healthyverse_tsa
Time Series Analysis and Nested Modeling of the Healthyverse Packages
Steven P. Sanderson II, MPH - Date: 09 July, 2025
This analysis follows a Nested Modeltime Workflow.
Get Data
glimpse(downloads_tbl)
## Rows: 144,864
## 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 2025-07-07 21:30:55, the file was birthed on: 2024-08-07 07:35:44.428716, and at report knit time is -8025.92 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 | 144864 |
| 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 | 104720 | 0.28 | 5 | 5 | 0 | 48 | 0 |
| r_arch | 104720 | 0.28 | 3 | 7 | 0 | 5 | 0 |
| r_os | 104720 | 0.28 | 7 | 15 | 0 | 23 | 0 |
| package | 0 | 1.00 | 7 | 13 | 0 | 8 | 0 |
| version | 0 | 1.00 | 5 | 17 | 0 | 60 | 0 |
| country | 12235 | 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 | 2025-07-07 | 2023-07-19 | 1688 |
Variable type: numeric
| skim_variable | n_missing | complete_rate | mean | sd | p0 | p25 | p50 | p75 | p100 | hist |
|---|---|---|---|---|---|---|---|---|---|---|
| size | 0 | 1 | 1131584.65 | 1513383.09 | 355 | 14701 | 293058 | 2367682 | 5677952 | ▇▁▂▁▁ |
| ip_id | 0 | 1 | 10454.41 | 18585.19 | 1 | 285 | 3037 | 11827 | 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 | 2025-07-07 21:30:55 | 2023-07-19 18:19:49 | 88973 |
Variable type: Timespan
| skim_variable | n_missing | complete_rate | min | max | median | n_unique |
|---|---|---|---|---|---|---|
| time | 0 | 1 | 0 | 59 | 16 | 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
