CODEWITHSUNDEEP
rtime-seriesarima
importm
importm

Reputation: 305

autoarima train and test, grouped by id in r

I'm trying to forecast time series with auto.arima. What I need is to split in train and test data to see the model metrics. my date range is between December 2016 and January 2020. I need the train data to be until December 2018 and the test after this.

Along with this, I need the metrics RMSE and MSE for each id. This is an example of my data: 

x<-    tibble::tribble(
      ~ID,        ~Date,  ~Value,
       1L, "01-12-2016",     48L,
       1L, "01-01-2017",  10055L,
       1L, "01-02-2017",    650L,
       1L, "01-03-2017",   8255L,
       1L, "01-04-2017",   3680L,
       1L, "01-05-2017",   2180L,
       1L, "01-06-2017",   2790L,
       1L, "01-07-2017",   3805L,
       1L, "01-08-2017",   2811L,
       1L, "01-09-2017",   -225L,
       1L, "01-10-2017",   -232L,
       1L, "01-11-2017",   -243L,
       1L, "01-12-2017",   -217L,
       1L, "01-01-2018",   -256L,
       1L, "01-02-2018",   -277L,
       1L, "01-03-2018",     -3L,
       1L, "01-04-2018",   -247L,
       1L, "01-05-2018",     88L,
       1L, "01-06-2018",   -260L,
       1L, "01-07-2018",   -228L,
       1L, "01-08-2018",   -285L,
       1L, "01-09-2018",   -321L,
       1L, "01-10-2018",   -265L,
       1L, "01-11-2018",   -302L,
       1L, "01-12-2018", -11968L,
       1L, "01-01-2019",   5435L,
       1L, "01-02-2019",   6694L,
       1L, "01-03-2019",   4750L,
       1L, "01-04-2019",   3747L,
       1L, "01-05-2019",   3727L,
       1L, "01-06-2019",   3252L,
       1L, "01-07-2019",   1691L,
       1L, "01-08-2019",   2489L,
       1L, "01-09-2019",   -182L,
       1L, "01-10-2019",   3926L,
       1L, "01-11-2019",    326L,
       1L, "01-12-2019",  -1047L,
       1L, "01-01-2020",     14L,
       2L, "01-12-2016",   -241L,
       2L, "01-01-2017",   -262L,
       2L, "01-02-2017",   -231L,
       2L, "01-03-2017",   -203L,
       2L, "01-04-2017",   -226L,
       2L, "01-05-2017",   -223L,
       2L, "01-06-2017",   -300L,
       2L, "01-07-2017",   -259L,
       2L, "01-08-2017",   -241L,
       2L, "01-09-2017",   -225L,
       2L, "01-10-2017",   -227L,
       2L, "01-11-2017",   -243L,
       2L, "01-12-2017",   -217L,
       2L, "01-01-2018",   -256L,
       2L, "01-02-2018",   -277L,
       2L, "01-03-2018",      0L,
       2L, "01-04-2018",   -247L,
       2L, "01-05-2018",   -274L,
       2L, "01-06-2018",   -264L,
       2L, "01-07-2018",   -227L,
       2L, "01-08-2018",   -275L,
       2L, "01-09-2018",   -325L,
       2L, "01-10-2018",   -269L,
       2L, "01-11-2018",   -306L,
       2L, "01-12-2018",   -264L,
       2L, "01-01-2019",   -308L,
       2L, "01-02-2019",   -332L,
       2L, "01-03-2019",   -260L,
       2L, "01-04-2019",   -300L,
       2L, "01-05-2019",   -302L,
       2L, "01-06-2019",   -291L,
       2L, "01-07-2019",   -284L,
       2L, "01-08-2019",   -288L,
       2L, "01-09-2019",   -272L,
       2L, "01-10-2019",      0L,
       2L, "01-11-2019",      0L,
       2L, "01-12-2019", -17107L,
       2L, "01-01-2020",   3500L,
       3L, "01-12-2016",   1940L,
       3L, "01-01-2017",   1753L,
       3L, "01-02-2017",   2758L,
       3L, "01-03-2017",   2539L,
       3L, "01-04-2017",  -9078L,
       3L, "01-05-2017",   5215L,
       3L, "01-06-2017",   1796L,
       3L, "01-07-2017",  -8424L,
       3L, "01-08-2017",  19868L,
       3L, "01-09-2017",  10707L,
       3L, "01-10-2017",   8985L,
       3L, "01-11-2017",   3058L,
       3L, "01-12-2017",   2469L,
       3L, "01-01-2018",     21L,
       3L, "01-02-2018",   1039L,
       3L, "01-03-2018",   2875L,
       3L, "01-04-2018",  -2678L,
       3L, "01-05-2018",   1515L,
       3L, "01-06-2018",   2651L,
       3L, "01-07-2018",  -5014L,
       3L, "01-08-2018",    299L,
       3L, "01-09-2018",   1755L,
       3L, "01-10-2018",   5009L,
       3L, "01-11-2018",   2857L,
       3L, "01-12-2018",   2909L,
       3L, "01-01-2019",   1353L,
       3L, "01-02-2019",   2337L,
       3L, "01-03-2019",   3019L,
       3L, "01-04-2019",   -531L,
       3L, "01-05-2019",  -1055L,
       3L, "01-06-2019",   1706L,
       3L, "01-07-2019",   -507L,
       3L, "01-08-2019",   2234L,
       3L, "01-09-2019",    890L,
       3L, "01-10-2019",     94L,
       3L, "01-11-2019",  -1781L,
       3L, "01-12-2019", 102590L,
       3L, "01-01-2020",    471L,
       4L, "01-12-2016",   2658L,
       4L, "01-01-2017",   2344L,
       4L, "01-02-2017",   2728L,
       4L, "01-03-2017",    -58L,
       4L, "01-04-2017",   -226L,
       4L, "01-05-2017",     -5L,
       4L, "01-06-2017",   -300L,
       4L, "01-07-2017",   -259L,
       4L, "01-08-2017",   -241L,
       4L, "01-09-2017",   -225L,
       4L, "01-10-2017",   -229L,
       4L, "01-11-2017",   -243L,
       4L, "01-12-2017",   -217L,
       4L, "01-01-2018",   -245L,
       4L, "01-02-2018",   -277L,
       4L, "01-03-2018",   -155L,
       4L, "01-04-2018",   5437L,
       4L, "01-05-2018",   2866L,
       4L, "01-06-2018",   3091L,
       4L, "01-07-2018",   3669L,
       4L, "01-08-2018",    311L,
       4L, "01-09-2018",   4120L,
       4L, "01-10-2018",   2357L,
       4L, "01-11-2018",  -4759L,
       4L, "01-12-2018",   4220L,
       4L, "01-01-2019",   2730L,
       4L, "01-02-2019",   2515L,
       4L, "01-03-2019",   2560L,
       4L, "01-04-2019",   2864L,
       4L, "01-05-2019",   1935L,
       4L, "01-06-2019",    938L,
       4L, "01-07-2019",   3268L,
       4L, "01-08-2019",   3232L,
       4L, "01-09-2019",   3347L,
       4L, "01-10-2019",   4241L,
       4L, "01-11-2019",   -247L,
       4L, "01-12-2019",    179L,
       4L, "01-01-2020",   2542L
      )

