healthyverse_tsa
Time Series Analysis, Modeling and Forecasting of the Healthyverse Packages
Steven P. Sanderson II, MPH - Date: 2026-07-23
Introduction
This analysis follows a Nested Modeltime Workflow from modeltime
along with using the NNS package. I use this to monitor the
downloads of all of my packages:
Get Data
glimpse(downloads_tbl)
Rows: 184,539
Columns: 11
$ date <date> 2020-11-23, 2020-11-23, 2020-11-23, 2020-11-23, 2020-11-23,…
$ time <Period> 15H 36M 55S, 11H 26M 39S, 23H 34M 44S, 18H 39M 32S, 9H 0M…
$ date_time <dttm> 2020-11-23 15:36:55, 2020-11-23 11:26:39, 2020-11-23 23:34:…
$ size <int> 4858294, 4858294, 4858301, 4858295, 361, 4863722, 4864794, 4…
$ r_version <chr> NA, "4.0.3", "3.5.3", "3.5.2", NA, NA, NA, NA, NA, NA, NA, N…
$ r_arch <chr> NA, "x86_64", "x86_64", "x86_64", NA, NA, NA, NA, NA, NA, NA…
$ r_os <chr> NA, "mingw32", "mingw32", "linux-gnu", NA, NA, NA, NA, NA, N…
$ package <chr> "healthyR.data", "healthyR.data", "healthyR.data", "healthyR…
$ version <chr> "1.0.0", "1.0.0", "1.0.0", "1.0.0", "1.0.0", "1.0.0", "1.0.0…
$ country <chr> "US", "US", "US", "GB", "US", "US", "DE", "HK", "JP", "US", …
$ ip_id <int> 2069, 2804, 78827, 27595, 90474, 90474, 42435, 74, 7655, 638…
The last day in the data set is 2026-07-21 23:50:31, the file was birthed on: 2025-10-31 10:47:59.603742, and at report knit time is 6321.04 hours old. Happy analyzing!
Now that we have our data lets take a look at it using the skimr
package.
skim(downloads_tbl)
| Name | downloads_tbl |
| Number of rows | 184539 |
| Number of columns | 11 |
| _______________________ | |
| Column type frequency: | |
| character | 6 |
| Date | 1 |
| numeric | 2 |
| POSIXct | 1 |
| Timespan | 1 |
| ________________________ | |
| Group variables | None |
Data summary
Variable type: character
| skim_variable | n_missing | complete_rate | min | max | empty | n_unique | whitespace |
|---|---|---|---|---|---|---|---|
| r_version | 138372 | 0.25 | 5 | 17 | 0 | 54 | 0 |
| r_arch | 138372 | 0.25 | 1 | 7 | 0 | 7 | 0 |
| r_os | 138372 | 0.25 | 7 | 33 | 0 | 33 | 0 |
| package | 0 | 1.00 | 7 | 13 | 0 | 8 | 0 |
| version | 0 | 1.00 | 5 | 17 | 0 | 63 | 0 |
| country | 18015 | 0.90 | 2 | 2 | 0 | 170 | 0 |
Variable type: Date
| skim_variable | n_missing | complete_rate | min | max | median | n_unique |
|---|---|---|---|---|---|---|
| date | 0 | 1 | 2020-11-23 | 2026-07-21 | 2024-02-14 | 2060 |
Variable type: numeric
| skim_variable | n_missing | complete_rate | mean | sd | p0 | p25 | p50 | p75 | p100 | hist |
|---|---|---|---|---|---|---|---|---|---|---|
| size | 0 | 1 | 1133217.14 | 1473785.69 | 355 | 43661 | 325606 | 2334859 | 5677952 | ▇▁▂▁▁ |
| ip_id | 0 | 1 | 12106.51 | 25530.05 | 1 | 157 | 2697 | 11959 | 429286 | ▇▁▁▁▁ |
Variable type: POSIXct
| skim_variable | n_missing | complete_rate | min | max | median | n_unique |
|---|---|---|---|---|---|---|
| date_time | 0 | 1 | 2020-11-23 09:00:41 | 2026-07-21 23:50:31 | 2024-02-14 09:38:09 | 118216 |
Variable type: Timespan
| skim_variable | n_missing | complete_rate | min | max | median | n_unique |
|---|---|---|---|---|---|---|
| time | 0 | 1 | 0 | 59 | 12H 12M 23S | 60 |
We can see that the following columns are missing a lot of data and for
us are most likely not useful anyways, so we will drop them
c(r_version, r_arch, r_os)
Plots
Now lets take a look at a time-series plot of the total daily downloads by package. We will use a log scale and place a vertical line at each version release for each package.
[[1]]
[[2]]
[[3]]
[[4]]
[[5]]
[[6]]
[[7]]
[[8]]
Now lets take a look at some time series decomposition graphs.