## -148.78 -36.03 -11.32 26.71 816.07
##
## Coefficients:
## Estimate Std. Error
## (Intercept) -1.718e+02 6.584e+01
## date 1.060e-02 3.486e-03
## lag(value, 1) 1.044e-01 2.411e-02
## lag(value, 7) 9.437e-02 2.495e-02
## lag(value, 14) 8.667e-02 2.494e-02
## lag(value, 21) 6.554e-02 2.511e-02
## lag(value, 28) 7.064e-02 2.499e-02
## lag(value, 35) 6.782e-02 2.508e-02
## lag(value, 42) 5.648e-02 2.519e-02
## lag(value, 49) 6.537e-02 2.506e-02
## month(date, label = TRUE).L -9.740e+00 5.109e+00
## month(date, label = TRUE).Q 3.369e+00 5.057e+00
## month(date, label = TRUE).C -1.331e+01 5.125e+00
## month(date, label = TRUE)^4 -6.816e+00 5.113e+00
## month(date, label = TRUE)^5 -1.131e+01 5.105e+00
## month(date, label = TRUE)^6 -4.068e+00 5.162e+00
## month(date, label = TRUE)^7 -7.101e+00 5.065e+00
## month(date, label = TRUE)^8 -3.011e+00 5.054e+00
## month(date, label = TRUE)^9 5.271e+00 5.044e+00
## month(date, label = TRUE)^10 2.629e+00 5.050e+00
## month(date, label = TRUE)^11 -3.686e+00 5.023e+00
## fourier_vec(date, type = "sin", K = 1, period = 7) -1.180e+01 2.307e+00
## fourier_vec(date, type = "cos", K = 1, period = 7) 8.115e+00 2.427e+00
## t value Pr(>|t|)
## (Intercept) -2.610 0.009142 **
## date 3.042 0.002389 **
## lag(value, 1) 4.331 1.58e-05 ***
## lag(value, 7) 3.782 0.000161 ***
## lag(value, 14) 3.475 0.000525 ***
## lag(value, 21) 2.609 0.009152 **
## lag(value, 28) 2.827 0.004753 **
## lag(value, 35) 2.704 0.006926 **
## lag(value, 42) 2.242 0.025092 *
## lag(value, 49) 2.609 0.009178 **
## month(date, label = TRUE).L -1.907 0.056757 .
## month(date, label = TRUE).Q 0.666 0.505392
## month(date, label = TRUE).C -2.596 0.009504 **
## month(date, label = TRUE)^4 -1.333 0.182691
## month(date, label = TRUE)^5 -2.216 0.026844 *
## month(date, label = TRUE)^6 -0.788 0.430823
## month(date, label = TRUE)^7 -1.402 0.161088
## month(date, label = TRUE)^8 -0.596 0.551423
## month(date, label = TRUE)^9 1.045 0.296219
## month(date, label = TRUE)^10 0.521 0.602696
## month(date, label = TRUE)^11 -0.734 0.463117
## fourier_vec(date, type = "sin", K = 1, period = 7) -5.117 3.47e-07 ***
## fourier_vec(date, type = "cos", K = 1, period = 7) 3.344 0.000845 ***
## ---
## Signif. codes: 0 '***' 0.001 '**' 0.01 '*' 0.05 '.' 0.1 ' ' 1
##
## Residual standard error: 58.73 on 1616 degrees of freedom
## (49 observations deleted due to missingness)
## Multiple R-squared: 0.2379, Adjusted R-squared: 0.2275
## F-statistic: 22.93 on 22 and 1616 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 = 4.06441815427408"
## [1] "NNS.ARMA(... method = 'lin' , seasonal.factor = c( 21, 63 ) ...)"
## [1] "CURRENT lin OBJECTIVE FUNCTION = 3.99997022059981"
## [1] "BEST method = 'lin', seasonal.factor = c( 21, 63 )"
## [1] "BEST lin OBJECTIVE FUNCTION = 3.99997022059981"
## [1] "CURRNET METHOD: nonlin"
## [1] "COPY LATEST PARAMETERS DIRECTLY FOR NNS.ARMA() IF ERROR:"
## [1] "NNS.ARMA(... method = 'nonlin' , seasonal.factor = c( 21, 63 ) ...)"
## [1] "CURRENT nonlin OBJECTIVE FUNCTION = 5.71904938021195"
## [1] "BEST method = 'nonlin' PATH MEMBER = c( 21, 63 )"
## [1] "BEST nonlin OBJECTIVE FUNCTION = 5.71904938021195"
## [1] "CURRNET METHOD: both"
## [1] "COPY LATEST PARAMETERS DIRECTLY FOR NNS.ARMA() IF ERROR:"
## [1] "NNS.ARMA(... method = 'both' , seasonal.factor = c( 21, 63 ) ...)"
