fitlm, how to use two datasets with identical size and timestamps

6 次查看（过去 30 天）

Dharmesh Joshi 2022-9-20

0
链接

此问题的直接链接

https://ww2.mathworks.cn/matlabcentral/answers/1808800-fitlm-how-to-use-two-datasets-with-identical-size-and-timestamps

评论： Dharmesh Joshi 2022-10-10

I have two data sets. Each dataset has the sensor values and a timestamp.

I would like to use fitlm to determin a linear function between both these sets of data, but the data used in this function should only be data that has timestamp in both two data sets, to that when data is being compared it at the same timestamp.

The issue is that a certain periods the dataset might have some data samples missing a certain time stamps.

This is my current code.

            ref_timetable_time=[ref_data_timestamp]';
            ref_timetable_ug=[ref_data_sensordata]';           
            ref_timetable = timetable(ref_timetable_time,ref_timetable_ug);
            timerperiod_Ref=ref_timetable(timerperiod,:);
            timetable_time=[live_t]';
            timetable_ug=[no2_ppb];           
            timetable_ug = timetable(timetable_time,timetable_ug);
            timerperiod_ug=timetable_ug(timerperiod,:);
            hourlyaveraged_no2_data = retime(timerperiod_ug,'hourly','mean');                       
            %model sensor data to ref data
            inModels{k} = fitlm( timerperiod_Ref.ref_timetable_ug, hourlyaveraged_no2_data.timetable_ug, "linear", "VarNames", ["Ref Data", "SensorValue"] );

What can be done so that dataset are equal in both size and timestamps?

Thank You

9 个评论
显示 7更早的评论隐藏 7更早的评论

dpb 2022-9-21

编辑：dpb 2022-9-21

retime will certainly "fill in" and synchronize two timeseries(*), but I see no reasonable way to create missing data prior to the beginning of the sensor time trace back to the time of the origin of the other trace--there isn't anything to go on to even guess what that might have been with the erratic nature of what data are there. I'd venture the only sensible option here is to begin both observations at the innitial time of the sensor trace and go on from there.

But, that done, I've no idea of what you must be thinking with the stated objective of "use fitlm to determine a linear function between both these sets of data" since there really appears to be almost no correlation whatever between the two traces. What model are you propoosing and what purpose would this fit be supposed to serve?

(*) There appears to be/may be a problem in the sensor data around Aug 21; the plot above shows a break in the line. That would indicate a NaN or Inf in either the data or the date vector -- plot() will draw a straight line between points if there were just a gap in time, but either variable being ~isfinite() will result in those points not being drawn; hence the apparent break. That would also produce NaN in the output of retime for those locations if don't use one of the options to handle missing data, explicitly.