia commented 8 years ago

The consensus we reached was to not support the non-equidistance intervals for now.

Do we want to remove/comment out the DailyBucketGenerator class or just leave it there unused? Or maybe just add a comment that the code is currently not used/may be used in a future version?

frbl commented 8 years ago

Let's remove it.

emerencia commented 8 years ago

Agreed.

frbl commented 8 years ago

fixed in #71