healthyverse_tsa
Time Series Analysis, Modeling and Forecasting of the Healthyverse
Packages Steven P. Sanderson II, MPH - Date: 2025-12-12
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: 162,689
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 2025-12-10 21:45:18, the file was birthed on: 2022-07-02 23:58:17.511888, and at report knit time is 3.016178^{4} 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 | 162689 |
| 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 | 119350 | 0.27 | 5 | 7 | 0 | 50 | 0 |
| r_arch | 119350 | 0.27 | 1 | 7 | 0 | 6 | 0 |
| r_os | 119350 | 0.27 | 7 | 19 | 0 | 24 | 0 |
| package | 0 | 1.00 | 7 | 13 | 0 | 8 | 0 |
| version | 0 | 1.00 | 5 | 17 | 0 | 62 | 0 |
| country | 15259 | 0.91 | 2 | 2 | 0 | 166 | 0 |
Variable type: Date
| skim_variable | n_missing | complete_rate | min | max | median | n_unique |
|---|---|---|---|---|---|---|
| date | 0 | 1 | 2020-11-23 | 2025-12-10 | 2023-11-06 | 1837 |
Variable type: numeric
| skim_variable | n_missing | complete_rate | mean | sd | p0 | p25 | p50 | p75 | p100 | hist |
|---|---|---|---|---|---|---|---|---|---|---|
| size | 0 | 1 | 1124236.46 | 1487771.95 | 355 | 27709 | 310297 | 2352686 | 5677952 | ▇▁▂▁▁ |
| ip_id | 0 | 1 | 11326.99 | 21968.41 | 1 | 236 | 2889 | 11961 | 299146 | ▇▁▁▁▁ |
Variable type: POSIXct
| skim_variable | n_missing | complete_rate | min | max | median | n_unique |
|---|---|---|---|---|---|---|
| date_time | 0 | 1 | 2020-11-23 09:00:41 | 2025-12-10 21:45:18 | 2023-11-06 02:23:35 | 102695 |
Variable type: Timespan
| skim_variable | n_missing | complete_rate | min | max | median | n_unique |
|---|---|---|---|---|---|---|
| time | 0 | 1 | 0 | 59 | 12H 6M 41S | 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
-146.39 -36.48 -11.25 27.07 819.98
Coefficients:
Estimate Std. Error
(Intercept) -1.818e+02 6.026e+01
date 1.115e-02 3.195e-03
lag(value, 1) 1.095e-01 2.318e-02
lag(value, 7) 8.916e-02 2.388e-02
lag(value, 14) 7.577e-02 2.385e-02
lag(value, 21) 8.111e-02 2.392e-02
lag(value, 28) 6.861e-02 2.383e-02
lag(value, 35) 5.499e-02 2.389e-02
lag(value, 42) 6.360e-02 2.401e-02
lag(value, 49) 6.144e-02 2.391e-02
month(date, label = TRUE).L -1.068e+01 5.010e+00
month(date, label = TRUE).Q 3.918e-01 4.930e+00
month(date, label = TRUE).C -1.596e+01 4.960e+00
month(date, label = TRUE)^4 -6.181e+00 4.937e+00
month(date, label = TRUE)^5 -6.748e+00 4.894e+00
month(date, label = TRUE)^6 1.218e+00 4.906e+00
month(date, label = TRUE)^7 -4.515e+00 4.842e+00
month(date, label = TRUE)^8 -4.035e+00 4.815e+00
month(date, label = TRUE)^9 2.764e+00 4.828e+00
month(date, label = TRUE)^10 9.390e-01 4.845e+00
month(date, label = TRUE)^11 -4.086e+00 4.831e+00
fourier_vec(date, type = "sin", K = 1, period = 7) -1.129e+01 2.210e+00
fourier_vec(date, type = "cos", K = 1, period = 7) 7.284e+00 2.290e+00
t value Pr(>|t|)
(Intercept) -3.017 0.002586 **
date 3.491 0.000493 ***
lag(value, 1) 4.726 2.47e-06 ***
lag(value, 7) 3.733 0.000195 ***
lag(value, 14) 3.177 0.001512 **
lag(value, 21) 3.391 0.000711 ***
lag(value, 28) 2.879 0.004035 **
lag(value, 35) 2.302 0.021457 *
lag(value, 42) 2.649 0.008153 **
lag(value, 49) 2.570 0.010254 *
month(date, label = TRUE).L -2.131 0.033217 *
month(date, label = TRUE).Q 0.079 0.936672
month(date, label = TRUE).C -3.218 0.001316 **
month(date, label = TRUE)^4 -1.252 0.210736
month(date, label = TRUE)^5 -1.379 0.168099
month(date, label = TRUE)^6 0.248 0.803901
month(date, label = TRUE)^7 -0.932 0.351248
month(date, label = TRUE)^8 -0.838 0.402144
month(date, label = TRUE)^9 0.572 0.567064
month(date, label = TRUE)^10 0.194 0.846333
month(date, label = TRUE)^11 -0.846 0.397797
fourier_vec(date, type = "sin", K = 1, period = 7) -5.105 3.66e-07 ***
fourier_vec(date, type = "cos", K = 1, period = 7) 3.180 0.001497 **
---
Signif. codes: 0 '***' 0.001 '**' 0.01 '*' 0.05 '.' 0.1 ' ' 1
Residual standard error: 59.26 on 1765 degrees of freedom
(49 observations deleted due to missingness)
Multiple R-squared: 0.2297, Adjusted R-squared: 0.2201
F-statistic: 23.92 on 22 and 1765 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( 1 ) ...)"
