healthyverse_tsa
Time Series Analysis and Nested Modeling of the Healthyverse Packages
Steven P. Sanderson II, MPH - Date: 09 June, 2025
This analysis follows a Nested Modeltime Workflow.
Get Data
glimpse(downloads_tbl)
## Rows: 141,817
## 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 2025-06-07 23:26:08, the file was birthed on: 2024-08-07 07:35:44, and at report knit time is -7307.84 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 | 141817 |
| 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 | 102284 | 0.28 | 5 | 5 | 0 | 47 | 0 |
| r_arch | 102284 | 0.28 | 3 | 7 | 0 | 5 | 0 |
| r_os | 102284 | 0.28 | 7 | 15 | 0 | 22 | 0 |
| package | 0 | 1.00 | 7 | 13 | 0 | 8 | 0 |
| version | 0 | 1.00 | 5 | 17 | 0 | 60 | 0 |
| country | 12052 | 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 | 2025-06-07 | 2023-06-30 | 1658 |
Variable type: numeric
| skim_variable | n_missing | complete_rate | mean | sd | p0 | p25 | p50 | p75 | p100 | hist |
|---|---|---|---|---|---|---|---|---|---|---|
| size | 0 | 1 | 1133393.91 | 1517075.59 | 355 | 14701 | 289703 | 2367728 | 5677952 | ▇▁▂▁▁ |
| ip_id | 0 | 1 | 10432.09 | 18579.77 | 1 | 284 | 3039 | 11684 | 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 | 2025-06-07 23:26:08 | 2023-06-30 19:38:29 | 86870 |
Variable type: Timespan
| skim_variable | n_missing | complete_rate | min | max | median | n_unique |
|---|---|---|---|---|---|---|
| time | 0 | 1 | 0 | 59 | 12H 6M 47S | 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
