healthyverse_tsa
Time Series Analysis and Nested Modeling of the Healthyverse Packages
Steven P. Sanderson II, MPH - Date: 18 April, 2025
This analysis follows a Nested Modeltime Workflow.
Get Data
glimpse(downloads_tbl)
## Rows: 137,359
## 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 2025-04-16 23:58:30, the file was birthed on: 2024-08-07 07:35:44, and at report knit time is -6060.38 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 | 137359 |
| 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 | 98865 | 0.28 | 5 | 5 | 0 | 46 | 0 |
| r_arch | 98865 | 0.28 | 3 | 7 | 0 | 5 | 0 |
| r_os | 98865 | 0.28 | 7 | 15 | 0 | 21 | 0 |
| package | 0 | 1.00 | 7 | 13 | 0 | 8 | 0 |
| version | 0 | 1.00 | 5 | 17 | 0 | 60 | 0 |
| country | 11571 | 0.92 | 2 | 2 | 0 | 163 | 0 |
Variable type: Date
| skim_variable | n_missing | complete_rate | min | max | median | n_unique |
|---|---|---|---|---|---|---|
| date | 0 | 1 | 2020-11-23 | 2025-04-16 | 2023-06-02 | 1606 |
Variable type: numeric
| skim_variable | n_missing | complete_rate | mean | sd | p0 | p25 | p50 | p75 | p100 | hist |
|---|---|---|---|---|---|---|---|---|---|---|
| size | 0 | 1 | 1135594.72 | 1522675.63 | 355 | 14701 | 274998 | 2367774 | 5677952 | ▇▁▂▁▁ |
| ip_id | 0 | 1 | 10375.55 | 18408.15 | 1 | 303 | 3064 | 11721 | 209747 | ▇▁▁▁▁ |
Variable type: POSIXct
| skim_variable | n_missing | complete_rate | min | max | median | n_unique |
|---|---|---|---|---|---|---|
| date_time | 0 | 1 | 2020-11-23 09:00:41 | 2025-04-16 23:58:30 | 2023-06-02 22:32:54 | 83632 |
Variable type: Timespan
| skim_variable | n_missing | complete_rate | min | max | median | n_unique |
|---|---|---|---|---|---|---|
| time | 0 | 1 | 0 | 59 | 12H 7M 29S | 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.
Now lets take a look at some time series decomposition graphs.
