Prophet: Encounter "Stan does not support NA" when using prophet function

[cawoodjm]

OS: Windows 10 Pro, R version: 3.3.3, Gui: RStudio
Screen shot attached of RStudio code and execution, plus the used data set pm.xlsx

The issue can be seen in the attached images. When I load a similar Wikipedia page traffic data set, and use the prophet function I receive the following error. The steps used was sourced from the following URL. (https://facebookincubator.github.io/prophet/docs/quick_start.html#r-api)

"Error in FUN(x[[i]],...) : Stan does not support NA (in t) in data failed to preprocess the data; optimization not done
Error in matrix(m$params[[name]], nrow = n.iteration) : 'data' must be of a vector type, was 'NULL'"

When I validate my source data set I do not see null or NA values. Would this error be contributed to a formatting error when the prophet function runs?

pm.xlsx

[bletham]

It looks like parsing the dates is failing. We use the zoo package to convert the ds column to date format, and you can see that this is introducing a bunch of NAs if you run:

ds <- zoo::as.Date(history$ds)

You can see the issue by trying to run it on a single date:

> zoo::as.Date('8/13/2015')
Error in charToDate(x) : 
  character string is not in a standard unambiguous format

The solution is to convert it to date format before passing it in to Prophet.

https://github.com/facebookincubator/prophet/commit/68f88e2b733d2f91769359e7e245472ccc87a2bf raises an exception if the date parsing fails, which should make this issue less opaque in the future.

[JoeReis]

I've tried the steps of correcting date and y formats and removing nulls, and still get this error message:
Error in FUN(X[[i]], ...) : Stan does not support NA (in y) in data failed to preprocess the data; optimization not done Error in matrix(m$params[[name]], nrow = n.iteration) : 'data' must be of a vector type, was 'NULL' Called from: matrix(m$params[[name]], nrow = n.iteration)

Also, my code is in a loop and tried to fix things with the suggestions in this link (https://github.com/facebookincubator/prophet/issues/93), but still to no avail.

Only breaking on loops of 12 iterations.

[bletham]

Can you check what version of prophet you are using? Is there any chance you can post the data that is failing so that I can replicate?

[JoeReis]

I'm using prophet version 0.1.1.

I managed to pinpoint the issue, which occurs with non-unique y-values. For example, it breaks if the y-value is all 0.

[bletham]

Thanks, we do currently scale data by dividing by max(abs(y)), which divides by zero if all y's are 0. This needs to be fixed.

[bletham]

I'm going to close the issue here to consolidate it with #251 , where it will be fixed.