healthyverse_tsa
Time Series Analysis, Modeling and Forecasting of the Healthyverse
Packages Steven P. Sanderson II, MPH - Date: 2026-04-03
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: 174,042
Columns: 11
$ date <date> 2020-11-23, 2020-11-23, 2020-11-23, 2020-11-23, 2020-11-23,…
$ time <Period> 15H 36M 55S, 11H 26M 39S, 23H 34M 44S, 18H 39M 32S, 9H 0M…
$ date_time <dttm> 2020-11-23 15:36:55, 2020-11-23 11:26:39, 2020-11-23 23:34:…
$ size <int> 4858294, 4858294, 4858301, 4858295, 361, 4863722, 4864794, 4…
$ r_version <chr> NA, "4.0.3", "3.5.3", "3.5.2", NA, NA, NA, NA, NA, NA, NA, N…
$ r_arch <chr> NA, "x86_64", "x86_64", "x86_64", NA, NA, NA, NA, NA, NA, NA…
$ r_os <chr> NA, "mingw32", "mingw32", "linux-gnu", NA, NA, NA, NA, NA, N…
$ package <chr> "healthyR.data", "healthyR.data", "healthyR.data", "healthyR…
$ version <chr> "1.0.0", "1.0.0", "1.0.0", "1.0.0", "1.0.0", "1.0.0", "1.0.0…
$ country <chr> "US", "US", "US", "GB", "US", "US", "DE", "HK", "JP", "US", …
$ ip_id <int> 2069, 2804, 78827, 27595, 90474, 90474, 42435, 74, 7655, 638…
The last day in the data set is 2026-04-01 23:24:50, the file was birthed on: 2025-10-31 10:47:59.603742, and at report knit time is 3656.61 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 | 174042 |
| 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 | 129196 | 0.26 | 5 | 7 | 0 | 51 | 0 |
| r_arch | 129196 | 0.26 | 1 | 7 | 0 | 6 | 0 |
| r_os | 129196 | 0.26 | 7 | 19 | 0 | 24 | 0 |
| package | 0 | 1.00 | 7 | 13 | 0 | 8 | 0 |
| version | 0 | 1.00 | 5 | 17 | 0 | 63 | 0 |
| country | 16161 | 0.91 | 2 | 2 | 0 | 167 | 0 |
Variable type: Date
| skim_variable | n_missing | complete_rate | min | max | median | n_unique |
|---|---|---|---|---|---|---|
| date | 0 | 1 | 2020-11-23 | 2026-04-01 | 2024-01-10 | 1949 |
Variable type: numeric
| skim_variable | n_missing | complete_rate | mean | sd | p0 | p25 | p50 | p75 | p100 | hist |
|---|---|---|---|---|---|---|---|---|---|---|
| size | 0 | 1 | 1126813.35 | 1478444.58 | 355 | 43530 | 325156 | 2333727 | 5677952 | ▇▁▂▁▁ |
| ip_id | 0 | 1 | 11455.81 | 22866.32 | 1 | 199 | 2741 | 11721 | 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 | 2026-04-01 23:24:50 | 2024-01-10 05:09:02 | 110684 |
Variable type: Timespan
| skim_variable | n_missing | complete_rate | min | max | median | n_unique |
|---|---|---|---|---|---|---|
| time | 0 | 1 | 0 | 59 | 17 | 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
-150.38 -37.45 -11.57 27.69 826.76
Coefficients:
Estimate Std. Error
(Intercept) -1.749e+02 5.448e+01
date 1.087e-02 2.882e-03
lag(value, 1) 9.354e-02 2.263e-02
lag(value, 7) 7.634e-02 2.348e-02
lag(value, 14) 6.414e-02 2.336e-02
lag(value, 21) 9.138e-02 2.344e-02
lag(value, 28) 7.594e-02 2.334e-02