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
## -147.98 -35.49 -10.57 26.70 813.18
##
## Coefficients:
## Estimate Std. Error
## (Intercept) -1.895e+02 7.263e+01
## date 1.149e-02 3.853e-03
## lag(value, 1) 1.063e-01 2.473e-02
## lag(value, 7) 9.931e-02 2.568e-02
## lag(value, 14) 9.637e-02 2.571e-02
## lag(value, 21) 6.487e-02 2.586e-02
## lag(value, 28) 6.390e-02 2.575e-02
## lag(value, 35) 6.948e-02 2.581e-02
## lag(value, 42) 4.951e-02 2.591e-02
## lag(value, 49) 7.016e-02 2.575e-02
## month(date, label = TRUE).L -1.041e+01 5.140e+00
## month(date, label = TRUE).Q 2.583e+00 5.201e+00
## month(date, label = TRUE).C -1.209e+01 5.214e+00
## month(date, label = TRUE)^4 -6.676e+00 5.199e+00
## month(date, label = TRUE)^5 -1.245e+01 5.206e+00
## month(date, label = TRUE)^6 -2.807e+00 5.267e+00
## month(date, label = TRUE)^7 -6.292e+00 5.169e+00
## month(date, label = TRUE)^8 -4.673e+00 5.176e+00
## month(date, label = TRUE)^9 5.750e+00 5.194e+00
## month(date, label = TRUE)^10 4.235e+00 5.271e+00
## month(date, label = TRUE)^11 -5.741e+00 5.340e+00
## fourier_vec(date, type = "sin", K = 1, period = 7) -1.160e+01 2.384e+00
## fourier_vec(date, type = "cos", K = 1, period = 7) 7.951e+00 2.512e+00
## t value Pr(>|t|)
## (Intercept) -2.609 0.009159 **
## date 2.982 0.002912 **
## lag(value, 1) 4.298 1.83e-05 ***
## lag(value, 7) 3.868 0.000114 ***
## lag(value, 14) 3.748 0.000184 ***
## lag(value, 21) 2.508 0.012230 *
## lag(value, 28) 2.482 0.013165 *
## lag(value, 35) 2.692 0.007183 **
## lag(value, 42) 1.911 0.056199 .
## lag(value, 49) 2.725 0.006510 **
## month(date, label = TRUE).L -2.025 0.043013 *
## month(date, label = TRUE).Q 0.497 0.619534
## month(date, label = TRUE).C -2.318 0.020586 *
## month(date, label = TRUE)^4 -1.284 0.199339
## month(date, label = TRUE)^5 -2.391 0.016916 *
## month(date, label = TRUE)^6 -0.533 0.594128
## month(date, label = TRUE)^7 -1.217 0.223750
## month(date, label = TRUE)^8 -0.903 0.366772
## month(date, label = TRUE)^9 1.107 0.268387
## month(date, label = TRUE)^10 0.804 0.421789
## month(date, label = TRUE)^11 -1.075 0.282498
## fourier_vec(date, type = "sin", K = 1, period = 7) -4.865 1.26e-06 ***
## fourier_vec(date, type = "cos", K = 1, period = 7) 3.165 0.001583 **
## ---
## Signif. codes: 0 '***' 0.001 '**' 0.01 '*' 0.05 '.' 0.1 ' ' 1
##
## Residual standard error: 58.78 on 1534 degrees of freedom
## (49 observations deleted due to missingness)
## Multiple R-squared: 0.2479, Adjusted R-squared: 0.2371
## F-statistic: 22.98 on 22 and 1534 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
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( 63 ) ...)"