## -146.72 -35.81 -11.05 26.84 814.94
##
## Coefficients:
## Estimate Std. Error
## (Intercept) -1.538e+02 6.801e+01
## date 9.637e-03 3.603e-03
## lag(value, 1) 1.050e-01 2.433e-02
## lag(value, 7) 9.664e-02 2.521e-02
## lag(value, 14) 8.917e-02 2.520e-02
## lag(value, 21) 6.880e-02 2.527e-02
## lag(value, 28) 7.085e-02 2.517e-02
## lag(value, 35) 6.726e-02 2.526e-02
## lag(value, 42) 5.100e-02 2.540e-02
## lag(value, 49) 6.913e-02 2.526e-02
## month(date, label = TRUE).L -9.569e+00 5.106e+00
## month(date, label = TRUE).Q 4.197e+00 5.114e+00
## month(date, label = TRUE).C -1.353e+01 5.128e+00
## month(date, label = TRUE)^4 -7.450e+00 5.163e+00
## month(date, label = TRUE)^5 -1.091e+01 5.113e+00
## month(date, label = TRUE)^6 -3.169e+00 5.208e+00
## month(date, label = TRUE)^7 -7.680e+00 5.090e+00
## month(date, label = TRUE)^8 -3.635e+00 5.094e+00
## month(date, label = TRUE)^9 6.147e+00 5.103e+00
## month(date, label = TRUE)^10 3.124e+00 5.077e+00
## month(date, label = TRUE)^11 -5.042e+00 5.189e+00
## fourier_vec(date, type = "sin", K = 1, period = 7) -1.148e+01 2.329e+00
## fourier_vec(date, type = "cos", K = 1, period = 7) 7.972e+00 2.449e+00
## t value Pr(>|t|)
## (Intercept) -2.262 0.023856 *
## date 2.675 0.007555 **
## lag(value, 1) 4.314 1.70e-05 ***
## lag(value, 7) 3.834 0.000131 ***
## lag(value, 14) 3.539 0.000413 ***
## lag(value, 21) 2.722 0.006551 **
## lag(value, 28) 2.815 0.004937 **
## lag(value, 35) 2.663 0.007826 **
## lag(value, 42) 2.008 0.044792 *
## lag(value, 49) 2.736 0.006290 **
## month(date, label = TRUE).L -1.874 0.061091 .
## month(date, label = TRUE).Q 0.821 0.411978
## month(date, label = TRUE).C -2.638 0.008420 **
## month(date, label = TRUE)^4 -1.443 0.149221
## month(date, label = TRUE)^5 -2.134 0.032977 *
## month(date, label = TRUE)^6 -0.608 0.542959
## month(date, label = TRUE)^7 -1.509 0.131561
## month(date, label = TRUE)^8 -0.714 0.475617
## month(date, label = TRUE)^9 1.204 0.228581
## month(date, label = TRUE)^10 0.615 0.538348
## month(date, label = TRUE)^11 -0.972 0.331391
## fourier_vec(date, type = "sin", K = 1, period = 7) -4.928 9.17e-07 ***
## fourier_vec(date, type = "cos", K = 1, period = 7) 3.256 0.001154 **
## ---
## Signif. codes: 0 '***' 0.001 '**' 0.01 '*' 0.05 '.' 0.1 ' ' 1
##
## Residual standard error: 58.68 on 1586 degrees of freedom
## (49 observations deleted due to missingness)
## Multiple R-squared: 0.2394, Adjusted R-squared: 0.2288
## F-statistic: 22.69 on 22 and 1586 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( 21 ) ...)"
## [1] "CURRENT lin OBJECTIVE FUNCTION = 2.7430823524398"
## [1] "NNS.ARMA(... method = 'lin' , seasonal.factor = c( 21, 63 ) ...)"
## [1] "CURRENT lin OBJECTIVE FUNCTION = 2.46052619688832"
## [1] "NNS.ARMA(... method = 'lin' , seasonal.factor = c( 21, 63, 91 ) ...)"
## [1] "CURRENT lin OBJECTIVE FUNCTION = 2.36262847890346"
## [1] "BEST method = 'lin', seasonal.factor = c( 21, 63, 91 )"
## [1] "BEST lin OBJECTIVE FUNCTION = 2.36262847890346"
## [1] "CURRNET METHOD: nonlin"
## [1] "COPY LATEST PARAMETERS DIRECTLY FOR NNS.ARMA() IF ERROR:"
## [1] "NNS.ARMA(... method = 'nonlin' , seasonal.factor = c( 21, 63, 91 ) ...)"
