I have two stationery time series that I want to see are they correlated or not. I decided to work with cross correlation, there is a good answer in Correlation between two time series but I dont know how to calculate p-value(as http://stats.stackexchange.com/users/11032/michael-chernick said) and decide that its a good answer, if there are correlated do something. I dont know how to do this. I want to code it with C++ and use it for simulation in Omnet++ . Could someone help me to find a way or introduce a document to me to understand the mathematical way to calculate it and program it in C++? P.s : My data are the number of packets which send from two nodes to an OLT in a network continuously. something like this:
from time 0 to 10:
series 1 : 2 5 3 7 9 2 0 1 8 4
series 2: 3 4 6 0 5 9 2 5 3 7 . series 2 has 6 min delay. or maybe there is no delay in other case, I must calculate the delay also (I can calculate delay with omnet++ but I dont know are they correlated at the end or not)
http://stats.stackexchange.com/users/11032/michael-chernick; http://stats.stackexchange.com/questions/29096/correlation-between-two-time-series
Hello Shirin,
Is it must to apply your code in C++? I have some discussions and functions to apply in MATLAB. They could be used to solve your problem and calculate cross correlation between two time series with delay.
Please check the links below:
http://www.mathworks.com/matlabcentral/newsreader/view_thread/282900
http://www.mathworks.com/help/econ/crosscorr.html
Good luck
Thanks you Hasan,
Yes it must be in C++ but if I understand how it works mathematically and know the exact formula, I can implement it in C++ by myself. Thanks for the links.
You can find the formulas within the discussions. By the way, MATLAB is adopting C code, therefore, you will be able to transfer it to C++ easily.
do you know Granger causality test?
It is a correlation test that considers a lag between two time series.
We have a paper using this: "Uncovering Causal Relationships between Bugs and Software Metrics", CSMR'12
Try Cross-Recurrence, in this way you can forget about models and delays, simply you can check the observed cross recurrence against a set of random shuffled versions of the same data, here below a practical application:
http://www.biomedcentral.com/1472-6807/9/48/abstract
a theoretical introduction to the theme: (together with software):
http://www.recurrence-plot.tk/
and (attached) a demonstration of the efficiency of the method in picking up very subtle signals..
Thanks a lot for you help. Now I am reading all these information to figure out the best way for my problem, The two time series that I have are coming from different sources and may be similar or may not be. maybe shifted maybe not and there is no noise. The final decision must be used in Omnet++ simulator and is a part of simulation code. Till now the fastest and easiest way that is suitable for Omnet++ is Hassan khaleel 's answer which is cross-correlation.
Thanks a lot , I learned a lot from your answers.
Did you figure out how to calculate the p value for cross-correlations?
- Badges
- Science topic
- Similar topics
- Engineering
- Electrical Engineering
- Signal Processing
- Image Processing