## [1] "CURRENT lin OBJECTIVE FUNCTION = 2.96421994505375"
## [1] "NNS.ARMA(... method = 'lin' , seasonal.factor = c( 63, 91 ) ...)"
## [1] "CURRENT lin OBJECTIVE FUNCTION = 2.50730127755907"
## [1] "NNS.ARMA(... method = 'lin' , seasonal.factor = c( 63, 91, 70 ) ...)"
## [1] "CURRENT lin OBJECTIVE FUNCTION = 2.44827147523047"
## [1] "BEST method = 'lin', seasonal.factor = c( 63, 91, 70 )"
## [1] "BEST lin OBJECTIVE FUNCTION = 2.44827147523047"
## [1] "CURRNET METHOD: nonlin"
## [1] "COPY LATEST PARAMETERS DIRECTLY FOR NNS.ARMA() IF ERROR:"
## [1] "NNS.ARMA(... method = 'nonlin' , seasonal.factor = c( 63, 91, 70 ) ...)"
## [1] "CURRENT nonlin OBJECTIVE FUNCTION = 4.22298939920344"
## [1] "BEST method = 'nonlin' PATH MEMBER = c( 63, 91, 70 )"
## [1] "BEST nonlin OBJECTIVE FUNCTION = 4.22298939920344"
## [1] "CURRNET METHOD: both"
## [1] "COPY LATEST PARAMETERS DIRECTLY FOR NNS.ARMA() IF ERROR:"
## [1] "NNS.ARMA(... method = 'both' , seasonal.factor = c( 63, 91, 70 ) ...)"
## [1] "CURRENT both OBJECTIVE FUNCTION = 3.14654436267833"
## [1] "BEST method = 'both' PATH MEMBER = c( 63, 91, 70 )"
## [1] "BEST both OBJECTIVE FUNCTION = 3.14654436267833"
## 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( 63 ) ...)"
## [1] "CURRENT lin OBJECTIVE FUNCTION = 2.44321524718395"
## [1] "NNS.ARMA(... method = 'lin' , seasonal.factor = c( 63, 98 ) ...)"
## [1] "CURRENT lin OBJECTIVE FUNCTION = 1.89689991368037"
## [1] "BEST method = 'lin', seasonal.factor = c( 63, 98 )"
## [1] "BEST lin OBJECTIVE FUNCTION = 1.89689991368037"
## [1] "CURRNET METHOD: nonlin"
## [1] "COPY LATEST PARAMETERS DIRECTLY FOR NNS.ARMA() IF ERROR:"
## [1] "NNS.ARMA(... method = 'nonlin' , seasonal.factor = c( 63, 98 ) ...)"
## [1] "CURRENT nonlin OBJECTIVE FUNCTION = 2.28585135198188"
## [1] "BEST method = 'nonlin' PATH MEMBER = c( 63, 98 )"
## [1] "BEST nonlin OBJECTIVE FUNCTION = 2.28585135198188"
## [1] "CURRNET METHOD: both"
## [1] "COPY LATEST PARAMETERS DIRECTLY FOR NNS.ARMA() IF ERROR:"
## [1] "NNS.ARMA(... method = 'both' , seasonal.factor = c( 63, 98 ) ...)"
## [1] "CURRENT both OBJECTIVE FUNCTION = 1.94739191680271"
## [1] "BEST method = 'both' PATH MEMBER = c( 63, 98 )"
## [1] "BEST both OBJECTIVE FUNCTION = 1.94739191680271"
## 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( 98 ) ...)"
## [1] "CURRENT lin OBJECTIVE FUNCTION = 2.10106552447094"
## [1] "BEST method = 'lin', seasonal.factor = c( 98 )"
## [1] "BEST lin OBJECTIVE FUNCTION = 2.10106552447094"
## [1] "CURRNET METHOD: nonlin"
## [1] "COPY LATEST PARAMETERS DIRECTLY FOR NNS.ARMA() IF ERROR:"
## [1] "NNS.ARMA(... method = 'nonlin' , seasonal.factor = c( 98 ) ...)"
## [1] "CURRENT nonlin OBJECTIVE FUNCTION = 2.46760759916857"
## [1] "BEST method = 'nonlin' PATH MEMBER = c( 98 )"
## [1] "BEST nonlin OBJECTIVE FUNCTION = 2.46760759916857"
## [1] "CURRNET METHOD: both"
## [1] "COPY LATEST PARAMETERS DIRECTLY FOR NNS.ARMA() IF ERROR:"
## [1] "NNS.ARMA(... method = 'both' , seasonal.factor = c( 98 ) ...)"
## [1] "CURRENT both OBJECTIVE FUNCTION = 2.14131028554055"
## [1] "BEST method = 'both' PATH MEMBER = c( 98 )"
## [1] "BEST both OBJECTIVE FUNCTION = 2.14131028554055"
## 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( 63 ) ...)"