## [1] "CURRENT nonlin OBJECTIVE FUNCTION = 4.44580036577609"
## [1] "BEST method = 'nonlin' PATH MEMBER = c( 21, 63, 91 )"
## [1] "BEST nonlin OBJECTIVE FUNCTION = 4.44580036577609"
## [1] "CURRNET METHOD: both"
## [1] "COPY LATEST PARAMETERS DIRECTLY FOR NNS.ARMA() IF ERROR:"
## [1] "NNS.ARMA(... method = 'both' , seasonal.factor = c( 21, 63, 91 ) ...)"
## [1] "CURRENT both OBJECTIVE FUNCTION = 3.00473680136983"
## [1] "BEST method = 'both' PATH MEMBER = c( 21, 63, 91 )"
## [1] "BEST both OBJECTIVE FUNCTION = 3.00473680136983"
## 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 = 3.68227576034834"
## [1] "BEST method = 'lin', seasonal.factor = c( 63 )"
## [1] "BEST lin OBJECTIVE FUNCTION = 3.68227576034834"
## [1] "CURRNET METHOD: nonlin"
## [1] "COPY LATEST PARAMETERS DIRECTLY FOR NNS.ARMA() IF ERROR:"
## [1] "NNS.ARMA(... method = 'nonlin' , seasonal.factor = c( 63 ) ...)"
## [1] "CURRENT nonlin OBJECTIVE FUNCTION = 4.05378278722203"
## [1] "BEST method = 'nonlin' PATH MEMBER = c( 63 )"
## [1] "BEST nonlin OBJECTIVE FUNCTION = 4.05378278722203"
## [1] "CURRNET METHOD: both"
## [1] "COPY LATEST PARAMETERS DIRECTLY FOR NNS.ARMA() IF ERROR:"
## [1] "NNS.ARMA(... method = 'both' , seasonal.factor = c( 63 ) ...)"
## [1] "CURRENT both OBJECTIVE FUNCTION = 3.74243003055983"
## [1] "BEST method = 'both' PATH MEMBER = c( 63 )"
## [1] "BEST both OBJECTIVE FUNCTION = 3.74243003055983"
## 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( 63 ) ...)"
## [1] "CURRENT lin OBJECTIVE FUNCTION = 2.55411104733254"
## [1] "NNS.ARMA(... method = 'lin' , seasonal.factor = c( 63, 84 ) ...)"
## [1] "CURRENT lin OBJECTIVE FUNCTION = 2.32252538886903"
## [1] "NNS.ARMA(... method = 'lin' , seasonal.factor = c( 63, 84, 91 ) ...)"
## [1] "CURRENT lin OBJECTIVE FUNCTION = 2.22588949750934"
## [1] "BEST method = 'lin', seasonal.factor = c( 63, 84, 91 )"
## [1] "BEST lin OBJECTIVE FUNCTION = 2.22588949750934"
## [1] "CURRNET METHOD: nonlin"
## [1] "COPY LATEST PARAMETERS DIRECTLY FOR NNS.ARMA() IF ERROR:"
## [1] "NNS.ARMA(... method = 'nonlin' , seasonal.factor = c( 63, 84, 91 ) ...)"
## [1] "CURRENT nonlin OBJECTIVE FUNCTION = 2.31469558509605"
## [1] "BEST method = 'nonlin' PATH MEMBER = c( 63, 84, 91 )"
## [1] "BEST nonlin OBJECTIVE FUNCTION = 2.31469558509605"
## [1] "CURRNET METHOD: both"
## [1] "COPY LATEST PARAMETERS DIRECTLY FOR NNS.ARMA() IF ERROR:"
## [1] "NNS.ARMA(... method = 'both' , seasonal.factor = c( 63, 84, 91 ) ...)"
## [1] "CURRENT both OBJECTIVE FUNCTION = 2.12231407694237"
## [1] "BEST method = 'both' PATH MEMBER = c( 63, 84, 91 )"
## [1] "BEST both OBJECTIVE FUNCTION = 2.12231407694237"
## 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( 21 ) ...)"
## [1] "CURRENT lin OBJECTIVE FUNCTION = 3.16511889505517"
## [1] "NNS.ARMA(... method = 'lin' , seasonal.factor = c( 21, 56 ) ...)"
## [1] "CURRENT lin OBJECTIVE FUNCTION = 2.8181038950464"
## [1] "NNS.ARMA(... method = 'lin' , seasonal.factor = c( 21, 56, 63 ) ...)"
## [1] "CURRENT lin OBJECTIVE FUNCTION = 2.69743722439823"
## [1] "BEST method = 'lin', seasonal.factor = c( 21, 56, 63 )"
## [1] "BEST lin OBJECTIVE FUNCTION = 2.69743722439823"
## [1] "CURRNET METHOD: nonlin"
## [1] "COPY LATEST PARAMETERS DIRECTLY FOR NNS.ARMA() IF ERROR:"
## [1] "NNS.ARMA(... method = 'nonlin' , seasonal.factor = c( 21, 56, 63 ) ...)"