[[1]]
[[2]]
[[3]]
[[4]]
[[5]]
[[6]]
[[7]]
[[8]]
[[1]]
[[2]]
[[3]]
[[4]]
[[5]]
[[6]]
[[7]]
[[8]]
Seasonal Diagnostics:
[[1]]
[[2]]
[[3]]
[[4]]
[[5]]
[[6]]
[[7]]
[[8]]
ACF and PACF Diagnostics:
[[1]]
[[2]]
[[3]]
[[4]]
[[5]]
[[6]]
[[7]]
[[8]]
Feature Engineering
Now that we have our basic data and a shot of what it looks like, let’s
add some features to our data which can be very helpful in modeling.
Lets start by making a tibble that is aggregated by the day and
package, as we are going to be interested in forecasting the next 4
weeks or 28 days for each package. First lets get our base data.
Call:
stats::lm(formula = .formula, data = df)
Residuals:
Min 1Q Median 3Q Max
-152.98 -38.41 -12.07 28.39 827.42
Coefficients:
Estimate Std. Error
(Intercept) -1.423e+02 4.978e+01
date 9.228e-03 2.631e-03
lag(value, 1) 9.301e-02 2.201e-02
lag(value, 7) 6.719e-02 2.260e-02
lag(value, 14) 7.002e-02 2.246e-02
lag(value, 21) 8.681e-02 2.255e-02
lag(value, 28) 7.905e-02 2.247e-02
lag(value, 35) 3.689e-02 2.252e-02
lag(value, 42) 6.213e-02 2.266e-02
lag(value, 49) 8.113e-02 2.266e-02
month(date, label = TRUE).L -8.484e+00 4.752e+00
month(date, label = TRUE).Q 8.970e-01 4.670e+00
month(date, label = TRUE).C -1.559e+01 4.728e+00
month(date, label = TRUE)^4 -9.181e+00 4.732e+00
month(date, label = TRUE)^5 -4.505e+00 4.731e+00
month(date, label = TRUE)^6 -7.166e-01 4.737e+00
month(date, label = TRUE)^7 -3.324e+00 4.689e+00
month(date, label = TRUE)^8 -4.130e+00 4.664e+00
month(date, label = TRUE)^9 1.298e+00 4.667e+00
month(date, label = TRUE)^10 7.561e-01 4.696e+00
month(date, label = TRUE)^11 -9.325e-01 4.598e+00
fourier_vec(date, type = "sin", K = 1, period = 7) -1.074e+01 2.098e+00
fourier_vec(date, type = "cos", K = 1, period = 7) 7.684e+00 2.158e+00
t value Pr(>|t|)
(Intercept) -2.859 0.004293 **
date 3.508 0.000462 ***
lag(value, 1) 4.225 2.49e-05 ***
lag(value, 7) 2.974 0.002977 **
lag(value, 14) 3.118 0.001849 **
lag(value, 21) 3.850 0.000122 ***
lag(value, 28) 3.518 0.000445 ***
lag(value, 35) 1.638 0.101657
lag(value, 42) 2.741 0.006175 **
lag(value, 49) 3.581 0.000351 ***
month(date, label = TRUE).L -1.785 0.074372 .
month(date, label = TRUE).Q 0.192 0.847681
month(date, label = TRUE).C -3.297 0.000993 ***
month(date, label = TRUE)^4 -1.940 0.052515 .
month(date, label = TRUE)^5 -0.952 0.341109
month(date, label = TRUE)^6 -0.151 0.879777
month(date, label = TRUE)^7 -0.709 0.478506
month(date, label = TRUE)^8 -0.886 0.375951
month(date, label = TRUE)^9 0.278 0.780885
month(date, label = TRUE)^10 0.161 0.872106
month(date, label = TRUE)^11 -0.203 0.839316
fourier_vec(date, type = "sin", K = 1, period = 7) -5.120 3.36e-07 ***
fourier_vec(date, type = "cos", K = 1, period = 7) 3.561 0.000378 ***
---
Signif. codes: 0 '***' 0.001 '**' 0.01 '*' 0.05 '.' 0.1 ' ' 1
Residual standard error: 60.27 on 1988 degrees of freedom
(49 observations deleted due to missingness)
Multiple R-squared: 0.1997, Adjusted R-squared: 0.1908
F-statistic: 22.55 on 22 and 1988 DF, p-value: < 2.2e-16
NNS Forecasting
This is something I have been wanting to try for a while. The NNS
package is a great package for forecasting time series data.