## [1] "CURRENT lin OBJECTIVE FUNCTION = 2.40087402589534"
## [1] "NNS.ARMA(... method = 'lin' , seasonal.factor = c( 63, 98 ) ...)"
## [1] "CURRENT lin OBJECTIVE FUNCTION = 2.0101134594531"
## [1] "BEST method = 'lin', seasonal.factor = c( 63, 98 )"
## [1] "BEST lin OBJECTIVE FUNCTION = 2.0101134594531"
## [1] "CURRNET METHOD: nonlin"
## [1] "COPY LATEST PARAMETERS DIRECTLY FOR NNS.ARMA() IF ERROR:"
## [1] "NNS.ARMA(... method = 'nonlin' , seasonal.factor = c( 63, 98 ) ...)"
## [1] "CURRENT nonlin OBJECTIVE FUNCTION = 3.19364655255016"
## [1] "BEST method = 'nonlin' PATH MEMBER = c( 63, 98 )"
## [1] "BEST nonlin OBJECTIVE FUNCTION = 3.19364655255016"
## [1] "CURRNET METHOD: both"
## [1] "COPY LATEST PARAMETERS DIRECTLY FOR NNS.ARMA() IF ERROR:"
## [1] "NNS.ARMA(... method = 'both' , seasonal.factor = c( 63, 98 ) ...)"
## [1] "CURRENT both OBJECTIVE FUNCTION = 2.47653323791533"
## [1] "BEST method = 'both' PATH MEMBER = c( 63, 98 )"
## [1] "BEST both OBJECTIVE FUNCTION = 2.47653323791533"
## Package: healthyverse
## [1] "CURRNET METHOD: lin"
## [1] "COPY LATEST PARAMETERS DIRECTLY FOR NNS.ARMA() IF ERROR:"
## [1] "NNS.ARMA(... method = 'lin' , seasonal.factor = c( 91 ) ...)"
## [1] "CURRENT lin OBJECTIVE FUNCTION = 3.80463938202456"
## [1] "BEST method = 'lin', seasonal.factor = c( 91 )"
## [1] "BEST lin OBJECTIVE FUNCTION = 3.80463938202456"
## [1] "CURRNET METHOD: nonlin"
## [1] "COPY LATEST PARAMETERS DIRECTLY FOR NNS.ARMA() IF ERROR:"
## [1] "NNS.ARMA(... method = 'nonlin' , seasonal.factor = c( 91 ) ...)"
## [1] "CURRENT nonlin OBJECTIVE FUNCTION = 5.38626154815699"
## [1] "BEST method = 'nonlin' PATH MEMBER = c( 91 )"
## [1] "BEST nonlin OBJECTIVE FUNCTION = 5.38626154815699"
## [1] "CURRNET METHOD: both"
## [1] "COPY LATEST PARAMETERS DIRECTLY FOR NNS.ARMA() IF ERROR:"
## [1] "NNS.ARMA(... method = 'both' , seasonal.factor = c( 91 ) ...)"
## [1] "CURRENT both OBJECTIVE FUNCTION = 4.34040228738022"
## [1] "BEST method = 'both' PATH MEMBER = c( 91 )"
## [1] "BEST both OBJECTIVE FUNCTION = 4.34040228738022"
## Package: RandomWalker
## [1] "CURRNET METHOD: lin"
## [1] "COPY LATEST PARAMETERS DIRECTLY FOR NNS.ARMA() IF ERROR:"
## [1] "NNS.ARMA(... method = 'lin' , seasonal.factor = c( 42 ) ...)"
## [1] "CURRENT lin OBJECTIVE FUNCTION = 1.8832062859861"
## [1] "BEST method = 'lin', seasonal.factor = c( 42 )"
## [1] "BEST lin OBJECTIVE FUNCTION = 1.8832062859861"
## [1] "CURRNET METHOD: nonlin"
## [1] "COPY LATEST PARAMETERS DIRECTLY FOR NNS.ARMA() IF ERROR:"
## [1] "NNS.ARMA(... method = 'nonlin' , seasonal.factor = c( 42 ) ...)"
## [1] "CURRENT nonlin OBJECTIVE FUNCTION = 2.23924128184997"
## [1] "BEST method = 'nonlin' PATH MEMBER = c( 42 )"
## [1] "BEST nonlin OBJECTIVE FUNCTION = 2.23924128184997"
## [1] "CURRNET METHOD: both"
## [1] "COPY LATEST PARAMETERS DIRECTLY FOR NNS.ARMA() IF ERROR:"
## [1] "NNS.ARMA(... method = 'both' , seasonal.factor = c( 42 ) ...)"
## [1] "CURRENT both OBJECTIVE FUNCTION = 1.82202040181459"
## [1] "BEST method = 'both' PATH MEMBER = c( 42 )"
## [1] "BEST both OBJECTIVE FUNCTION = 1.82202040181459"
## Package: tidyAML
## [1] "CURRNET METHOD: lin"
## [1] "COPY LATEST PARAMETERS DIRECTLY FOR NNS.ARMA() IF ERROR:"
## [1] "NNS.ARMA(... method = 'lin' , seasonal.factor = c( 28 ) ...)"