[1] "CURRENT lin OBJECTIVE FUNCTION = 176.569523303937"
[1] "BEST method = 'lin' PATH MEMBER = c( 1 )"
[1] "BEST lin OBJECTIVE FUNCTION = 176.569523303937"
[1] "CURRNET METHOD: nonlin"
[1] "COPY LATEST PARAMETERS DIRECTLY FOR NNS.ARMA() IF ERROR:"
[1] "NNS.ARMA(... method = 'nonlin' , seasonal.factor = c( 1 ) ...)"
[1] "CURRENT nonlin OBJECTIVE FUNCTION = 25.3448088330466"
[1] "BEST method = 'nonlin' PATH MEMBER = c( 1 )"
[1] "BEST nonlin OBJECTIVE FUNCTION = 25.3448088330466"
[1] "CURRNET METHOD: both"
[1] "COPY LATEST PARAMETERS DIRECTLY FOR NNS.ARMA() IF ERROR:"
[1] "NNS.ARMA(... method = 'both' , seasonal.factor = c( 1 ) ...)"
[1] "CURRENT both OBJECTIVE FUNCTION = 22.1424663968508"
[1] "BEST method = 'both' PATH MEMBER = c( 1 )"
[1] "BEST both OBJECTIVE FUNCTION = 22.1424663968508"
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( 3 ) ...)"
[1] "CURRENT lin OBJECTIVE FUNCTION = 20.9138420000945"
[1] "BEST method = 'lin' PATH MEMBER = c( 3 )"
[1] "BEST lin OBJECTIVE FUNCTION = 20.9138420000945"
[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 = 14.5908630375361"
[1] "BEST method = 'nonlin' PATH MEMBER = c( 3 )"
[1] "BEST nonlin OBJECTIVE FUNCTION = 14.5908630375361"
[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 = 48.0877344990938"
[1] "BEST method = 'both' PATH MEMBER = c( 3 )"
[1] "BEST both OBJECTIVE FUNCTION = 48.0877344990938"
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( 11 ) ...)"
[1] "CURRENT lin OBJECTIVE FUNCTION = 8.9122926658068"
[1] "BEST method = 'lin' PATH MEMBER = c( 11 )"
[1] "BEST lin OBJECTIVE FUNCTION = 8.9122926658068"
[1] "CURRNET METHOD: nonlin"
[1] "COPY LATEST PARAMETERS DIRECTLY FOR NNS.ARMA() IF ERROR:"
[1] "NNS.ARMA(... method = 'nonlin' , seasonal.factor = c( 11 ) ...)"
[1] "CURRENT nonlin OBJECTIVE FUNCTION = 8.71905526248609"
[1] "BEST method = 'nonlin' PATH MEMBER = c( 11 )"
[1] "BEST nonlin OBJECTIVE FUNCTION = 8.71905526248609"
[1] "CURRNET METHOD: both"
[1] "COPY LATEST PARAMETERS DIRECTLY FOR NNS.ARMA() IF ERROR:"
[1] "NNS.ARMA(... method = 'both' , seasonal.factor = c( 11 ) ...)"
[1] "CURRENT both OBJECTIVE FUNCTION = 8.35463837626691"
[1] "BEST method = 'both' PATH MEMBER = c( 11 )"
[1] "BEST both OBJECTIVE FUNCTION = 8.35463837626691"
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( 11 ) ...)"