## [1] "CURRENT nonlin OBJECTIVE FUNCTION = 5.17111413785284"
## [1] "BEST method = 'nonlin' PATH MEMBER = c( 21, 56, 63 )"
## [1] "BEST nonlin OBJECTIVE FUNCTION = 5.17111413785284"
## [1] "CURRNET METHOD: both"
## [1] "COPY LATEST PARAMETERS DIRECTLY FOR NNS.ARMA() IF ERROR:"
## [1] "NNS.ARMA(... method = 'both' , seasonal.factor = c( 21, 56, 63 ) ...)"
## [1] "CURRENT both OBJECTIVE FUNCTION = 3.16246033198328"
## [1] "BEST method = 'both' PATH MEMBER = c( 21, 56, 63 )"
## [1] "BEST both OBJECTIVE FUNCTION = 3.16246033198328"
## Package: healthyverse
## [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 = 4.19172370298605"
## [1] "NNS.ARMA(... method = 'lin' , seasonal.factor = c( 28, 56 ) ...)"
## [1] "CURRENT lin OBJECTIVE FUNCTION = 4.01373480488995"
## [1] "BEST method = 'lin', seasonal.factor = c( 28, 56 )"
## [1] "BEST lin OBJECTIVE FUNCTION = 4.01373480488995"
## [1] "CURRNET METHOD: nonlin"
## [1] "COPY LATEST PARAMETERS DIRECTLY FOR NNS.ARMA() IF ERROR:"
## [1] "NNS.ARMA(... method = 'nonlin' , seasonal.factor = c( 28, 56 ) ...)"
## [1] "CURRENT nonlin OBJECTIVE FUNCTION = 4.83478312995496"
## [1] "BEST method = 'nonlin' PATH MEMBER = c( 28, 56 )"
## [1] "BEST nonlin OBJECTIVE FUNCTION = 4.83478312995496"
## [1] "CURRNET METHOD: both"
## [1] "COPY LATEST PARAMETERS DIRECTLY FOR NNS.ARMA() IF ERROR:"
## [1] "NNS.ARMA(... method = 'both' , seasonal.factor = c( 28, 56 ) ...)"
## [1] "CURRENT both OBJECTIVE FUNCTION = 4.0085728844063"
## [1] "BEST method = 'both' PATH MEMBER = c( 28, 56 )"
## [1] "BEST both OBJECTIVE FUNCTION = 4.0085728844063"
## Package: RandomWalker
## [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 = 3.2257661577086"
## [1] "BEST method = 'lin', seasonal.factor = c( 63 )"
## [1] "BEST lin OBJECTIVE FUNCTION = 3.2257661577086"
## [1] "CURRNET METHOD: nonlin"
## [1] "COPY LATEST PARAMETERS DIRECTLY FOR NNS.ARMA() IF ERROR:"
## [1] "NNS.ARMA(... method = 'nonlin' , seasonal.factor = c( 63 ) ...)"
## [1] "CURRENT nonlin OBJECTIVE FUNCTION = 3.25568514794957"
## [1] "BEST method = 'nonlin' PATH MEMBER = c( 63 )"
## [1] "BEST nonlin OBJECTIVE FUNCTION = 3.25568514794957"
## [1] "CURRNET METHOD: both"
## [1] "COPY LATEST PARAMETERS DIRECTLY FOR NNS.ARMA() IF ERROR:"
## [1] "NNS.ARMA(... method = 'both' , seasonal.factor = c( 63 ) ...)"
## [1] "CURRENT both OBJECTIVE FUNCTION = 3.06234849163741"
## [1] "BEST method = 'both' PATH MEMBER = c( 63 )"
## [1] "BEST both OBJECTIVE FUNCTION = 3.06234849163741"
## Package: tidyAML
## [1] "CURRNET METHOD: lin"
## [1] "COPY LATEST PARAMETERS DIRECTLY FOR NNS.ARMA() IF ERROR:"
## [1] "NNS.ARMA(... method = 'lin' , seasonal.factor = c( 84 ) ...)"
## [1] "CURRENT lin OBJECTIVE FUNCTION = 6.43697401810396"
## [1] "NNS.ARMA(... method = 'lin' , seasonal.factor = c( 84, 91 ) ...)"
## [1] "CURRENT lin OBJECTIVE FUNCTION = 6.34962386146168"
## [1] "BEST method = 'lin', seasonal.factor = c( 84, 91 )"
## [1] "BEST lin OBJECTIVE FUNCTION = 6.34962386146168"
## [1] "CURRNET METHOD: nonlin"
## [1] "COPY LATEST PARAMETERS DIRECTLY FOR NNS.ARMA() IF ERROR:"
## [1] "NNS.ARMA(... method = 'nonlin' , seasonal.factor = c( 84, 91 ) ...)"
## [1] "CURRENT nonlin OBJECTIVE FUNCTION = 4.25366230586835"
## [1] "BEST method = 'nonlin' PATH MEMBER = c( 84, 91 )"
## [1] "BEST nonlin OBJECTIVE FUNCTION = 4.25366230586835"
## [1] "CURRNET METHOD: both"
## [1] "COPY LATEST PARAMETERS DIRECTLY FOR NNS.ARMA() IF ERROR:"
## [1] "NNS.ARMA(... method = 'both' , seasonal.factor = c( 84, 91 ) ...)"
