1,683
edits
Changes
Talk:Finding relations between variables in time series (edit)
Revision as of 01:09, 22 May 2022
, 01:09, 22 May 2022→todo suggestion
mp April 7th at 1:51 PM
mp April 7th at 1:51 PM
@ oh an additional thought, post chat — I think a “fast up / slow down” pattern would be reflected in a Markov probability distribution where any given number has a low probability of later numbers being higher than it. e.g. if prior was 7, then 7-or-lower are likely (gradual decline), but 8+ very unlikely.I was struggling to remember the language here, but I think a “first order” Markov model is where probability distributions at any given step are based only according to the previous step (and no further “memory” in the system). A “second order” is influenced by the two previous steps (a bit more memory).
@ oh an additional thought, post chat — I think a “fast up / slow down” pattern would be reflected in a Markov probability distribution where any given number has a low probability of later numbers being higher than it. e.g. if prior was 7, then 7-or-lower are likely (gradual decline), but 8+ very unlikely.I was struggling to remember the language here, but I think a “first order” Markov model is where probability distributions at any given step are based only according to the previous step (and no further “memory” in the system). A “second order” is influenced by the two previous steps (a bit more memory).
First! Carefully compare variance internally to days and between days. If internal is too high this variable has too poor "distinctness". Can also look for "stability" in derivatives and between any time measured such as 12 hours or a month... or maybe strongly unlikely Markov model? Also maybe first, multimodality and outlier and anomalies.
== potential sources of solutions ==
== potential sources of solutions ==