markovchain
markovchain copied to clipboard
old issues regarding inference
Dear Dr Spedicato
Many, many thanks for the providing the markovchain package!! I have always wanted to teach myself a bit of markov chains but derving the TPM from a 'relative frequency' approach of manually counting the transitions always put me off for the data i wanted to play with! So thanks!!
So I have been playing around with your package... and there has been some joy... problems and confusion. Hence, i have some questions, i hope you wont mind responding. I am not a statistician so forgive me for some crazy/silly questions. Im more into hydrology.
i have attached a script of what i have been experimenting with, which is drought class transition data for a met station in a catchment in Zimbabwe, Africa. My issues are as follows:
Package Errors
- I am facing problems with running the function > verifyMarkovProperty(class_an). As in the attached script. This error is popping up:
Error in [<-
(*tmp*
, i, 1, value = SSO[i]) : subscript out of bounds
i have tried modifying my data format many times and it keeps popping up.. Can you please check it?
- Some times when i run the tests in the package with subsets of the data, i get NAs when running other assess... functions e.g. - assessOrder, assessStationarity.... what could be the issues?
suggestion
if you are calculating the fundmental matrix behind the scenes, could you consider giving the user to pop it out, as a function!!
question from package
- Dr Spedicato, how is that first passage time interpreted? i am a bit confused with it! I thought it would be one value, >= one? You could examplify with what i have implemented in the script attached!
Maybe, i am blinded by the desire of wanting to answer the question, e.g. if i start in drought state D4, D3, D2, (as in my case) how long will it take to transit to the good state, D0? -- is First Passage Time for answering such a question?
- The warnings "In chisq.test(mat) : Chi-squared approximation may be incorrect" keep pop up not matter what data i have tried. Should they be taken seriously or its just info? when would you say they need to be taken seriously - maybe, in terms of the test statistic or degrees of freedom?
general questions
some general questions from a beginner...
-
Using your R package, if i want to set up a TPM using saying one/two more series, how will i do about it? For example, drought classes transition conditioned on global indicies, either ENSO or the Indian diapole or both?
-
I would like to run some simulations with the TPM for say 1-5 years of my data, but the TPM is drawn from the same data.... is there a way to go around the complications that arise from this approach?
-
the homogenous TPM maintains the time step of the original data right?
Thanks for the package!!!
Regards
Donald