lag(value, 35) 4.589e-02 2.340e-02
lag(value, 42) 6.389e-02 2.354e-02
lag(value, 49) 7.568e-02 2.345e-02
month(date, label = TRUE).L -9.300e+00 4.756e+00
month(date, label = TRUE).Q -1.020e+00 4.776e+00
month(date, label = TRUE).C -1.463e+01 4.790e+00
month(date, label = TRUE)^4 -8.247e+00 4.788e+00
month(date, label = TRUE)^5 -5.221e+00 4.793e+00
month(date, label = TRUE)^6 -3.400e-01 4.832e+00
month(date, label = TRUE)^7 -3.574e+00 4.757e+00
month(date, label = TRUE)^8 -5.005e+00 4.757e+00
month(date, label = TRUE)^9 3.230e+00 4.822e+00
month(date, label = TRUE)^10 1.046e+00 4.881e+00
month(date, label = TRUE)^11 -4.344e+00 4.883e+00
fourier_vec(date, type = "sin", K = 1, period = 7) -1.133e+01 2.152e+00
fourier_vec(date, type = "cos", K = 1, period = 7) 7.386e+00 2.225e+00
t value Pr(>|t|)
(Intercept) -3.210 0.001348 **
date 3.771 0.000168 ***
lag(value, 1) 4.134 3.73e-05 ***
lag(value, 7) 3.251 0.001169 **
lag(value, 14) 2.746 0.006095 **
lag(value, 21) 3.898 0.000100 ***
lag(value, 28) 3.253 0.001161 **
lag(value, 35) 1.960 0.050093 .
lag(value, 42) 2.714 0.006700 **
lag(value, 49) 3.228 0.001270 **
month(date, label = TRUE).L -1.955 0.050705 .
month(date, label = TRUE).Q -0.214 0.830848
month(date, label = TRUE).C -3.054 0.002290 **
month(date, label = TRUE)^4 -1.723 0.085119 .
month(date, label = TRUE)^5 -1.089 0.276212
month(date, label = TRUE)^6 -0.070 0.943912
month(date, label = TRUE)^7 -0.751 0.452486
month(date, label = TRUE)^8 -1.052 0.292868
month(date, label = TRUE)^9 0.670 0.502971
month(date, label = TRUE)^10 0.214 0.830286
month(date, label = TRUE)^11 -0.890 0.373759
fourier_vec(date, type = "sin", K = 1, period = 7) -5.266 1.55e-07 ***
fourier_vec(date, type = "cos", K = 1, period = 7) 3.319 0.000921 ***
---
Signif. codes: 0 '***' 0.001 '**' 0.01 '*' 0.05 '.' 0.1 ' ' 1
Residual standard error: 59.94 on 1877 degrees of freedom
(49 observations deleted due to missingness)
Multiple R-squared: 0.215, Adjusted R-squared: 0.2058
F-statistic: 23.37 on 22 and 1877 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( 9 ) ...)"
[1] "CURRENT lin OBJECTIVE FUNCTION = 49.4161511080069"
[1] "BEST method = 'lin' PATH MEMBER = c( 9 )"
[1] "BEST lin OBJECTIVE FUNCTION = 49.4161511080069"
[1] "CURRNET METHOD: nonlin"
[1] "COPY LATEST PARAMETERS DIRECTLY FOR NNS.ARMA() IF ERROR:"
[1] "NNS.ARMA(... method = 'nonlin' , seasonal.factor = c( 9 ) ...)"
[1] "CURRENT nonlin OBJECTIVE FUNCTION = 8.06837966657089"
[1] "BEST method = 'nonlin' PATH MEMBER = c( 9 )"
[1] "BEST nonlin OBJECTIVE FUNCTION = 8.06837966657089"
[1] "CURRNET METHOD: both"
[1] "COPY LATEST PARAMETERS DIRECTLY FOR NNS.ARMA() IF ERROR:"
[1] "NNS.ARMA(... method = 'both' , seasonal.factor = c( 9 ) ...)"
