3 ms·
"Time series" is such an over-subscribed term. What sorts of time series is this actually useful for? For instance, will it be able to predict dynamics for a m
by uoaei 2y ago
"Time series" is such an over-subscribed term. What sorts of time series is this actually useful for?
For instance, will it be able to predict dynamics for a machine with thousands of sensors?
- techwizrd 2y agoSpecifically, its referring to univariate, contiguous point forecasts. Honestly, I'm a little puzzled by the benchmarks.
- sarusso 2y agoEven if it was for multivariate time series, the model would first need to infer what machine are we talking about, then its working conditions, and only then make a reasonable forecast based on an hypothesis of its dynamics. I don’t know, seems pretty hard.
- uoaei 2y agoIndeed. An issue I ran into over and over while doing research for semiconductor manufacturing. My complaint was more illustrative than earnest.