library(NNS)
data_list <- base_data |>
select(package, value) |>
group_split(package)
data_list |>
imap(
\(x, idx) {
obj <- x
x <- obj |> pull(value) |> tail(7*52)
train_set_size <- length(x) - 56
pkg <- obj |> pluck(1) |> unique()
# sf <- NNS.seas(x, modulo = 7, plot = FALSE)$periods
seas <- t(
sapply(
1:25,
function(i) c(
i,
sqrt(
mean((
NNS.ARMA(x,
h = 28,
training.set = train_set_size,
method = "lin",
seasonal.factor = i,
plot=FALSE
) - tail(x, 28)) ^ 2)))
)
)
colnames(seas) <- c("Period", "RMSE")
sf <- seas[which.min(seas[, 2]), 1]
cat(paste0("Package: ", pkg, "\n"))
NNS.ARMA.optim(
variable = x,
h = 28,
training.set = train_set_size,
#seasonal.factor = seq(12, 60, 7),
seasonal.factor = sf,
pred.int = 0.95,
plot = TRUE
)
title(
sub = paste0("\n",
"Package: ", pkg, " - NNS Optimization")
)
}
)
Package: healthyR
[1] "CURRNET METHOD: lin"
[1] "COPY LATEST PARAMETERS DIRECTLY FOR NNS.ARMA() IF ERROR:"
[1] "NNS.ARMA(... method = 'lin' , seasonal.factor = c( 21 ) ...)"
[1] "CURRENT lin OBJECTIVE FUNCTION = 6.36814853440202"
[1] "BEST method = 'lin' PATH MEMBER = c( 21 )"
[1] "BEST lin OBJECTIVE FUNCTION = 6.36814853440202"
[1] "CURRNET METHOD: nonlin"
[1] "COPY LATEST PARAMETERS DIRECTLY FOR NNS.ARMA() IF ERROR:"
[1] "NNS.ARMA(... method = 'nonlin' , seasonal.factor = c( 21 ) ...)"
[1] "CURRENT nonlin OBJECTIVE FUNCTION = 11.9735120024927"
[1] "BEST method = 'nonlin' PATH MEMBER = c( 21 )"
[1] "BEST nonlin OBJECTIVE FUNCTION = 11.9735120024927"
[1] "CURRNET METHOD: both"
[1] "COPY LATEST PARAMETERS DIRECTLY FOR NNS.ARMA() IF ERROR:"
[1] "NNS.ARMA(... method = 'both' , seasonal.factor = c( 21 ) ...)"
[1] "CURRENT both OBJECTIVE FUNCTION = 8.65489368389751"
[1] "BEST method = 'both' PATH MEMBER = c( 21 )"
[1] "BEST both OBJECTIVE FUNCTION = 8.65489368389751"
Package: healthyR.ai
[1] "CURRNET METHOD: lin"
[1] "COPY LATEST PARAMETERS DIRECTLY FOR NNS.ARMA() IF ERROR:"
[1] "NNS.ARMA(... method = 'lin' , seasonal.factor = c( 14 ) ...)"
[1] "CURRENT lin OBJECTIVE FUNCTION = 4.53953217528045"
[1] "BEST method = 'lin' PATH MEMBER = c( 14 )"
[1] "BEST lin OBJECTIVE FUNCTION = 4.53953217528045"
[1] "CURRNET METHOD: nonlin"
[1] "COPY LATEST PARAMETERS DIRECTLY FOR NNS.ARMA() IF ERROR:"
[1] "NNS.ARMA(... method = 'nonlin' , seasonal.factor = c( 14 ) ...)"
[1] "CURRENT nonlin OBJECTIVE FUNCTION = 2.82679282361839"
[1] "BEST method = 'nonlin' PATH MEMBER = c( 14 )"
[1] "BEST nonlin OBJECTIVE FUNCTION = 2.82679282361839"
[1] "CURRNET METHOD: both"
[1] "COPY LATEST PARAMETERS DIRECTLY FOR NNS.ARMA() IF ERROR:"
[1] "NNS.ARMA(... method = 'both' , seasonal.factor = c( 14 ) ...)"
[1] "CURRENT both OBJECTIVE FUNCTION = 3.42813576857545"
[1] "BEST method = 'both' PATH MEMBER = c( 14 )"
[1] "BEST both OBJECTIVE FUNCTION = 3.42813576857545"
Package: healthyR.data
[1] "CURRNET METHOD: lin"
[1] "COPY LATEST PARAMETERS DIRECTLY FOR NNS.ARMA() IF ERROR:"
[1] "NNS.ARMA(... method = 'lin' , seasonal.factor = c( 22 ) ...)"
[1] "CURRENT lin OBJECTIVE FUNCTION = 7.83687671912085"
[1] "BEST method = 'lin' PATH MEMBER = c( 22 )"
[1] "BEST lin OBJECTIVE FUNCTION = 7.83687671912085"
[1] "CURRNET METHOD: nonlin"
[1] "COPY LATEST PARAMETERS DIRECTLY FOR NNS.ARMA() IF ERROR:"
[1] "NNS.ARMA(... method = 'nonlin' , seasonal.factor = c( 22 ) ...)"