## [1] "CURRENT lin OBJECTIVE FUNCTION = 4.17654966954133"
## [1] "BEST method = 'lin', seasonal.factor = c( 28 )"
## [1] "BEST lin OBJECTIVE FUNCTION = 4.17654966954133"
## [1] "CURRNET METHOD: nonlin"
## [1] "COPY LATEST PARAMETERS DIRECTLY FOR NNS.ARMA() IF ERROR:"
## [1] "NNS.ARMA(... method = 'nonlin' , seasonal.factor = c( 28 ) ...)"
## [1] "CURRENT nonlin OBJECTIVE FUNCTION = 2.7576357499475"
## [1] "BEST method = 'nonlin' PATH MEMBER = c( 28 )"
## [1] "BEST nonlin OBJECTIVE FUNCTION = 2.7576357499475"
## [1] "CURRNET METHOD: both"
## [1] "COPY LATEST PARAMETERS DIRECTLY FOR NNS.ARMA() IF ERROR:"
## [1] "NNS.ARMA(... method = 'both' , seasonal.factor = c( 28 ) ...)"
## [1] "CURRENT both OBJECTIVE FUNCTION = 2.6564814874381"
## [1] "BEST method = 'both' PATH MEMBER = c( 28 )"
## [1] "BEST both OBJECTIVE FUNCTION = 2.6564814874381"
## Package: TidyDensity
## [1] "CURRNET METHOD: lin"
## [1] "COPY LATEST PARAMETERS DIRECTLY FOR NNS.ARMA() IF ERROR:"
## [1] "NNS.ARMA(... method = 'lin' , seasonal.factor = c( 63 ) ...)"
## [1] "CURRENT lin OBJECTIVE FUNCTION = 2.91839912048994"
## [1] "NNS.ARMA(... method = 'lin' , seasonal.factor = c( 63, 77 ) ...)"
## [1] "CURRENT lin OBJECTIVE FUNCTION = 2.89641011559762"
## [1] "BEST method = 'lin', seasonal.factor = c( 63, 77 )"
## [1] "BEST lin OBJECTIVE FUNCTION = 2.89641011559762"
## [1] "CURRNET METHOD: nonlin"
## [1] "COPY LATEST PARAMETERS DIRECTLY FOR NNS.ARMA() IF ERROR:"
## [1] "NNS.ARMA(... method = 'nonlin' , seasonal.factor = c( 63, 77 ) ...)"
## [1] "CURRENT nonlin OBJECTIVE FUNCTION = 3.72253043687994"
## [1] "BEST method = 'nonlin' PATH MEMBER = c( 63, 77 )"
## [1] "BEST nonlin OBJECTIVE FUNCTION = 3.72253043687994"
## [1] "CURRNET METHOD: both"
## [1] "COPY LATEST PARAMETERS DIRECTLY FOR NNS.ARMA() IF ERROR:"
## [1] "NNS.ARMA(... method = 'both' , seasonal.factor = c( 63, 77 ) ...)"
## [1] "CURRENT both OBJECTIVE FUNCTION = 3.13079196938131"
## [1] "BEST method = 'both' PATH MEMBER = c( 63, 77 )"
## [1] "BEST both OBJECTIVE FUNCTION = 3.13079196938131"
## [[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$sd
data_transformed_tbl <- data_liv_tbl %>%
# get standardization
mutate(value_trans = standard_vec(value_trans, silent = TRUE)$standard_scaled) %>%
select(-value)
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 %>%
# 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 [1,599 × 2]> <tibble [28 × 2]> <split [1571|28]>
## 2 healthyR <tibble [1,592 × 2]> <tibble [28 × 2]> <split [1564|28]>
## 3 healthyR.ts <tibble [1,536 × 2]> <tibble [28 × 2]> <split [1508|28]>
## 4 healthyverse <tibble [1,507 × 2]> <tibble [28 × 2]> <split [1479|28]>
## 5 healthyR.ai <tibble [1,331 × 2]> <tibble [28 × 2]> <split [1303|28]>
## 6 TidyDensity <tibble [1,182 × 2]> <tibble [28 × 2]> <split [1154|28]>
## 7 tidyAML <tibble [790 × 2]> <tibble [28 × 2]> <split [762|28]>
## 8 RandomWalker <tibble [212 × 2]> <tibble [28 × 2]> <split [184|28]>
Now it is time to make some recipes and models using the modeltime workflow.