x<-as.data.frame(x)

I tried with dplyr: 

x %>% 
  group_by(ID) %>% 
  do(fit=auto.arima(.$Value,seasonal = F,stepwise = F,approximation = F),
     fit_forecast=forecast(auto.arima(.$Value,seasonal = F,stepwise = F,approximation = F),h=12))

But I don't know how to add step of train and test and the metrics. Does anyone know how to solve it? Thanks !

Upvotes: 0

Views: 993

Answers (2)

Rob Hyndman
Rob Hyndman

Reputation: 31800

I've already answered the same question at https://community.rstudio.com/t/autoarima-train-and-test-grouped-by-id-in-r/66400, but for the record here it is again.

This is much easier to do with the tsibble and fable packages, like this.

library(dplyr)
library(tsibble)
library(lubridate)
library(fable)

# Turn x into a tsibble object
x <- x %>%
  mutate(Date = yearmonth(dmy(Date))) %>%
  as_tsibble(index = Date, key = ID)

# Use filter to create training set
# Then fit non-seasonal ARIMA models
fit <- x %>%
  filter(Date <= yearmonth("2018 Dec")) %>%
  model(ARIMA(Value ~ PDQ(0,0,0), stepwise=FALSE, approximation=FALSE))

# Now forecast the test set and compute RMSE and MSE
fit %>%
  forecast(h = 13) %>%
  accuracy(x) %>%
  mutate(MSE = RMSE^2) %>%
  select(ID, RMSE, MSE)
#> # A tibble: 4 x 3
#>      ID   RMSE        MSE
#>   <int>  <dbl>      <dbl>
#> 1     1  9696.  94011195.
#> 2     2  4792.  22964718.
#> 3     3 27899. 778326636.
#> 4     4  1776.   3153398.

Created on 2020-05-18 by the reprex package (v0.3.0)

Upvotes: 2

Shan R
Shan R

Reputation: 541

You simply subset your dataframe into training and test. Typically 80/20 split or 90/10 split. But in your case, since you seem to want specific dates, you can even hardcode the number of rows.

library("lubridate")
x$Date <- dmy(x$Date)
x <- x[order(x$Date),]
rownames(x) <- NULL
x.train <- x[1:139,]
x.text <- x[-(1:139),]

Typically the data is also sampled to get the 80/20 or 90/10. But in a timeseries, you want to want to simply split instead of sampling.

Upvotes: 0

Related Questions