[1] "CURRENT lin OBJECTIVE FUNCTION = 14.7944473567488"
[1] "BEST method = 'lin' PATH MEMBER = c( 11 )"
[1] "BEST lin OBJECTIVE FUNCTION = 14.7944473567488"
[1] "CURRNET METHOD: nonlin"
[1] "COPY LATEST PARAMETERS DIRECTLY FOR NNS.ARMA() IF ERROR:"
[1] "NNS.ARMA(... method = 'nonlin' , seasonal.factor = c( 11 ) ...)"
[1] "CURRENT nonlin OBJECTIVE FUNCTION = 24.5861935875251"
[1] "BEST method = 'nonlin' PATH MEMBER = c( 11 )"
[1] "BEST nonlin OBJECTIVE FUNCTION = 24.5861935875251"
[1] "CURRNET METHOD: both"
[1] "COPY LATEST PARAMETERS DIRECTLY FOR NNS.ARMA() IF ERROR:"
[1] "NNS.ARMA(... method = 'both' , seasonal.factor = c( 11 ) ...)"
[1] "CURRENT both OBJECTIVE FUNCTION = 20.5918707350301"
[1] "BEST method = 'both' PATH MEMBER = c( 11 )"
[1] "BEST both OBJECTIVE FUNCTION = 20.5918707350301"
Package: healthyverse
[1] "CURRNET METHOD: lin"
[1] "COPY LATEST PARAMETERS DIRECTLY FOR NNS.ARMA() IF ERROR:"
[1] "NNS.ARMA(... method = 'lin' , seasonal.factor = c( 4 ) ...)"
[1] "CURRENT lin OBJECTIVE FUNCTION = 10.9414658384525"
[1] "BEST method = 'lin' PATH MEMBER = c( 4 )"
[1] "BEST lin OBJECTIVE FUNCTION = 10.9414658384525"
[1] "CURRNET METHOD: nonlin"
[1] "COPY LATEST PARAMETERS DIRECTLY FOR NNS.ARMA() IF ERROR:"
[1] "NNS.ARMA(... method = 'nonlin' , seasonal.factor = c( 4 ) ...)"
[1] "CURRENT nonlin OBJECTIVE FUNCTION = 4.16594400914149"
[1] "BEST method = 'nonlin' PATH MEMBER = c( 4 )"
[1] "BEST nonlin OBJECTIVE FUNCTION = 4.16594400914149"
[1] "CURRNET METHOD: both"
[1] "COPY LATEST PARAMETERS DIRECTLY FOR NNS.ARMA() IF ERROR:"
[1] "NNS.ARMA(... method = 'both' , seasonal.factor = c( 4 ) ...)"
[1] "CURRENT both OBJECTIVE FUNCTION = 5.28954857371325"
[1] "BEST method = 'both' PATH MEMBER = c( 4 )"
[1] "BEST both OBJECTIVE FUNCTION = 5.28954857371325"
Package: RandomWalker
[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 = 9.71367670987443"
[1] "BEST method = 'lin' PATH MEMBER = c( 7 )"
[1] "BEST lin OBJECTIVE FUNCTION = 9.71367670987443"
[1] "CURRNET METHOD: nonlin"
[1] "COPY LATEST PARAMETERS DIRECTLY FOR NNS.ARMA() IF ERROR:"
[1] "NNS.ARMA(... method = 'nonlin' , seasonal.factor = c( 7 ) ...)"
[1] "CURRENT nonlin OBJECTIVE FUNCTION = 4.55830718323456"
[1] "BEST method = 'nonlin' PATH MEMBER = c( 7 )"
[1] "BEST nonlin OBJECTIVE FUNCTION = 4.55830718323456"
[1] "CURRNET METHOD: both"
[1] "COPY LATEST PARAMETERS DIRECTLY FOR NNS.ARMA() IF ERROR:"
[1] "NNS.ARMA(... method = 'both' , seasonal.factor = c( 7 ) ...)"
[1] "CURRENT both OBJECTIVE FUNCTION = 5.0221386561647"
[1] "BEST method = 'both' PATH MEMBER = c( 7 )"
[1] "BEST both OBJECTIVE FUNCTION = 5.0221386561647"
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 = 55.9824852532643"
[1] "BEST method = 'lin' PATH MEMBER = c( 2 )"
[1] "BEST lin OBJECTIVE FUNCTION = 55.9824852532643"
[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 = 51.0786998186261"
[1] "BEST method = 'nonlin' PATH MEMBER = c( 2 )"
[1] "BEST nonlin OBJECTIVE FUNCTION = 51.0786998186261"
[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 = 95.0599023826288"
[1] "BEST method = 'both' PATH MEMBER = c( 2 )"
[1] "BEST both OBJECTIVE FUNCTION = 95.0599023826288"
Package: TidyDensity
[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 = 10.0192470313871"
[1] "BEST method = 'lin' PATH MEMBER = c( 7 )"
[1] "BEST lin OBJECTIVE FUNCTION = 10.0192470313871"
[1] "CURRNET METHOD: nonlin"
[1] "COPY LATEST PARAMETERS DIRECTLY FOR NNS.ARMA() IF ERROR:"
[1] "NNS.ARMA(... method = 'nonlin' , seasonal.factor = c( 7 ) ...)"