[1] "CURRENT both OBJECTIVE FUNCTION = 12.0257366658761"
[1] "BEST method = 'both' PATH MEMBER = c( 9 )"
[1] "BEST both OBJECTIVE FUNCTION = 12.0257366658761"
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( 4 ) ...)"
[1] "CURRENT lin OBJECTIVE FUNCTION = 10.9495849147152"
[1] "BEST method = 'lin' PATH MEMBER = c( 4 )"
[1] "BEST lin OBJECTIVE FUNCTION = 10.9495849147152"
[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 = 19.5399310985138"
[1] "BEST method = 'nonlin' PATH MEMBER = c( 4 )"
[1] "BEST nonlin OBJECTIVE FUNCTION = 19.5399310985138"
[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 = 29.9278931568024"
[1] "BEST method = 'both' PATH MEMBER = c( 4 )"
[1] "BEST both OBJECTIVE FUNCTION = 29.9278931568024"
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( 13 ) ...)"
[1] "CURRENT lin OBJECTIVE FUNCTION = 11.3366668224515"
[1] "BEST method = 'lin' PATH MEMBER = c( 13 )"
[1] "BEST lin OBJECTIVE FUNCTION = 11.3366668224515"
[1] "CURRNET METHOD: nonlin"
[1] "COPY LATEST PARAMETERS DIRECTLY FOR NNS.ARMA() IF ERROR:"
[1] "NNS.ARMA(... method = 'nonlin' , seasonal.factor = c( 13 ) ...)"
[1] "CURRENT nonlin OBJECTIVE FUNCTION = 12.6153402278014"
[1] "BEST method = 'nonlin' PATH MEMBER = c( 13 )"
[1] "BEST nonlin OBJECTIVE FUNCTION = 12.6153402278014"
[1] "CURRNET METHOD: both"
[1] "COPY LATEST PARAMETERS DIRECTLY FOR NNS.ARMA() IF ERROR:"
[1] "NNS.ARMA(... method = 'both' , seasonal.factor = c( 13 ) ...)"
[1] "CURRENT both OBJECTIVE FUNCTION = 11.1585031460308"
[1] "BEST method = 'both' PATH MEMBER = c( 13 )"
[1] "BEST both OBJECTIVE FUNCTION = 11.1585031460308"
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( 19 ) ...)"
[1] "CURRENT lin OBJECTIVE FUNCTION = 22.9338928771366"
[1] "BEST method = 'lin' PATH MEMBER = c( 19 )"
[1] "BEST lin OBJECTIVE FUNCTION = 22.9338928771366"
[1] "CURRNET METHOD: nonlin"
[1] "COPY LATEST PARAMETERS DIRECTLY FOR NNS.ARMA() IF ERROR:"
[1] "NNS.ARMA(... method = 'nonlin' , seasonal.factor = c( 19 ) ...)"
[1] "CURRENT nonlin OBJECTIVE FUNCTION = 15.3894163987613"
[1] "BEST method = 'nonlin' PATH MEMBER = c( 19 )"
[1] "BEST nonlin OBJECTIVE FUNCTION = 15.3894163987613"
[1] "CURRNET METHOD: both"
[1] "COPY LATEST PARAMETERS DIRECTLY FOR NNS.ARMA() IF ERROR:"
[1] "NNS.ARMA(... method = 'both' , seasonal.factor = c( 19 ) ...)"