## [1] "CURRENT both OBJECTIVE FUNCTION = 5.14228495373323"
## [1] "BEST method = 'both' PATH MEMBER = c( 21, 63 )"
## [1] "BEST both OBJECTIVE FUNCTION = 5.14228495373323"
## 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( 7 ) ...)"
## [1] "CURRENT lin OBJECTIVE FUNCTION = 3.60033053662806"
## [1] "NNS.ARMA(... method = 'lin' , seasonal.factor = c( 7, 42 ) ...)"
## [1] "CURRENT lin OBJECTIVE FUNCTION = 3.19460828948664"
## [1] "NNS.ARMA(... method = 'lin' , seasonal.factor = c( 7, 42, 49 ) ...)"
## [1] "CURRENT lin OBJECTIVE FUNCTION = 2.93008969213022"
## [1] "NNS.ARMA(... method = 'lin' , seasonal.factor = c( 7, 42, 49, 63 ) ...)"
## [1] "CURRENT lin OBJECTIVE FUNCTION = 2.58776659762755"
## [1] "NNS.ARMA(... method = 'lin' , seasonal.factor = c( 7, 42, 49, 63, 98 ) ...)"
## [1] "CURRENT lin OBJECTIVE FUNCTION = 2.51456616972992"
## [1] "BEST method = 'lin', seasonal.factor = c( 7, 42, 49, 63, 98 )"
## [1] "BEST lin OBJECTIVE FUNCTION = 2.51456616972992"
## [1] "CURRNET METHOD: nonlin"
## [1] "COPY LATEST PARAMETERS DIRECTLY FOR NNS.ARMA() IF ERROR:"
## [1] "NNS.ARMA(... method = 'nonlin' , seasonal.factor = c( 7, 42, 49, 63, 98 ) ...)"
## [1] "CURRENT nonlin OBJECTIVE FUNCTION = 12.0371882649282"
## [1] "BEST method = 'nonlin' PATH MEMBER = c( 7, 42, 49, 63, 98 )"
## [1] "BEST nonlin OBJECTIVE FUNCTION = 12.0371882649282"
## [1] "CURRNET METHOD: both"
## [1] "COPY LATEST PARAMETERS DIRECTLY FOR NNS.ARMA() IF ERROR:"
## [1] "NNS.ARMA(... method = 'both' , seasonal.factor = c( 7, 42, 49, 63, 98 ) ...)"
## [1] "CURRENT both OBJECTIVE FUNCTION = 5.34961924971309"
## [1] "BEST method = 'both' PATH MEMBER = c( 7, 42, 49, 63, 98 )"
## [1] "BEST both OBJECTIVE FUNCTION = 5.34961924971309"
## 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( 77 ) ...)"
## [1] "CURRENT lin OBJECTIVE FUNCTION = 3.1150174215318"
## [1] "NNS.ARMA(... method = 'lin' , seasonal.factor = c( 77, 42 ) ...)"
## [1] "CURRENT lin OBJECTIVE FUNCTION = 2.80398201202113"
## [1] "BEST method = 'lin', seasonal.factor = c( 77, 42 )"
## [1] "BEST lin OBJECTIVE FUNCTION = 2.80398201202113"
## [1] "CURRNET METHOD: nonlin"
## [1] "COPY LATEST PARAMETERS DIRECTLY FOR NNS.ARMA() IF ERROR:"
## [1] "NNS.ARMA(... method = 'nonlin' , seasonal.factor = c( 77, 42 ) ...)"
## [1] "CURRENT nonlin OBJECTIVE FUNCTION = 2.47617737784653"
## [1] "BEST method = 'nonlin' PATH MEMBER = c( 77, 42 )"
## [1] "BEST nonlin OBJECTIVE FUNCTION = 2.47617737784653"
## [1] "CURRNET METHOD: both"
## [1] "COPY LATEST PARAMETERS DIRECTLY FOR NNS.ARMA() IF ERROR:"
## [1] "NNS.ARMA(... method = 'both' , seasonal.factor = c( 77, 42 ) ...)"
## [1] "CURRENT both OBJECTIVE FUNCTION = 2.37975545264338"
## [1] "BEST method = 'both' PATH MEMBER = c( 77, 42 )"
## [1] "BEST both OBJECTIVE FUNCTION = 2.37975545264338"
## 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 = 2.180961870352"
## [1] "NNS.ARMA(... method = 'lin' , seasonal.factor = c( 63, 49 ) ...)"