[1] "CURRENT nonlin OBJECTIVE FUNCTION = 9.19230475783876"
[1] "BEST method = 'nonlin' PATH MEMBER = c( 7 )"
[1] "BEST nonlin OBJECTIVE FUNCTION = 9.19230475783876"
[1] "CURRNET METHOD: both"
[1] "COPY LATEST PARAMETERS DIRECTLY FOR NNS.ARMA() IF ERROR:"
[1] "NNS.ARMA(... method = 'both' , seasonal.factor = c( 7 ) ...)"
[1] "CURRENT both OBJECTIVE FUNCTION = 7.28727379984017"
[1] "BEST method = 'both' PATH MEMBER = c( 7 )"
[1] "BEST both OBJECTIVE FUNCTION = 7.28727379984017"
[[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 [1,829 × 50]> <tibble [28 × 50]> <split [1801|28]>
2 healthyR <tibble [1,820 × 50]> <tibble [28 × 50]> <split [1792|28]>
3 healthyR.ts <tibble [1,765 × 50]> <tibble [28 × 50]> <split [1737|28]>
4 healthyverse <tibble [1,736 × 50]> <tibble [28 × 50]> <split [1708|28]>
5 healthyR.ai <tibble [1,562 × 50]> <tibble [28 × 50]> <split [1534|28]>
6 TidyDensity <tibble [1,413 × 50]> <tibble [28 × 50]> <split [1385|28]>
7 tidyAML <tibble [1,020 × 50]> <tibble [28 × 50]> <split [992|28]>
8 RandomWalker <tibble [443 × 50]> <tibble [28 × 50]> <split [415|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 | 0.7422748 | 104.40889 | 0.7202115 | 151.84502 | 0.8922461 | 0.0968377 |
| healthyR.data | 2 | LM | Test | 0.6987373 | 192.97301 | 0.6779682 | 117.83168 | 0.8215401 | 0.0023428 |
| healthyR.data | 3 | EARTH | Test | 0.7777029 | 107.95896 | 0.7545866 | 162.80981 | 0.9348008 | 0.0324979 |
| healthyR.data | 4 | NNAR | Test | 0.7203570 | 206.72307 | 0.6989452 | 117.45817 | 0.8613187 | 0.0003843 |
| healthyR | 1 | ARIMA | Test | 0.6206681 | 287.08154 | 0.8913261 | 139.52289 | 0.7499832 | 0.0046133 |
| healthyR | 2 | LM | Test | 0.5400077 | 386.18720 | 0.7754916 | 107.96261 | 0.6462881 | 0.0788073 |
| healthyR | 3 | EARTH | Test | 0.7319908 | 789.92605 | 1.0511939 | 93.29021 | 0.9226627 | 0.0142535 |
| healthyR | 4 | NNAR | Test | 0.6259143 | 340.46524 | 0.8988600 | 136.10928 | 0.7197956 | 0.0380494 |
| healthyR.ts | 1 | ARIMA | Test | 0.5880230 | 219.13526 | 0.7099583 | 134.78962 | 0.7344289 | 0.0441402 |
| healthyR.ts | 2 | LM | Test | 0.7228873 | 272.47634 | 0.8727888 | 140.92345 | 0.8642904 | 0.0054960 |
| healthyR.ts | 3 | EARTH | Test | 0.5221763 | 187.38077 | 0.6304573 | 109.96546 | 0.6736939 | 0.0533812 |
| healthyR.ts | 4 | NNAR | Test | 0.7691787 | 260.00469 | 0.9286793 | 153.10259 | 0.9273533 | 0.0016830 |
| healthyverse | 1 | ARIMA | Test | 0.7027015 | 105.61902 | 0.7636533 | 133.61702 | 0.9206856 | 0.0033245 |
| healthyverse | 2 | LM | Test | 0.7694518 | 154.47515 | 0.8361936 | 131.74455 | 0.9033708 | 0.0068024 |
| healthyverse | 3 | EARTH | Test | 0.7074516 | 95.42670 | 0.7688155 | 137.13031 | 0.9399709 | NA |
| healthyverse | 4 | NNAR | Test | 0.7082230 | 144.40144 | 0.7696537 | 125.94000 | 0.8697719 | 0.0433339 |
