Should time series have regular sample periods?

[markNZed]

Consider the attached example of time series realTraffic/speed_6005 and we can see a straight line that reflects the following data points, starting line 700:

2015-09-04 22:34:00,88
2015-09-04 22:41:00,92
2015-09-08 10:44:00,94
2015-09-08 10:49:00,94

Notice the time step goes from 5-10 minutes to 4 days and back again.

This is not indicated as an anomaly, nor does the Numenta detector indicate an anomaly (I guess it is just processing samples and not encoding the time of the samples?)

Should the time series have regular sample periods or not?