[1] "CURRENT both OBJECTIVE FUNCTION = 13.3834672437266"
[1] "BEST method = 'both' PATH MEMBER = c( 19 )"
[1] "BEST both OBJECTIVE FUNCTION = 13.3834672437266"
Package: healthyverse
[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 = 72.5270370117118"
[1] "BEST method = 'lin' PATH MEMBER = c( 3 )"
[1] "BEST lin OBJECTIVE FUNCTION = 72.5270370117118"
[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 = 40.715170547556"
[1] "BEST method = 'nonlin' PATH MEMBER = c( 3 )"
[1] "BEST nonlin OBJECTIVE FUNCTION = 40.715170547556"
[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 = 30.1735408390841"
[1] "BEST method = 'both' PATH MEMBER = c( 3 )"
[1] "BEST both OBJECTIVE FUNCTION = 30.1735408390841"
Package: RandomWalker
[1] "CURRNET METHOD: lin"
[1] "COPY LATEST PARAMETERS DIRECTLY FOR NNS.ARMA() IF ERROR:"
[1] "NNS.ARMA(... method = 'lin' , seasonal.factor = c( 17 ) ...)"
[1] "CURRENT lin OBJECTIVE FUNCTION = 6.49616822458263"
[1] "BEST method = 'lin' PATH MEMBER = c( 17 )"
[1] "BEST lin OBJECTIVE FUNCTION = 6.49616822458263"
[1] "CURRNET METHOD: nonlin"
[1] "COPY LATEST PARAMETERS DIRECTLY FOR NNS.ARMA() IF ERROR:"
[1] "NNS.ARMA(... method = 'nonlin' , seasonal.factor = c( 17 ) ...)"
[1] "CURRENT nonlin OBJECTIVE FUNCTION = 7.3596880201861"
[1] "BEST method = 'nonlin' PATH MEMBER = c( 17 )"
[1] "BEST nonlin OBJECTIVE FUNCTION = 7.3596880201861"
[1] "CURRNET METHOD: both"
[1] "COPY LATEST PARAMETERS DIRECTLY FOR NNS.ARMA() IF ERROR:"
[1] "NNS.ARMA(... method = 'both' , seasonal.factor = c( 17 ) ...)"
[1] "CURRENT both OBJECTIVE FUNCTION = 7.53157588661384"
[1] "BEST method = 'both' PATH MEMBER = c( 17 )"
[1] "BEST both OBJECTIVE FUNCTION = 7.53157588661384"
Package: tidyAML
[1] "CURRNET METHOD: lin"
[1] "COPY LATEST PARAMETERS DIRECTLY FOR NNS.ARMA() IF ERROR:"
[1] "NNS.ARMA(... method = 'lin' , seasonal.factor = c( 14 ) ...)"
[1] "CURRENT lin OBJECTIVE FUNCTION = 10.8828299247299"
[1] "BEST method = 'lin' PATH MEMBER = c( 14 )"
[1] "BEST lin OBJECTIVE FUNCTION = 10.8828299247299"
[1] "CURRNET METHOD: nonlin"
[1] "COPY LATEST PARAMETERS DIRECTLY FOR NNS.ARMA() IF ERROR:"
[1] "NNS.ARMA(... method = 'nonlin' , seasonal.factor = c( 14 ) ...)"
[1] "CURRENT nonlin OBJECTIVE FUNCTION = 21.7415509624376"
[1] "BEST method = 'nonlin' PATH MEMBER = c( 14 )"
[1] "BEST nonlin OBJECTIVE FUNCTION = 21.7415509624376"
[1] "CURRNET METHOD: both"
[1] "COPY LATEST PARAMETERS DIRECTLY FOR NNS.ARMA() IF ERROR:"
[1] "NNS.ARMA(... method = 'both' , seasonal.factor = c( 14 ) ...)"
[1] "CURRENT both OBJECTIVE FUNCTION = 17.0455268231883"
[1] "BEST method = 'both' PATH MEMBER = c( 14 )"
[1] "BEST both OBJECTIVE FUNCTION = 17.0455268231883"
Package: TidyDensity
[1] "CURRNET METHOD: lin"
[1] "COPY LATEST PARAMETERS DIRECTLY FOR NNS.ARMA() IF ERROR:"
[1] "NNS.ARMA(... method = 'lin' , seasonal.factor = c( 12 ) ...)"