[1] "CURRENT nonlin OBJECTIVE FUNCTION = 9.58780315230894"
[1] "BEST method = 'nonlin' PATH MEMBER = c( 22 )"
[1] "BEST nonlin OBJECTIVE FUNCTION = 9.58780315230894"
[1] "CURRNET METHOD: both"
[1] "COPY LATEST PARAMETERS DIRECTLY FOR NNS.ARMA() IF ERROR:"
[1] "NNS.ARMA(... method = 'both' , seasonal.factor = c( 22 ) ...)"
[1] "CURRENT both OBJECTIVE FUNCTION = 9.5056413963395"
[1] "BEST method = 'both' PATH MEMBER = c( 22 )"
[1] "BEST both OBJECTIVE FUNCTION = 9.5056413963395"
Package: healthyR.ts
[1] "CURRNET METHOD: lin"
[1] "COPY LATEST PARAMETERS DIRECTLY FOR NNS.ARMA() IF ERROR:"
[1] "NNS.ARMA(... method = 'lin' , seasonal.factor = c( 3 ) ...)"
[1] "CURRENT lin OBJECTIVE FUNCTION = 23.4437966358135"
[1] "BEST method = 'lin' PATH MEMBER = c( 3 )"
[1] "BEST lin OBJECTIVE FUNCTION = 23.4437966358135"
[1] "CURRNET METHOD: nonlin"
[1] "COPY LATEST PARAMETERS DIRECTLY FOR NNS.ARMA() IF ERROR:"
[1] "NNS.ARMA(... method = 'nonlin' , seasonal.factor = c( 3 ) ...)"
[1] "CURRENT nonlin OBJECTIVE FUNCTION = 9.20051414154505"
[1] "BEST method = 'nonlin' PATH MEMBER = c( 3 )"
[1] "BEST nonlin OBJECTIVE FUNCTION = 9.20051414154505"
[1] "CURRNET METHOD: both"
[1] "COPY LATEST PARAMETERS DIRECTLY FOR NNS.ARMA() IF ERROR:"
[1] "NNS.ARMA(... method = 'both' , seasonal.factor = c( 3 ) ...)"
[1] "CURRENT both OBJECTIVE FUNCTION = 11.2968847436722"
[1] "BEST method = 'both' PATH MEMBER = c( 3 )"
[1] "BEST both OBJECTIVE FUNCTION = 11.2968847436722"
Package: healthyverse
[1] "CURRNET METHOD: lin"
[1] "COPY LATEST PARAMETERS DIRECTLY FOR NNS.ARMA() IF ERROR:"
[1] "NNS.ARMA(... method = 'lin' , seasonal.factor = c( 10 ) ...)"
[1] "CURRENT lin OBJECTIVE FUNCTION = 7.82010991065232"
[1] "BEST method = 'lin' PATH MEMBER = c( 10 )"
[1] "BEST lin OBJECTIVE FUNCTION = 7.82010991065232"
[1] "CURRNET METHOD: nonlin"
[1] "COPY LATEST PARAMETERS DIRECTLY FOR NNS.ARMA() IF ERROR:"
[1] "NNS.ARMA(... method = 'nonlin' , seasonal.factor = c( 10 ) ...)"
[1] "CURRENT nonlin OBJECTIVE FUNCTION = 16.3257001717509"
[1] "BEST method = 'nonlin' PATH MEMBER = c( 10 )"
[1] "BEST nonlin OBJECTIVE FUNCTION = 16.3257001717509"
[1] "CURRNET METHOD: both"
[1] "COPY LATEST PARAMETERS DIRECTLY FOR NNS.ARMA() IF ERROR:"
[1] "NNS.ARMA(... method = 'both' , seasonal.factor = c( 10 ) ...)"
[1] "CURRENT both OBJECTIVE FUNCTION = 9.33336831632444"
[1] "BEST method = 'both' PATH MEMBER = c( 10 )"
[1] "BEST both OBJECTIVE FUNCTION = 9.33336831632444"
Package: RandomWalker
[1] "CURRNET METHOD: lin"
[1] "COPY LATEST PARAMETERS DIRECTLY FOR NNS.ARMA() IF ERROR:"
[1] "NNS.ARMA(... method = 'lin' , seasonal.factor = c( 13 ) ...)"
[1] "CURRENT lin OBJECTIVE FUNCTION = 8.75107132816099"
[1] "BEST method = 'lin' PATH MEMBER = c( 13 )"
[1] "BEST lin OBJECTIVE FUNCTION = 8.75107132816099"
[1] "CURRNET METHOD: nonlin"
[1] "COPY LATEST PARAMETERS DIRECTLY FOR NNS.ARMA() IF ERROR:"
[1] "NNS.ARMA(... method = 'nonlin' , seasonal.factor = c( 13 ) ...)"