## [1] "CURRENT lin OBJECTIVE FUNCTION = 2.11402508580056"
## [1] "BEST method = 'lin', seasonal.factor = c( 63, 49 )"
## [1] "BEST lin OBJECTIVE FUNCTION = 2.11402508580056"
## [1] "CURRNET METHOD: nonlin"
## [1] "COPY LATEST PARAMETERS DIRECTLY FOR NNS.ARMA() IF ERROR:"
## [1] "NNS.ARMA(... method = 'nonlin' , seasonal.factor = c( 63, 49 ) ...)"
## [1] "CURRENT nonlin OBJECTIVE FUNCTION = 14.348684603633"
## [1] "BEST method = 'nonlin' PATH MEMBER = c( 63, 49 )"
## [1] "BEST nonlin OBJECTIVE FUNCTION = 14.348684603633"
## [1] "CURRNET METHOD: both"
## [1] "COPY LATEST PARAMETERS DIRECTLY FOR NNS.ARMA() IF ERROR:"
## [1] "NNS.ARMA(... method = 'both' , seasonal.factor = c( 63, 49 ) ...)"
## [1] "CURRENT both OBJECTIVE FUNCTION = 5.58645665743536"
## [1] "BEST method = 'both' PATH MEMBER = c( 63, 49 )"
## [1] "BEST both OBJECTIVE FUNCTION = 5.58645665743536"
## Package: healthyverse
## [1] "CURRNET METHOD: lin"
## [1] "COPY LATEST PARAMETERS DIRECTLY FOR NNS.ARMA() IF ERROR:"
## [1] "NNS.ARMA(... method = 'lin' , seasonal.factor = c( 77 ) ...)"
## [1] "CURRENT lin OBJECTIVE FUNCTION = 4.40572305802886"
## [1] "NNS.ARMA(... method = 'lin' , seasonal.factor = c( 77, 63 ) ...)"
## [1] "CURRENT lin OBJECTIVE FUNCTION = 3.49730878545913"
## [1] "BEST method = 'lin', seasonal.factor = c( 77, 63 )"
## [1] "BEST lin OBJECTIVE FUNCTION = 3.49730878545913"
## [1] "CURRNET METHOD: nonlin"
## [1] "COPY LATEST PARAMETERS DIRECTLY FOR NNS.ARMA() IF ERROR:"
## [1] "NNS.ARMA(... method = 'nonlin' , seasonal.factor = c( 77, 63 ) ...)"
## [1] "CURRENT nonlin OBJECTIVE FUNCTION = 4.79994825981482"
## [1] "BEST method = 'nonlin' PATH MEMBER = c( 77, 63 )"
## [1] "BEST nonlin OBJECTIVE FUNCTION = 4.79994825981482"
## [1] "CURRNET METHOD: both"
## [1] "COPY LATEST PARAMETERS DIRECTLY FOR NNS.ARMA() IF ERROR:"
## [1] "NNS.ARMA(... method = 'both' , seasonal.factor = c( 77, 63 ) ...)"
## [1] "CURRENT both OBJECTIVE FUNCTION = 4.11550241366429"
## [1] "BEST method = 'both' PATH MEMBER = c( 77, 63 )"
## [1] "BEST both OBJECTIVE FUNCTION = 4.11550241366429"
## Package: RandomWalker
## [1] "CURRNET METHOD: lin"
## [1] "COPY LATEST PARAMETERS DIRECTLY FOR NNS.ARMA() IF ERROR:"
## [1] "NNS.ARMA(... method = 'lin' , seasonal.factor = c( 77 ) ...)"
## [1] "CURRENT lin OBJECTIVE FUNCTION = 2.04863313671662"
## [1] "NNS.ARMA(... method = 'lin' , seasonal.factor = c( 77, 49 ) ...)"
## [1] "CURRENT lin OBJECTIVE FUNCTION = 1.87720402400082"
## [1] "NNS.ARMA(... method = 'lin' , seasonal.factor = c( 77, 49, 42 ) ...)"
## [1] "CURRENT lin OBJECTIVE FUNCTION = 1.78624622621924"
## [1] "BEST method = 'lin', seasonal.factor = c( 77, 49, 42 )"
## [1] "BEST lin OBJECTIVE FUNCTION = 1.78624622621924"
## [1] "CURRNET METHOD: nonlin"
## [1] "COPY LATEST PARAMETERS DIRECTLY FOR NNS.ARMA() IF ERROR:"
## [1] "NNS.ARMA(... method = 'nonlin' , seasonal.factor = c( 77, 49, 42 ) ...)"
## [1] "CURRENT nonlin OBJECTIVE FUNCTION = 1.60227694331498"
## [1] "BEST method = 'nonlin' PATH MEMBER = c( 77, 49, 42 )"
## [1] "BEST nonlin OBJECTIVE FUNCTION = 1.60227694331498"
## [1] "CURRNET METHOD: both"
## [1] "COPY LATEST PARAMETERS DIRECTLY FOR NNS.ARMA() IF ERROR:"
## [1] "NNS.ARMA(... method = 'both' , seasonal.factor = c( 77, 49, 42 ) ...)"