Modeltime Workflow
Recipe Object
recipe_base <- recipe(
value_trans ~ date
, data = extract_nested_test_split(nested_data_tbl)
)
recipe_base
recipe_date <- recipe_base %>%
step_mutate(date = as.numeric(date))
Models
# Models ------------------------------------------------------------------
# Auto ARIMA --------------------------------------------------------------
model_spec_arima_no_boost <- arima_reg() %>%
set_engine(engine = "auto_arima")
wflw_auto_arima <- workflow() %>%
add_recipe(recipe = recipe_base) %>%
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_base) %>%
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 | 0.8166529 | 175.01566 | 0.6519883 | 163.7885 | 0.9436430 | 0.1099733 |
| healthyR.data | 2 | LM | Test | 0.7797320 | 169.69811 | 0.6225119 | 146.5980 | 0.9057676 | 0.0493791 |
| healthyR.data | 3 | EARTH | Test | 1.0496093 | 330.43926 | 0.8379729 | 131.9468 | 1.2606270 | 0.0493791 |
| healthyR.data | 4 | NNAR | Test | 0.7440144 | 116.48452 | 0.5939962 | 171.9089 | 0.9355045 | 0.0000001 |
| healthyR | 1 | ARIMA | Test | 0.7388047 | 104.16298 | 0.7297873 | 178.2227 | 0.9099680 | 0.0256899 |
| healthyR | 2 | LM | Test | 0.7115738 | 97.97049 | 0.7028887 | 185.6999 | 0.8742374 | 0.0715654 |
| healthyR | 3 | EARTH | Test | 0.7111615 | 97.21340 | 0.7024814 | 177.0310 | 0.8729731 | 0.0715654 |
| healthyR | 4 | NNAR | Test | 0.7476649 | 100.15380 | 0.7385394 | 158.6677 | 0.9256955 | 0.0473272 |
| healthyR.ts | 1 | ARIMA | Test | 0.9143378 | 273.40720 | 0.7268053 | 148.3628 | 1.1273960 | 0.0455431 |
| healthyR.ts | 2 | LM | Test | 0.9345655 | 326.40447 | 0.7428843 | 141.1867 | 1.1680833 | 0.0455431 |
| healthyR.ts | 3 | EARTH | Test | 0.9335866 | 322.95318 | 0.7421061 | 141.8001 | 1.1648705 | NA |
| healthyR.ts | 4 | NNAR | Test | 0.9431758 | 166.34090 | 0.7497286 | 179.3385 | 1.1791870 | 0.0337781 |
| healthyverse | 1 | ARIMA | Test | 0.6692285 | 351.45408 | 0.6858813 | 111.7667 | 0.8381570 | 0.0154100 |
| healthyverse | 2 | LM | Test | 0.6698987 | 432.19256 | 0.6865681 | 105.4821 | 0.8229659 | 0.0025692 |
| healthyverse | 3 | EARTH | Test | 0.6511266 | 263.86872 | 0.6673289 | 116.1471 | 0.8487658 | 0.0025692 |
| healthyverse | 4 | NNAR | Test | 0.6677176 | 247.36087 | 0.6843327 | 122.0862 | 0.8722359 | 0.0139842 |
| healthyR.ai | 1 | ARIMA | Test | 0.7254637 | 148.82572 | 0.7240198 | 179.4298 | 0.9424998 | 0.0405568 |
| healthyR.ai | 2 | LM | Test | 0.6664376 | 146.02753 | 0.6651112 | 146.8749 | 0.8685641 | 0.0359816 |
| healthyR.ai | 3 | EARTH | Test | 0.6717265 | 123.98120 | 0.6703895 | 153.3813 | 0.8802402 | 0.0359816 |
| healthyR.ai | 4 | NNAR | Test | 0.7896146 | 173.76226 | 0.7880430 | 161.5622 | 1.0330173 | 0.1332527 |
| TidyDensity | 1 | ARIMA | Test | 0.6205283 | 308.70846 | 0.6902220 | 111.5305 | 0.7536063 | 0.0569331 |
| TidyDensity | 2 | LM | Test | 0.6419715 | 367.50284 | 0.7140734 | 102.3069 | 0.8171988 | 0.0977793 |