| healthyR.ai | 1 | ARIMA | Test | 0.8444300 | 107.75673 | 1.0638781 | 192.18573 | 1.0229354 | 0.0289958 |
| healthyR.ai | 2 | LM | Test | 1.1174644 | 188.67375 | 1.4078680 | 165.38302 | 1.3312110 | 0.0466877 |
| healthyR.ai | 3 | EARTH | Test | 0.7115773 | 99.68899 | 0.8965000 | 121.25409 | 0.9135717 | 0.0668488 |
| healthyR.ai | 4 | NNAR | Test | 1.0468650 | 171.88162 | 1.3189214 | 147.83037 | 1.2722429 | 0.0451345 |
| TidyDensity | 1 | ARIMA | Test | 1.0870601 | 275.21574 | 0.7429908 | 160.95607 | 1.2383084 | 0.0298920 |
| TidyDensity | 2 | LM | Test | 0.8672790 | 128.23374 | 0.5927733 | 157.02084 | 1.0558537 | 0.0117261 |
| TidyDensity | 3 | EARTH | Test | 1.1276602 | 296.97873 | 0.7707403 | 163.71620 | 1.2568951 | 0.0000847 |
| TidyDensity | 4 | NNAR | Test | 0.9397759 | 164.68228 | 0.6423240 | 167.98021 | 1.0985928 | 0.0067000 |
| tidyAML | 1 | ARIMA | Test | 0.6294300 | 108.74555 | 0.8270355 | 178.39134 | 0.7892309 | 0.0023750 |
| tidyAML | 2 | LM | Test | 0.5984350 | 219.90947 | 0.7863098 | 124.16996 | 0.7656480 | 0.0482413 |
| tidyAML | 3 | EARTH | Test | 0.5491441 | 152.72646 | 0.7215443 | 104.80243 | 0.6813093 | 0.1275852 |
| tidyAML | 4 | NNAR | Test | 0.6238863 | 215.07668 | 0.8197514 | 122.48301 | 0.7592986 | 0.0179053 |
| RandomWalker | 1 | ARIMA | Test | 0.6740869 | 160.64262 | 0.6767488 | 146.44368 | 0.7846548 | 0.2107023 |
| RandomWalker | 2 | LM | Test | 0.8042935 | 168.79856 | 0.8074695 | 160.15269 | 0.9270806 | 0.0011328 |
| RandomWalker | 3 | EARTH | Test | 0.9254765 | 218.19338 | 0.9291310 | 167.13028 | 0.9821949 | 0.0039693 |
| RandomWalker | 4 | NNAR | Test | 0.9227835 | 189.75732 | 0.9264273 | 173.69103 | 1.0451480 | 0.0643235 |
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… 2 LM Test 0.699 193. 0.678 118. 0.822 0.00234
2 healthyR 2 LM Test 0.540 386. 0.775 108. 0.646 0.0788
3 healthyR.ts 3 EARTH Test 0.522 187. 0.630 110. 0.674 0.0534
4 healthyverse 4 NNAR Test 0.708 144. 0.770 126. 0.870 0.0433
5 healthyR.ai 3 EARTH Test 0.712 99.7 0.896 121. 0.914 0.0668
6 TidyDensity 2 LM Test 0.867 128. 0.593 157. 1.06 0.0117
7 tidyAML 3 EARTH Test 0.549 153. 0.722 105. 0.681 0.128
8 RandomWalker 1 ARIMA Test 0.674 161. 0.677 146. 0.785 0.211
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 [1801|28]> <mdl_tm_t [1 × 5]>
2 healthyR <tibble> <tibble> <split [1792|28]> <mdl_tm_t [1 × 5]>
3 healthyR.ts <tibble> <tibble> <split [1737|28]> <mdl_tm_t [1 × 5]>
4 healthyverse <tibble> <tibble> <split [1708|28]> <mdl_tm_t [1 × 5]>
5 healthyR.ai <tibble> <tibble> <split [1534|28]> <mdl_tm_t [1 × 5]>
6 TidyDensity <tibble> <tibble> <split [1385|28]> <mdl_tm_t [1 × 5]>
7 tidyAML <tibble> <tibble> <split [992|28]> <mdl_tm_t [1 × 5]>
8 RandomWalker <tibble> <tibble> <split [415|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")
