Timeline for Check if my time series is forecastable using Shannon entropy
Current License: CC BY-SA 4.0
10 events
| when toggle format | what | by | license | comment | |
|---|---|---|---|---|---|
| Mar 28 at 8:01 | comment | added | user78229 | It is the case in many applied predictive situations that 'ground truth' for the model either isn't available or, more simply, does not exist. In such cases model validation is obtained from the usefulness of the results: does it predict? are greater (lesser) values of a metric helpful or insightful? And so on. Wrt time series information theoretic metrics such as permutation entropy may be more useful than Shannon entropy. See Brandmaier's papers on PE jstatsoft.org/article/view/v067i05 | |
| Mar 28 at 7:50 | answer | added | Cryo | timeline score: 0 | |
| Mar 27 at 21:47 | comment | added | Marco | Cool. What do you mean by to simulate what I'm planning to do, to understand what entropies to expect? | |
| Mar 27 at 21:44 | comment | added | Cryo | This will answer your 'can I' question as well | |
| Mar 27 at 21:44 | comment | added | Cryo | Makes sense. It feels to me like you will still need to know what meaningful difference in entropy is. So it might be a good idea to simulate what you are planning to do, to understand what entropies to expect | |
| Mar 27 at 21:06 | comment | added | Marco | @Cryo I'm thinking about analyzing different sections of my windowed time series to see which are the most forecastable, so that I can then test these most forecastable parts via machine learning. Thanks. | |
| Mar 27 at 15:10 | comment | added | Cryo | For example, you calculate entropy and you get 42.0.... now what? | |
| Mar 27 at 15:09 | comment | added | Cryo | Sure you can. What are you going to do with that entropy? Entropy here is a statistic that is derived from your data. What you need to understand is what values of this statistic are important for you, and whether this statistic is sensitive enough to things you care about. My advice would be to generate some synthetic data which is either completely noise, or time-correlated. Process that data as you wish (windowing etc), calculate entropy, and then decide whether the obtained signal is useful | |
| S Mar 25 at 23:25 | review | First questions | |||
| Mar 26 at 1:03 | |||||
| S Mar 25 at 23:25 | history | asked | Marco | CC BY-SA 4.0 |