Matching for timestamped data #1098
-
Hi, I am wondering whether it is possible to use stumpy for matching time series where the time information is supplied as a timestamp, rather than the raw position in the series. What I mean is a series where each element has the following form: (8, -10.369417) where the first element is the timestamp, in seconds. This is audio data - spectral peaks, in fact. So there could be gaps in the series, like this: ((8, -10.369417), (9, -11.487215), (12, -8.715628)...) Is it possible to do this? Thanks |
Beta Was this translation helpful? Give feedback.
Replies: 1 comment 2 replies
-
@relativeflux Thank you for your question and welcome to the STUMPY community. In general, as you've already identified, matrix profiles don't actually take "time" as an input and, instead, simply takes in a sequence of raw data points as input. In other words, matrix profiles assume that your data points are evenly spaced apart. The most natural thing for you to do is to impute the missing values in between and then apply the matrix profile thereafter. |
Beta Was this translation helpful? Give feedback.
@relativeflux Thank you for your question and welcome to the STUMPY community. In general, as you've already identified, matrix profiles don't actually take "time" as an input and, instead, simply takes in a sequence of raw data points as input. In other words, matrix profiles assume that your data points are evenly spaced apart. The most natural thing for you to do is to impute the missing values in between and then apply the matrix profile thereafter.