| TidyDensity | 3 | EARTH | Test | 0.6171700 | 269.02775 | 0.6864865 | 109.4917 | 0.7734796 | 0.0977793 |
| TidyDensity | 4 | NNAR | Test | 0.6868838 | 253.30594 | 0.7640300 | 145.4068 | 0.8267306 | 0.0083488 |
| tidyAML | 1 | ARIMA | Test | 0.6427715 | 317.61615 | 0.8458465 | 101.6648 | 0.7999587 | 0.0377784 |
| tidyAML | 2 | LM | Test | 0.6540296 | 310.78721 | 0.8606616 | 103.8010 | 0.8058125 | 0.0343169 |
| tidyAML | 3 | EARTH | Test | 0.6803634 | 337.60786 | 0.8953151 | 104.3609 | 0.8313325 | 0.0343169 |
| tidyAML | 4 | NNAR | Test | 0.6408142 | 294.51774 | 0.8432709 | 104.3341 | 0.7879498 | 0.0024578 |
| RandomWalker | 1 | ARIMA | Test | 1.2134277 | 116.25743 | 0.5920482 | 173.2953 | 1.4373721 | 0.0021978 |
| RandomWalker | 2 | LM | Test | 1.2237908 | 98.82794 | 0.5971046 | 193.1199 | 1.4389783 | 0.0879682 |
| RandomWalker | 3 | EARTH | Test | 1.2085690 | 110.11240 | 0.5896776 | 170.2913 | 1.4376996 | NA |
| RandomWalker | 4 | NNAR | Test | 1.4349838 | 280.29409 | 0.7001486 | 156.5642 | 1.6259132 | 0.0000682 |
Plot Models
nested_modeltime_tbl %>%
extract_nested_test_forecast() %>%
group_by(package) %>%
plot_modeltime_forecast(
.interactive = FALSE,
.conf_interval_show = FALSE,
.facet_scales = "free"
) +
theme_minimal() +
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… 2 LM Test 0.780 170. 0.623 147. 0.906 0.0494
## 2 healthyR 3 EARTH Test 0.711 97.2 0.702 177. 0.873 0.0716
## 3 healthyR.ts 1 ARIMA Test 0.914 273. 0.727 148. 1.13 0.0455
## 4 healthyverse 2 LM Test 0.670 432. 0.687 105. 0.823 0.00257
## 5 healthyR.ai 2 LM Test 0.666 146. 0.665 147. 0.869 0.0360
## 6 TidyDensity 1 ARIMA Test 0.621 309. 0.690 112. 0.754 0.0569
## 7 tidyAML 4 NNAR Test 0.641 295. 0.843 104. 0.788 0.00246
## 8 RandomWalker 1 ARIMA Test 1.21 116. 0.592 173. 1.44 0.00220
best_nested_modeltime_tbl %>%
extract_nested_test_forecast() %>%
#filter(!is.na(.model_id)) %>%
group_by(package) %>%
plot_modeltime_forecast(
.interactive = FALSE,
.conf_interval_alpha = 0.2,
.facet_scales = "free"
) +
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 [1571|28]> <mdl_tm_t [1 × 5]>
## 2 healthyR <tibble> <tibble> <split [1564|28]> <mdl_tm_t [1 × 5]>
## 3 healthyR.ts <tibble> <tibble> <split [1508|28]> <mdl_tm_t [1 × 5]>
## 4 healthyverse <tibble> <tibble> <split [1479|28]> <mdl_tm_t [1 × 5]>
## 5 healthyR.ai <tibble> <tibble> <split [1303|28]> <mdl_tm_t [1 × 5]>
## 6 TidyDensity <tibble> <tibble> <split [1154|28]> <mdl_tm_t [1 × 5]>
## 7 tidyAML <tibble> <tibble> <split [762|28]> <mdl_tm_t [1 × 5]>
## 8 RandomWalker <tibble> <tibble> <split [184|28]> <mdl_tm_t [1 × 5]>
nested_modeltime_refit_tbl %>%
extract_nested_future_forecast() %>%
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)) %>%
group_by(package) %>%
plot_modeltime_forecast(
.interactive = FALSE,
.conf_interval_alpha = 0.2,
.facet_scales = "free"
) +
theme_minimal() +
theme(legend.position = "bottom")