[1] "CURRENT nonlin OBJECTIVE FUNCTION = 10.4396103750543"
[1] "BEST method = 'nonlin' PATH MEMBER = c( 13 )"
[1] "BEST nonlin OBJECTIVE FUNCTION = 10.4396103750543"
[1] "CURRNET METHOD: both"
[1] "COPY LATEST PARAMETERS DIRECTLY FOR NNS.ARMA() IF ERROR:"
[1] "NNS.ARMA(... method = 'both' , seasonal.factor = c( 13 ) ...)"
[1] "CURRENT both OBJECTIVE FUNCTION = 9.97611176818114"
[1] "BEST method = 'both' PATH MEMBER = c( 13 )"
[1] "BEST both OBJECTIVE FUNCTION = 9.97611176818114"
Package: tidyAML
[1] "CURRNET METHOD: lin"
[1] "COPY LATEST PARAMETERS DIRECTLY FOR NNS.ARMA() IF ERROR:"
[1] "NNS.ARMA(... method = 'lin' , seasonal.factor = c( 2 ) ...)"
[1] "CURRENT lin OBJECTIVE FUNCTION = 102.113892975685"
[1] "BEST method = 'lin' PATH MEMBER = c( 2 )"
[1] "BEST lin OBJECTIVE FUNCTION = 102.113892975685"
[1] "CURRNET METHOD: nonlin"
[1] "COPY LATEST PARAMETERS DIRECTLY FOR NNS.ARMA() IF ERROR:"
[1] "NNS.ARMA(... method = 'nonlin' , seasonal.factor = c( 2 ) ...)"
[1] "CURRENT nonlin OBJECTIVE FUNCTION = 74.1293506241215"
[1] "BEST method = 'nonlin' PATH MEMBER = c( 2 )"
[1] "BEST nonlin OBJECTIVE FUNCTION = 74.1293506241215"
[1] "CURRNET METHOD: both"
[1] "COPY LATEST PARAMETERS DIRECTLY FOR NNS.ARMA() IF ERROR:"
[1] "NNS.ARMA(... method = 'both' , seasonal.factor = c( 2 ) ...)"
[1] "CURRENT both OBJECTIVE FUNCTION = 83.1111453645318"
[1] "BEST method = 'both' PATH MEMBER = c( 2 )"
[1] "BEST both OBJECTIVE FUNCTION = 83.1111453645318"
Package: TidyDensity
[1] "CURRNET METHOD: lin"
[1] "COPY LATEST PARAMETERS DIRECTLY FOR NNS.ARMA() IF ERROR:"
[1] "NNS.ARMA(... method = 'lin' , seasonal.factor = c( 17 ) ...)"
[1] "CURRENT lin OBJECTIVE FUNCTION = 9.28857389989922"
[1] "BEST method = 'lin' PATH MEMBER = c( 17 )"
[1] "BEST lin OBJECTIVE FUNCTION = 9.28857389989922"
[1] "CURRNET METHOD: nonlin"
[1] "COPY LATEST PARAMETERS DIRECTLY FOR NNS.ARMA() IF ERROR:"
[1] "NNS.ARMA(... method = 'nonlin' , seasonal.factor = c( 17 ) ...)"
[1] "CURRENT nonlin OBJECTIVE FUNCTION = 11.0412780654632"
[1] "BEST method = 'nonlin' PATH MEMBER = c( 17 )"
[1] "BEST nonlin OBJECTIVE FUNCTION = 11.0412780654632"
[1] "CURRNET METHOD: both"
[1] "COPY LATEST PARAMETERS DIRECTLY FOR NNS.ARMA() IF ERROR:"
[1] "NNS.ARMA(... method = 'both' , seasonal.factor = c( 17 ) ...)"
[1] "CURRENT both OBJECTIVE FUNCTION = 10.3130578841034"
[1] "BEST method = 'both' PATH MEMBER = c( 17 )"
[1] "BEST both OBJECTIVE FUNCTION = 10.3130578841034"
[[1]]
NULL
[[2]]
NULL
[[3]]
NULL
[[4]]
NULL
[[5]]
NULL
[[6]]
NULL
[[7]]
NULL
[[8]]
NULL
Pre-Processing
Now we are going to do some basic pre-processing.
data_padded_tbl <- base_data %>%
pad_by_time(
.date_var = date,
.pad_value = 0
)
# Get log interval and standardization parameters
log_params <- liv(data_padded_tbl$value, limit_lower = 0, offset = 1, silent = TRUE)
limit_lower <- log_params$limit_lower
limit_upper <- log_params$limit_upper
offset <- log_params$offset
data_liv_tbl <- data_padded_tbl %>%
# Get log interval transform
mutate(value_trans = liv(value, limit_lower = 0, offset = 1, silent = TRUE)$log_scaled)
# Get Standardization Params
std_params <- standard_vec(data_liv_tbl$value_trans, silent = TRUE)
std_mean <- std_params$mean
std_sd <- std_params$sd
data_transformed_tbl <- data_liv_tbl %>%
group_by(package) %>%
# get standardization
mutate(value_trans = standard_vec(value_trans, silent = TRUE)$standard_scaled) %>%
tk_augment_fourier(
.date_var = date,
.periods = c(7, 14, 30, 90, 180),
.K = 2
) %>%
tk_augment_timeseries_signature(
.date_var = date
) %>%
ungroup() %>%
select(-c(value, -year.iso))
Since this is panel data we can follow one of two different modeling strategies. We can search for a global model in the panel data or we can use nested forecasting finding the best model for each of the time series. Since we only have 5 panels, we will use nested forecasting.