## [1] "CURRENT both OBJECTIVE FUNCTION = 1.52631578009582"
## [1] "BEST method = 'both' PATH MEMBER = c( 77, 49, 42 )"
## [1] "BEST both OBJECTIVE FUNCTION = 1.52631578009582"
## 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 = 4.39880918004636"
## [1] "BEST method = 'lin', seasonal.factor = c( 84 )"
## [1] "BEST lin OBJECTIVE FUNCTION = 4.39880918004636"
## [1] "CURRNET METHOD: nonlin"
## [1] "COPY LATEST PARAMETERS DIRECTLY FOR NNS.ARMA() IF ERROR:"
## [1] "NNS.ARMA(... method = 'nonlin' , seasonal.factor = c( 84 ) ...)"
## [1] "CURRENT nonlin OBJECTIVE FUNCTION = 2.82357455153159"
## [1] "BEST method = 'nonlin' PATH MEMBER = c( 84 )"
## [1] "BEST nonlin OBJECTIVE FUNCTION = 2.82357455153159"
## [1] "CURRNET METHOD: both"
## [1] "COPY LATEST PARAMETERS DIRECTLY FOR NNS.ARMA() IF ERROR:"
## [1] "NNS.ARMA(... method = 'both' , seasonal.factor = c( 84 ) ...)"
## [1] "CURRENT both OBJECTIVE FUNCTION = 3.29418925930975"
## [1] "BEST method = 'both' PATH MEMBER = c( 84 )"
## [1] "BEST both OBJECTIVE FUNCTION = 3.29418925930975"
## 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 = 3.02132032187574"
## [1] "BEST method = 'lin', seasonal.factor = c( 63 )"
## [1] "BEST lin OBJECTIVE FUNCTION = 3.02132032187574"
## [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 = 9.59992274536936"
## [1] "BEST method = 'nonlin' PATH MEMBER = c( 63 )"
## [1] "BEST nonlin OBJECTIVE FUNCTION = 9.59992274536936"
## [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 = 5.60812282893831"
## [1] "BEST method = 'both' PATH MEMBER = c( 63 )"
## [1] "BEST both OBJECTIVE FUNCTION = 5.60812282893831"
## [[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 [1,680 × 2]> <tibble [28 × 2]> <split [1652|28]>
## 2 healthyR <tibble [1,674 × 2]> <tibble [28 × 2]> <split [1646|28]>
## 3 healthyR.ts <tibble [1,618 × 2]> <tibble [28 × 2]> <split [1590|28]>
## 4 healthyverse <tibble [1,588 × 2]> <tibble [28 × 2]> <split [1560|28]>
## 5 healthyR.ai <tibble [1,413 × 2]> <tibble [28 × 2]> <split [1385|28]>
## 6 TidyDensity <tibble [1,264 × 2]> <tibble [28 × 2]> <split [1236|28]>
## 7 tidyAML <tibble [872 × 2]> <tibble [28 × 2]> <split [844|28]>
## 8 RandomWalker <tibble [294 × 2]> <tibble [28 × 2]> <split [266|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 | 0.6387831 | 121.52354 | 0.8043779 | 124.99035 | 0.7748652 | 0.0066078 |
| healthyR.data | 2 | LM | Test | 0.6428732 | 127.06714 | 0.8095284 | 119.11102 | 0.7833459 | 0.0178960 |
| healthyR.data | 3 | NULL | NA | NA | NA | NA | NA | NA | NA |
| healthyR.data | 4 | NNAR | Test | 0.7023367 | 106.53792 | 0.8844069 | 191.94467 | 0.8212247 | 0.0075983 |
| healthyR | 1 | ARIMA | Test | 0.7165756 | 151.03138 | 0.7549643 | 151.49377 | 0.9424499 | 0.0212405 |
| healthyR | 2 | LM | Test | 0.7007660 | 100.86055 | 0.7383077 | 152.03003 | 0.9476467 | 0.0252489 |
| healthyR | 3 | NULL | NA | NA | NA | NA | NA | NA | NA |
| healthyR | 4 | NNAR | Test | 0.7072660 | 115.89929 | 0.7451560 | 149.91539 | 0.9495003 | 0.0012760 |
| healthyR.ts | 1 | ARIMA | Test | 0.8794214 | 138.80343 | 0.7119328 | 162.66473 | 1.1412892 | 0.0060599 |
| healthyR.ts | 2 | LM | Test | 0.8647180 | 156.47097 | 0.7000297 | 139.51937 | 1.1164574 | 0.0786569 |