## [1] "CURRENT both OBJECTIVE FUNCTION = 4.99287515264327"
## [1] "BEST method = 'both' PATH MEMBER = c( 84, 91 )"
## [1] "BEST both OBJECTIVE FUNCTION = 4.99287515264327"
## 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 = 3.74524832904839"
## [1] "NNS.ARMA(... method = 'lin' , seasonal.factor = c( 63, 56 ) ...)"
## [1] "CURRENT lin OBJECTIVE FUNCTION = 3.47356558778286"
## [1] "NNS.ARMA(... method = 'lin' , seasonal.factor = c( 63, 56, 35 ) ...)"
## [1] "CURRENT lin OBJECTIVE FUNCTION = 3.38286731954275"
## [1] "BEST method = 'lin', seasonal.factor = c( 63, 56, 35 )"
## [1] "BEST lin OBJECTIVE FUNCTION = 3.38286731954275"
## [1] "CURRNET METHOD: nonlin"
## [1] "COPY LATEST PARAMETERS DIRECTLY FOR NNS.ARMA() IF ERROR:"
## [1] "NNS.ARMA(... method = 'nonlin' , seasonal.factor = c( 63, 56, 35 ) ...)"
## [1] "CURRENT nonlin OBJECTIVE FUNCTION = 4.78422409893306"
## [1] "BEST method = 'nonlin' PATH MEMBER = c( 63, 56, 35 )"
## [1] "BEST nonlin OBJECTIVE FUNCTION = 4.78422409893306"
## [1] "CURRNET METHOD: both"
## [1] "COPY LATEST PARAMETERS DIRECTLY FOR NNS.ARMA() IF ERROR:"
## [1] "NNS.ARMA(... method = 'both' , seasonal.factor = c( 63, 56, 35 ) ...)"
## [1] "CURRENT both OBJECTIVE FUNCTION = 3.66678773858143"
## [1] "BEST method = 'both' PATH MEMBER = c( 63, 56, 35 )"
## [1] "BEST both OBJECTIVE FUNCTION = 3.66678773858143"
## [[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 [1,650 × 2]> <tibble [28 × 2]> <split [1622|28]>
## 2 healthyR <tibble [1,644 × 2]> <tibble [28 × 2]> <split [1616|28]>
## 3 healthyR.ts <tibble [1,588 × 2]> <tibble [28 × 2]> <split [1560|28]>
## 4 healthyverse <tibble [1,558 × 2]> <tibble [28 × 2]> <split [1530|28]>
## 5 healthyR.ai <tibble [1,383 × 2]> <tibble [28 × 2]> <split [1355|28]>
## 6 TidyDensity <tibble [1,234 × 2]> <tibble [28 × 2]> <split [1206|28]>
## 7 tidyAML <tibble [842 × 2]> <tibble [28 × 2]> <split [814|28]>
## 8 RandomWalker <tibble [264 × 2]> <tibble [28 × 2]> <split [236|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 | 0.5855370 | 102.30733 | 0.6266358 | 128.70690 | 0.8020992 | 0.2055497 |
| healthyR.data | 2 | LM | Test | 0.6377634 | 206.75981 | 0.6825279 | 123.39294 | 0.7939722 | 0.0044239 |
| healthyR.data | 3 | EARTH | Test | 0.6576065 | 138.72253 | 0.7037639 | 146.84330 | 0.8625510 | 0.0044239 |
| healthyR.data | 4 | NNAR | Test | 0.7156268 | 95.18227 | 0.7658566 | 168.71399 | 0.9708544 | 0.0169405 |
| healthyR | 1 | ARIMA | Test | 0.7148223 | 123.00235 | 0.8703759 | 164.91372 | 0.8893949 | 0.0557124 |
| healthyR | 2 | LM | Test | 0.6793562 | 97.70083 | 0.8271919 | 167.53139 | 0.8648480 | 0.0502272 |
| healthyR | 3 | EARTH | Test | 0.6345182 | 128.86998 | 0.7725966 | 134.83236 | 0.7845044 | 0.0502272 |
| healthyR | 4 | NNAR | Test | 0.7000701 | 149.57695 | 0.8524134 | 166.65682 | 0.8571585 | 0.0245626 |
| healthyR.ts | 1 | ARIMA | Test | 0.6372057 | 102.92148 | 0.8125609 | 154.61933 | 0.7857825 | 0.0276328 |
| healthyR.ts | 2 | LM | Test | 0.9049160 | 161.96352 | 1.1539436 | 154.63168 | 1.1203716 | 0.0276328 |
| healthyR.ts | 3 | EARTH | Test | 0.5489115 | 157.42151 | 0.6999687 | 93.41615 | 0.6972032 | 0.0276328 |