To do this we will use the nest_timeseries and
split_nested_timeseries functions to create a nested tibble.
horizon <- 4*7
nested_data_tbl <- data_transformed_tbl %>%
# 0. Filter out column where package is NA
filter(!is.na(package)) %>%
# 1. Extending: We'll predict n days into the future.
extend_timeseries(
.id_var = package,
.date_var = date,
.length_future = horizon
) %>%
# 2. Nesting: We'll group by id, and create a future dataset
# that forecasts n days of extended data and
# an actual dataset that contains n*2 days
nest_timeseries(
.id_var = package,
.length_future = horizon
#.length_actual = horizon*2
) %>%
# 3. Splitting: We'll take the actual data and create splits
# for accuracy and confidence interval estimation of n das (test)
# and the rest is training data
split_nested_timeseries(
.length_test = horizon
)
nested_data_tbl
# A tibble: 8 × 4
package .actual_data .future_data .splits
<fct> <list> <list> <list>
1 healthyR.data <tibble [2,048 × 50]> <tibble [28 × 50]> <split [2020|28]>
2 healthyR <tibble [2,042 × 50]> <tibble [28 × 50]> <split [2014|28]>
3 healthyR.ts <tibble [1,978 × 50]> <tibble [28 × 50]> <split [1950|28]>
4 healthyverse <tibble [1,878 × 50]> <tibble [28 × 50]> <split [1850|28]>
5 healthyR.ai <tibble [1,783 × 50]> <tibble [28 × 50]> <split [1755|28]>
6 TidyDensity <tibble [1,636 × 50]> <tibble [28 × 50]> <split [1608|28]>
7 tidyAML <tibble [1,240 × 50]> <tibble [28 × 50]> <split [1212|28]>
8 RandomWalker <tibble [664 × 50]> <tibble [28 × 50]> <split [636|28]>
Now it is time to make some recipes and models using the modeltime workflow.
Modeltime Workflow
Recipe Object
recipe_base <- recipe(
value_trans ~ .
, data = extract_nested_test_split(nested_data_tbl)
)
recipe_base
recipe_date <- recipe(
value_trans ~ date
, data = extract_nested_test_split(nested_data_tbl)
)
Models
# Models ------------------------------------------------------------------
# Auto ARIMA --------------------------------------------------------------
model_spec_arima_no_boost <- arima_reg() %>%
set_engine(engine = "auto_arima")
wflw_auto_arima <- workflow() %>%
add_recipe(recipe = recipe_date) %>%
add_model(model_spec_arima_no_boost)
# NNETAR ------------------------------------------------------------------
model_spec_nnetar <- nnetar_reg(
mode = "regression"
, seasonal_period = "auto"
) %>%
set_engine("nnetar")
wflw_nnetar <- workflow() %>%
add_recipe(recipe = recipe_base) %>%
add_model(model_spec_nnetar)
# TSLM --------------------------------------------------------------------
model_spec_lm <- linear_reg() %>%
set_engine("lm")
wflw_lm <- workflow() %>%
add_recipe(recipe = recipe_base) %>%
add_model(model_spec_lm)
# MARS --------------------------------------------------------------------
model_spec_mars <- mars(mode = "regression") %>%
set_engine("earth")
wflw_mars <- workflow() %>%
add_recipe(recipe = recipe_date) %>%
add_model(model_spec_mars)
Nested Modeltime Tables
nested_modeltime_tbl <- modeltime_nested_fit(
# Nested Data
nested_data = nested_data_tbl,
control = control_nested_fit(
verbose = TRUE,
allow_par = FALSE
),
# Add workflows
wflw_auto_arima,
wflw_lm,
wflw_mars,
wflw_nnetar
)
nested_modeltime_tbl <- nested_modeltime_tbl[!is.na(nested_modeltime_tbl$package),]
Model Accuracy
nested_modeltime_tbl %>%
extract_nested_test_accuracy() %>%
filter(!is.na(package)) %>%
knitr::kable()
