forecasting with tsibble and fable

tidyverse
22

Thanks @mitchelloharawild! In my original data set, average of three models is performing the best. I am trying to use your suggested code here to use average as my main model. Initially, when I tried the following, it gave me this error:

Error: Can't convert distribution> to character>.

path <- ("FC")
df %>% 
  model(
    snaive = SNAIVE(Sales ~ lag("year")),
    ets = ETS(Sales),
    arima = ARIMA(Sales)
  )%>%
  mutate(average = (ets + arima + snaive) / 3)%>%
  forecast(h = "3 years")%>% 
  write_csv(path) # Write to a csv

# Read in the CSV
read_csv(path)

Then I tried taking mean and variance as per your suggestion. Variance didn't work, but mean worked and I was able to save it also.

path <- ("FC")
df %>% 
  model(
    snaive = SNAIVE(Sales ~ lag("year")),
    ets = ETS(Sales),
    arima = ARIMA(Sales)
  )%>%
  mutate(average = (ets + arima + snaive) / 3)%>%
  forecast(h = "3 years")%>% 
  summarise(mean = mean(Sales))%>%
  write_csv(path) # Write to a csv

# Read in the CSV
df1 <- read_csv(path)

Then, I updated my packages to see if variance works and now even mean doesn't work using the same code as above and gives the following error:

Error: The result is not a valid tsibble.
i Do you need as_tibble() to work with data frame?

I am struggling with saving the final results.

Thanks for your help!

23

Any suggestion on the above problem of saving the results!
Has anyone encountered this issue?

Thanks for helping!

24

I see. This is fixed in the development versions of the package. I'm planning to submit an update to CRAN soon.

25

Thanks @mitchelloharawild! Do I need to change anything in terms of my code mainly for saving the forercasted values?

26

The code above should work with the next version.

27

Perfect! Thank you @mitchelloharawild! Please do let me know when you update the next version as I cannot save my results until then and cannot proceed further :frowning:

So, waiting on you! Thanks again for your help!

28

The updated v0.2.0 distributional package is now on CRAN which should fix the file writing issue.

29

Thanks @mitchelloharawild! But I still have the same problems even after updating all packages:

  1. Error: Can't convert distribution> to character>. - (If I don't use mean to see the forecasted values)
  2. Error: The result is not a valid tsibble. (If I use mean to see the values)
    i Do you need as_tibble() to work with data frame?

Thanks again for your help!

30

Can you check what version of distributional you have?

packageVersion("distributional")
31

Yes, it is 0.2.0.

packageVersion("distributional")
[1] ‘0.2.0’

32

Could you try as.character(distributional::dist_normal(0,1))?

33

Yes, it works! It gives the mean values. and probably I can also get the intervals.

Thank you @mitchelloharawild!

34

This topic was automatically closed 7 days after the last reply. New replies are no longer allowed.