0
$\begingroup$

Following is count data needed to model with a nonhomogeneous Poisson process (NHPP).

enter image description here

For each row-wise e.g. 8:15-8:30, every 12 intervals of 15 minutes size and 15 days of data collected. I have already fitted each row-wise data with other distributions.

But now the problem is to fit the same data with NHPP. I tried fitting following data as,

data = {{4, 13, 23, 0, 1, 3, 10, 17, 26, 20, 3, 0, 30, 0, 2}, {27, 2, 18, 15,
   24, 19, 3, 7, 13, 21, 14, 3, 6, 8, 22}, {28, 4, 23, 10, 26, 13, 3, 
  15, 9, 22, 16, 25, 5, 17, 21}, {24, 24, 26, 9, 25, 11, 23, 22, 11, 
  16, 19, 3, 17, 7, 17}, {3, 7, 26, 21, 29, 1, 0, 27, 11, 28, 22, 1, 
  27, 1, 14}, {10, 18, 13, 19, 18, 1, 27, 16, 28, 1, 29, 24, 10, 20, 
  15}, {25, 6, 5, 9, 21, 26, 25, 5, 14, 1, 28, 8, 28, 18, 18}, {7, 17,
   9, 30, 26, 16, 21, 27, 12, 17, 20, 26, 30, 12, 24}, {14, 16, 3, 27,
   19, 13, 8, 30, 8, 15, 4, 3, 20, 27, 0}, {23, 15, 28, 9, 29, 7, 30, 
  30, 21, 30, 27, 29, 23, 0, 19}, {4, 0, 9, 6, 18, 19, 1, 23, 1, 12, 
  12, 17, 6, 27, 29}, {22, 22, 9, 21, 1, 4, 25, 8, 30, 7, 20, 11, 14, 
  26, 22}}
  1. First by estimating the arrival rate in each interval with
    rate = Join[{0,Round[(Total@data[[#]]/(15*15))&/Range[12],0.01]}]//Flatten
  2. Here i considering the interval into min form {0, 15, 30,...,180}, by int = Table[15*i, {i, 0, 12}]
  3. Then I paired upper limit and rate in each interval as {{0,0}, {15,0.68},{30,0.9},...{180,1.08}}.
  4. Such list is then partition into following list of sublists piecewise = Partition[Transpose[{time1, rate}], 2, 1]
  5. LinearEqs = Simplify[InterpolatingPolynomial[piecewise[[#]], x]]&/@Range[Length[piecewise]] applied to get intensity function for each interval.
  6. lower limit as ll = Table[15*i, {i, 0, 11}] and upper limit as ul = Table[15*i, {i, 1, 12}].
  7. ExptdMean = Integrate[LinearEqs[[#]], {x, ll[[#]], ul[[#]]}] & /@ Range[12] for each interval.

But I have been told to do column-wise fitting over the time period (0,T] i.e. (8:00-11:00].

Now that put into me confusion whether it is from Day1 all the way from 8:00-11:00, then Day2's 8:00-11:00 and so on for 15 days.

If considering examples like below then I have count column as tot = Total /@ data set with ul as x and exposure as 15 mins.

enter image description here

But again I am not sure that 15 days of data can fit in such way.
I'm baffled about count data fitting with NHPP.

Please, if anyone can help me with the correct approach for fitting column-wise data to the NHPP model.

Thanks for your time.

$\endgroup$
  • $\begingroup$ You mention "the NHPP model" but you never state a specific model. One can't make inferences by just massaging the data in isolation of a specific model. While my answer to your question mathematica.stackexchange.com/questions/218618/… might be inadequate for your needs, it does at least indirectly state a model through the GeneralizedLinearModelFit function. (Note that your massaging of the data induces dependence among the massaged observations in which most models assume independence.) $\endgroup$ – JimB Apr 7 '20 at 16:25
  • $\begingroup$ @JimB I'm learning about NHPP. It would be courteous if some procedures I can learn specifically using InhomogeneousPoissoProcess[] with testing on such sort of data or data available for the experiments. Certainly, your answer to the previous post has noticed :) $\endgroup$ – step-by-step Apr 8 '20 at 12:59
  • 1
    $\begingroup$ Your questions are good. However, it seems to me that you first need advice on what statistical procedures to use as opposed to getting help with Mathematica. Because of that your question might be closed. I suggest asking your question on CrossValidated (stats.stackexchange.com). $\endgroup$ – JimB Apr 8 '20 at 16:19

Your Answer

By clicking “Post Your Answer”, you agree to our terms of service, privacy policy and cookie policy

Browse other questions tagged or ask your own question.