| package | .model_id | .model_desc | .type | mae | mape | mase | smape | rmse | rsq |
|---|---|---|---|---|---|---|---|---|---|
| healthyR.data | 1 | ARIMA | Test | 0.7888585 | 89.06619 | 0.9314941 | 163.40479 | 0.9531812 | 0.0221274 |
| healthyR.data | 2 | LM | Test | 0.7618324 | 128.73246 | 0.8995815 | 143.35205 | 0.8873318 | 0.0123283 |
| healthyR.data | 3 | EARTH | Test | 0.7839696 | 107.87955 | 0.9257213 | 171.31551 | 0.9121673 | 0.1834255 |
| healthyR.data | 4 | NNAR | Test | 0.6971001 | 108.73356 | 0.8231447 | 134.90007 | 0.8403623 | 0.1132506 |
| healthyR | 1 | ARIMA | Test | 0.7396618 | 201.17751 | 1.0296096 | 121.76302 | 0.9299677 | 0.1261359 |
| healthyR | 2 | LM | Test | 0.8751429 | 178.92508 | 1.2181994 | 159.40285 | 1.0412698 | 0.0194374 |
| healthyR | 3 | EARTH | Test | 5.0290070 | 3851.65242 | 7.0003804 | 161.63656 | 5.4649724 | 0.1142701 |
| healthyR | 4 | NNAR | Test | 0.8053616 | 150.92858 | 1.1210638 | 144.54163 | 0.9653820 | 0.1647437 |
| healthyR.ts | 1 | ARIMA | Test | 0.8920327 | 175.61497 | 1.1123468 | 182.50379 | 1.1138183 | 0.0022580 |
| healthyR.ts | 2 | LM | Test | 0.9046183 | 496.22772 | 1.1280408 | 160.41763 | 1.1647390 | 0.0055431 |
| healthyR.ts | 3 | EARTH | Test | 1.5113605 | 1057.26426 | 1.8846360 | 179.37789 | 1.7348687 | 0.0827803 |
| healthyR.ts | 4 | NNAR | Test | 0.9021510 | 551.64662 | 1.1249640 | 167.12731 | 1.1385314 | 0.0045473 |
| healthyverse | 1 | ARIMA | Test | 0.5492663 | 87.49420 | 0.8167106 | 45.44368 | 0.6704319 | 0.0613849 |
| healthyverse | 2 | LM | Test | 0.8928897 | 66.75835 | 1.3276485 | 85.46715 | 1.0266848 | 0.0127254 |
| healthyverse | 3 | EARTH | Test | 0.5356991 | 100.99786 | 0.7965375 | 43.02700 | 0.7363193 | 0.0870788 |
| healthyverse | 4 | NNAR | Test | 0.9587060 | 68.99313 | 1.4255116 | 90.84927 | 1.1023779 | 0.0000187 |
| healthyR.ai | 1 | ARIMA | Test | 0.6976092 | 82.52778 | 1.0273334 | 128.67748 | 0.9019774 | 0.0006089 |
| healthyR.ai | 2 | LM | Test | 0.7112766 | 104.14830 | 1.0474605 | 116.19825 | 0.9100444 | 0.0224145 |
| healthyR.ai | 3 | EARTH | Test | 0.8635459 | 120.54661 | 1.2716997 | 193.35800 | 1.0245289 | 0.0396521 |
| healthyR.ai | 4 | NNAR | Test | 0.7572702 | 112.96090 | 1.1151930 | 146.28154 | 0.9788257 | 0.0144130 |
| TidyDensity | 1 | ARIMA | Test | 1.0009376 | 195.92429 | 1.1844462 | 163.33391 | 1.1229610 | 0.0133943 |
| TidyDensity | 2 | LM | Test | 1.0824009 | 250.77526 | 1.2808447 | 171.11225 | 1.1950585 | 0.0070672 |
| TidyDensity | 3 | EARTH | Test | 1.0827829 | 226.82132 | 1.2812967 | 168.64921 | 1.1880744 | 0.1090895 |
| TidyDensity | 4 | NNAR | Test | 1.1279274 | 285.65708 | 1.3347179 | 170.48139 | 1.2303575 | 0.0011483 |
| tidyAML | 1 | ARIMA | Test | 0.6717428 | 122.90178 | 0.9348115 | 175.90026 | 0.8908308 | 0.0024108 |
| tidyAML | 2 | LM | Test | 0.5453803 | 217.69115 | 0.7589628 | 123.38510 | 0.7168719 | 0.3114899 |
| tidyAML | 3 | EARTH | Test | 0.8873507 | 401.57762 | 1.2348560 | 160.73085 | 1.0456612 | 0.0134403 |
| tidyAML | 4 | NNAR | Test | 0.6409848 | 259.86085 | 0.8920080 | 145.92108 | 0.8099703 | 0.1303754 |
| RandomWalker | 1 | ARIMA | Test | 0.8836147 | 128.15017 | 0.9311119 | 167.48151 | 1.0512404 | 0.0000259 |
| RandomWalker | 2 | LM | Test | 0.8986807 | 175.24802 | 0.9469878 | 182.08453 | 1.0282596 | 0.0034356 |
| RandomWalker | 3 | EARTH | Test | 0.8051439 | 99.87184 | 0.8484231 | 184.83193 | 0.9567502 | 0.1135821 |