| healthyR.ts | 3 | NULL | NA | NA | NA | NA | NA | NA | NA |
| healthyR.ts | 4 | NNAR | Test | 0.8357687 | 96.20768 | 0.6765938 | 174.71452 | 1.1226136 | 0.0893112 |
| healthyverse | 1 | ARIMA | Test | 0.6686801 | 228.61384 | 0.7549952 | 90.62072 | 0.8515577 | 0.0014272 |
| healthyverse | 2 | LM | Test | 0.6673951 | 232.01775 | 0.7535444 | 90.31555 | 0.8477696 | 0.0087579 |
| healthyverse | 3 | NULL | NA | NA | NA | NA | NA | NA | NA |
| healthyverse | 4 | NNAR | Test | 0.6570521 | 139.09648 | 0.7418662 | 101.32494 | 0.8665181 | 0.0218361 |
| healthyR.ai | 1 | ARIMA | Test | 0.6284083 | 104.63595 | 0.7283585 | 142.86817 | 0.7779658 | 0.0001081 |
| healthyR.ai | 2 | LM | Test | 0.6252170 | 101.68148 | 0.7246596 | 143.54901 | 0.7752920 | 0.0015936 |
| healthyR.ai | 3 | NULL | NA | NA | NA | NA | NA | NA | NA |
| healthyR.ai | 4 | NNAR | Test | 0.6282327 | 104.29692 | 0.7281549 | 143.41718 | 0.7739091 | 0.0048857 |
| TidyDensity | 1 | ARIMA | Test | 0.4972813 | 85.25293 | 0.7161799 | 107.17167 | 0.7142283 | 0.0123547 |
| TidyDensity | 2 | LM | Test | 0.4793384 | 126.22609 | 0.6903387 | 82.28127 | 0.6872720 | 0.0127604 |
| TidyDensity | 3 | NULL | NA | NA | NA | NA | NA | NA | NA |
| TidyDensity | 4 | NNAR | Test | 0.5424263 | 91.17799 | 0.7811973 | 124.34279 | 0.7490766 | 0.0019546 |
| tidyAML | 1 | ARIMA | Test | 0.5796674 | 140.11576 | 0.8061651 | 89.87722 | 0.7417726 | 0.0176508 |
| tidyAML | 2 | LM | Test | 0.5947299 | 165.19158 | 0.8271131 | 89.56647 | 0.7424083 | 0.0615052 |
| tidyAML | 3 | NULL | NA | NA | NA | NA | NA | NA | NA |
| tidyAML | 4 | NNAR | Test | 0.5807568 | 167.80904 | 0.8076802 | 87.54036 | 0.7223793 | 0.1041650 |
| RandomWalker | 1 | ARIMA | Test | 1.0605624 | 106.27020 | 0.6520745 | 155.27585 | 1.2565860 | 0.0246102 |
| RandomWalker | 2 | LM | Test | 1.1015392 | 100.44640 | 0.6772686 | 179.26902 | 1.2569323 | 0.0489512 |
| RandomWalker | 3 | NULL | NA | NA | NA | NA | NA | NA | NA |
| RandomWalker | 4 | NNAR | Test | 1.0495618 | 133.10017 | 0.6453109 | 140.97182 | 1.1989081 | 0.1292908 |
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… 1 ARIMA Test 0.639 122. 0.804 125. 0.775 0.00661
## 2 healthyR 1 ARIMA Test 0.717 151. 0.755 151. 0.942 0.0212
## 3 healthyR.ts 2 LM Test 0.865 156. 0.700 140. 1.12 0.0787
## 4 healthyverse 2 LM Test 0.667 232. 0.754 90.3 0.848 0.00876
## 5 healthyR.ai 4 NNAR Test 0.628 104. 0.728 143. 0.774 0.00489
## 6 TidyDensity 2 LM Test 0.479 126. 0.690 82.3 0.687 0.0128
## 7 tidyAML 4 NNAR Test 0.581 168. 0.808 87.5 0.722 0.104
## 8 RandomWalker 4 NNAR Test 1.05 133. 0.645 141. 1.20 0.129
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 [1652|28]> <mdl_tm_t [1 × 5]>
## 2 healthyR <tibble> <tibble> <split [1646|28]> <mdl_tm_t [1 × 5]>
## 3 healthyR.ts <tibble> <tibble> <split [1590|28]> <mdl_tm_t [1 × 5]>
## 4 healthyverse <tibble> <tibble> <split [1560|28]> <mdl_tm_t [1 × 5]>
## 5 healthyR.ai <tibble> <tibble> <split [1385|28]> <mdl_tm_t [1 × 5]>
## 6 TidyDensity <tibble> <tibble> <split [1236|28]> <mdl_tm_t [1 × 5]>
## 7 tidyAML <tibble> <tibble> <split [844|28]> <mdl_tm_t [1 × 5]>
## 8 RandomWalker <tibble> <tibble> <split [266|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")
