Another nixtlar user @Mariana Menchero! 🎉 Seray Mirasci <Seray.Mirasci@lut.fi> has a question about something that's not working for them. Details in thread. Thank you!

thanks for letting me know @Tracy Teal from a first glance, it's because it is a tibble and the CRAN version only supports data frames and tsibbles

Is it right that tibbles work in the development version? I can point them to that to try.

I haven't forgotten about this, is just that there is more than one issue with this use case.

Thanks, got it, and no problem! I'll just let them know we're looking into it.

1. It’s an excel file, so I’m assuming they’re using

read_excel

, which loads the data as tibble. No support for that on CRAN, but yes on the dev version on GitHub. However: 1. They use a sequence 1,2,3,… as unique_ids, so instead of having one series with 260 observations,

nixtlar

thinks it’s 260 series with 1 observation each. 2. Frequency is seconds (!), so they need to specify it since

nixtlar

can only deduce hourly and above. 3. When using

as.Date()

, the hms part disappears. And even after making the necessary changes, the call still fails 🤦‍♀️ I’m currently looking into it, and for the time being, we can tell them that.

great, thank you so much for looking into this

No, it was actually funny to find some many issues in such a small dataset 😅

😄 a what not to do example

Hi @Tracy Teal, as promised, the solution for this issue has now been implemented in

nixtlar

, so please ask the user to download the development version:

devtools::install_github("Nixtla/nixtlar")

For this particular case, it is better to build a new data frame, specifying the format of the data, which is taken every second.

Y_df <- readxl::read_excel("path/to/data/Output_Fyz_rev.xlsx", sheet = 1)

df <- data.frame(
 unique_id = "id", 
 ds = seq(as.POSIXct(Y_df$ds[1]), by = "1 sec", length.out = nrow(Y_df)), 
 y = as.numeric(Y_df$y)
)

df$ds <- format(df$ds, format = "%Y-%m-%d %H:%M:%S")

Given that

nixtlar

can’t deduce subhourly frequencies, it is necessary to set it directly with

freq="S"

.

fc <- nixtla_client_forecast(df, h=8, id_col="unique_id", freq="S", level=c(80,95))

The forecast should look like this, same as with the Python SDK.

Thanks so much @Mariana Menchero! I'll share this.