| RandomWalker | 4 | NNAR | Test | 1.0044807 | 218.83176 | 1.0584749 | 180.38876 | 1.1272290 | 0.0020915 |
Plot Models
nested_modeltime_tbl %>%
extract_nested_test_forecast() %>%
group_by(package) %>%
filter_by_time(.date_var = .index, .start_date = max(.index) - 60) %>%
ungroup() %>%
plot_modeltime_forecast(
.interactive = FALSE,
.conf_interval_show = FALSE,
.facet_scales = "free"
) +
theme_minimal() +
facet_wrap(~ package, nrow = 3) +
theme(legend.position = "bottom")
Best Model
best_nested_modeltime_tbl <- nested_modeltime_tbl %>%
modeltime_nested_select_best(
metric = "rmse",
minimize = TRUE,
filter_test_forecasts = TRUE
)
best_nested_modeltime_tbl %>%
extract_nested_best_model_report()
# Nested Modeltime Table
# A tibble: 8 × 10
package .model_id .model_desc .type mae mape mase smape rmse rsq
<fct> <int> <chr> <chr> <dbl> <dbl> <dbl> <dbl> <dbl> <dbl>
1 healthyR.da… 4 NNAR Test 0.697 109. 0.823 135. 0.840 1.13e-1
2 healthyR 1 ARIMA Test 0.740 201. 1.03 122. 0.930 1.26e-1
3 healthyR.ts 1 ARIMA Test 0.892 176. 1.11 183. 1.11 2.26e-3
4 healthyverse 1 ARIMA Test 0.549 87.5 0.817 45.4 0.670 6.14e-2
5 healthyR.ai 1 ARIMA Test 0.698 82.5 1.03 129. 0.902 6.09e-4
6 TidyDensity 1 ARIMA Test 1.00 196. 1.18 163. 1.12 1.34e-2
7 tidyAML 2 LM Test 0.545 218. 0.759 123. 0.717 3.11e-1
8 RandomWalker 3 EARTH Test 0.805 99.9 0.848 185. 0.957 1.14e-1
best_nested_modeltime_tbl %>%
extract_nested_test_forecast() %>%
#filter(!is.na(.model_id)) %>%
group_by(package) %>%
filter_by_time(.date_var = .index, .start_date = max(.index) - 60) %>%
ungroup() %>%
plot_modeltime_forecast(
.interactive = FALSE,
.conf_interval_alpha = 0.2,
.facet_scales = "free"
) +
facet_wrap(~ package, nrow = 3) +
theme_minimal() +
theme(legend.position = "bottom")
Refitting and Future Forecast
Now that we have the best models, we can make our future forecasts.
nested_modeltime_refit_tbl <- best_nested_modeltime_tbl %>%
modeltime_nested_refit(
control = control_nested_refit(verbose = TRUE)
)
nested_modeltime_refit_tbl
# Nested Modeltime Table
# A tibble: 8 × 5
package .actual_data .future_data .splits .modeltime_tables
<fct> <list> <list> <list> <list>
1 healthyR.data <tibble> <tibble> <split [2020|28]> <mdl_tm_t [1 × 5]>
2 healthyR <tibble> <tibble> <split [2014|28]> <mdl_tm_t [1 × 5]>
3 healthyR.ts <tibble> <tibble> <split [1950|28]> <mdl_tm_t [1 × 5]>
4 healthyverse <tibble> <tibble> <split [1850|28]> <mdl_tm_t [1 × 5]>
5 healthyR.ai <tibble> <tibble> <split [1755|28]> <mdl_tm_t [1 × 5]>
6 TidyDensity <tibble> <tibble> <split [1608|28]> <mdl_tm_t [1 × 5]>
7 tidyAML <tibble> <tibble> <split [1212|28]> <mdl_tm_t [1 × 5]>
8 RandomWalker <tibble> <tibble> <split [636|28]> <mdl_tm_t [1 × 5]>
nested_modeltime_refit_tbl %>%
extract_nested_future_forecast() %>%
group_by(package) %>%
mutate(across(.value:.conf_hi, .fns = ~ standard_inv_vec(
x = .,
mean = std_mean,
sd = std_sd
)$standard_inverse_value)) %>%
mutate(across(.value:.conf_hi, .fns = ~ liiv(
x = .,
limit_lower = limit_lower,
limit_upper = limit_upper,
offset = offset
)$rescaled_v)) %>%
filter_by_time(.date_var = .index, .start_date = max(.index) - 60) %>%
ungroup() %>%
plot_modeltime_forecast(
.interactive = FALSE,
.conf_interval_alpha = 0.2,
.facet_scales = "free"
) +
facet_wrap(~ package, nrow = 3) +
theme_minimal() +
theme(legend.position = "bottom")