| healthyR.ts | 4 | NNAR | Test | 0.7674924 | 109.87067 | 0.9787018 | 175.14898 | 0.9695342 | 0.0003261 |
| healthyverse | 1 | ARIMA | Test | 0.5973861 | 142.79604 | 1.0699211 | 78.62738 | 0.7592503 | 0.0141880 |
| healthyverse | 2 | LM | Test | 0.5832757 | 147.96670 | 1.0446493 | 75.66541 | 0.7129084 | 0.0305761 |
| healthyverse | 3 | EARTH | Test | 0.5890906 | 119.70138 | 1.0550638 | 78.17111 | 0.7650496 | NA |
| healthyverse | 4 | NNAR | Test | 0.6661535 | 100.58549 | 1.1930838 | 92.55448 | 0.8604255 | 0.0246351 |
| healthyR.ai | 1 | ARIMA | Test | 0.6395329 | 110.72057 | 0.9745449 | 172.58049 | 0.7969351 | 0.0234194 |
| healthyR.ai | 2 | LM | Test | 0.5678213 | 90.64147 | 0.8652681 | 124.13721 | 0.7337760 | 0.0463945 |
| healthyR.ai | 3 | EARTH | Test | 1.5170939 | 772.28622 | 2.3118063 | 117.30769 | 1.7342538 | 0.0463945 |
| healthyR.ai | 4 | NNAR | Test | 0.5892327 | 130.05376 | 0.8978955 | 125.50746 | 0.7586005 | 0.1139031 |
| TidyDensity | 1 | ARIMA | Test | 0.4935984 | 172.03605 | 1.0726726 | 112.53752 | 0.6396590 | 0.0239366 |
| TidyDensity | 2 | LM | Test | 0.6380096 | 267.23984 | 1.3865022 | 119.11747 | 0.8023462 | 0.0025582 |
| TidyDensity | 3 | EARTH | Test | 0.4761721 | 161.41729 | 1.0348023 | 111.38477 | 0.6208872 | 0.0025582 |
| TidyDensity | 4 | NNAR | Test | 0.4241940 | 110.95388 | 0.9218449 | 121.39702 | 0.5371477 | 0.0072082 |
| tidyAML | 1 | ARIMA | Test | 0.8063876 | 130.51856 | 0.8517556 | 101.19456 | 1.1166030 | 0.0011514 |
| tidyAML | 2 | LM | Test | 0.7847263 | 124.73835 | 0.8288757 | 100.31402 | 1.0904809 | 0.2796813 |
| tidyAML | 3 | EARTH | Test | 1.2695063 | 274.06927 | 1.3409298 | 120.40715 | 1.4888505 | 0.2796813 |
| tidyAML | 4 | NNAR | Test | 0.7546819 | 114.02068 | 0.7971410 | 100.72009 | 1.0658945 | 0.0064640 |
| RandomWalker | 1 | ARIMA | Test | 1.1015081 | 120.86909 | 0.6021124 | 123.70022 | 1.4039810 | 0.0172449 |
| RandomWalker | 2 | LM | Test | 1.2002443 | 118.06846 | 0.6560841 | 191.16879 | 1.3687606 | 0.0123125 |
| RandomWalker | 3 | EARTH | Test | 4.6809575 | 1181.24558 | 2.5587307 | 165.75031 | 5.2047281 | 0.0123125 |
| RandomWalker | 4 | NNAR | Test | 1.1558392 | 181.07405 | 0.6318112 | 156.68751 | 1.3411679 | 0.0401079 |
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 0.638 207. 0.683 123. 0.794 0.00442
## 2 healthyR 3 EARTH Test 0.635 129. 0.773 135. 0.785 0.0502
## 3 healthyR.ts 3 EARTH Test 0.549 157. 0.700 93.4 0.697 0.0276
## 4 healthyverse 2 LM Test 0.583 148. 1.04 75.7 0.713 0.0306
## 5 healthyR.ai 2 LM Test 0.568 90.6 0.865 124. 0.734 0.0464
## 6 TidyDensity 4 NNAR Test 0.424 111. 0.922 121. 0.537 0.00721
## 7 tidyAML 4 NNAR Test 0.755 114. 0.797 101. 1.07 0.00646
## 8 RandomWalker 4 NNAR Test 1.16 181. 0.632 157. 1.34 0.0401
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 [1622|28]> <mdl_tm_t [1 × 5]>
## 2 healthyR <tibble> <tibble> <split [1616|28]> <mdl_tm_t [1 × 5]>
## 3 healthyR.ts <tibble> <tibble> <split [1560|28]> <mdl_tm_t [1 × 5]>
## 4 healthyverse <tibble> <tibble> <split [1530|28]> <mdl_tm_t [1 × 5]>
## 5 healthyR.ai <tibble> <tibble> <split [1355|28]> <mdl_tm_t [1 × 5]>
## 6 TidyDensity <tibble> <tibble> <split [1206|28]> <mdl_tm_t [1 × 5]>
## 7 tidyAML <tibble> <tibble> <split [814|28]> <mdl_tm_t [1 × 5]>
## 8 RandomWalker <tibble> <tibble> <split [236|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")