[1] "CURRENT lin OBJECTIVE FUNCTION = 27.30996456205"
[1] "BEST method = 'lin' PATH MEMBER = c( 12 )"
[1] "BEST lin OBJECTIVE FUNCTION = 27.30996456205"
[1] "CURRNET METHOD: nonlin"
[1] "COPY LATEST PARAMETERS DIRECTLY FOR NNS.ARMA() IF ERROR:"
[1] "NNS.ARMA(... method = 'nonlin' , seasonal.factor = c( 12 ) ...)"
[1] "CURRENT nonlin OBJECTIVE FUNCTION = 4.19615487231713"
[1] "BEST method = 'nonlin' PATH MEMBER = c( 12 )"
[1] "BEST nonlin OBJECTIVE FUNCTION = 4.19615487231713"
[1] "CURRNET METHOD: both"
[1] "COPY LATEST PARAMETERS DIRECTLY FOR NNS.ARMA() IF ERROR:"
[1] "NNS.ARMA(... method = 'both' , seasonal.factor = c( 12 ) ...)"
[1] "CURRENT both OBJECTIVE FUNCTION = 6.63176462278477"
[1] "BEST method = 'both' PATH MEMBER = c( 12 )"
[1] "BEST both OBJECTIVE FUNCTION = 6.63176462278477"
[[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 [1,938 × 50]> <tibble [28 × 50]> <split [1910|28]>
2 healthyR <tibble [1,932 × 50]> <tibble [28 × 50]> <split [1904|28]>
3 healthyR.ts <tibble [1,868 × 50]> <tibble [28 × 50]> <split [1840|28]>
4 healthyverse <tibble [1,814 × 50]> <tibble [28 × 50]> <split [1786|28]>
5 healthyR.ai <tibble [1,674 × 50]> <tibble [28 × 50]> <split [1646|28]>
6 TidyDensity <tibble [1,525 × 50]> <tibble [28 × 50]> <split [1497|28]>
7 tidyAML <tibble [1,131 × 50]> <tibble [28 × 50]> <split [1103|28]>
8 RandomWalker <tibble [555 × 50]> <tibble [28 × 50]> <split [527|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 | 0.8242392 | 121.30399 | 0.7044495 | 174.33072 | 0.9988770 | 0.0032335 |
| healthyR.data | 2 | LM | Test | 0.7825210 | 148.37654 | 0.6687944 | 140.18645 | 1.0175742 | 0.0003613 |
| healthyR.data | 3 | EARTH | Test | 0.8310477 | 175.73553 | 0.7102685 | 132.35598 | 1.0863959 | 0.1921224 |
| healthyR.data | 4 | NNAR | Test | 0.7991947 | 179.82609 | 0.6830448 | 140.17863 | 1.0204628 | 0.0026202 |
| healthyR | 1 | ARIMA | Test | 0.8302436 | 1387.80159 | 0.7973388 | 135.36780 | 1.1398173 | 0.0061143 |
| healthyR | 2 | LM | Test | 0.8915471 | 536.45841 | 0.8562126 | 152.07271 | 1.1834100 | 0.0561822 |
| healthyR | 3 | EARTH | Test | 0.7960861 | 2243.63534 | 0.7645350 | 122.35384 | 1.1093113 | 0.1603213 |
| healthyR | 4 | NNAR | Test | 0.8396737 | 948.22791 | 0.8063951 | 139.17705 | 1.1826287 | 0.0068016 |
| healthyR.ts | 1 | ARIMA | Test | 0.5017220 | 284.13160 | 0.5420679 | 126.06488 | 0.7363862 | 0.1781458 |
| healthyR.ts | 2 | LM | Test | 0.6534916 | 422.93963 | 0.7060420 | 149.62237 | 0.8375889 | 0.0196666 |
| healthyR.ts | 3 | EARTH | Test | 0.7039487 | 338.47427 | 0.7605566 | 145.21175 | 0.8974595 | 0.0757851 |
| healthyR.ts | 4 | NNAR | Test | 0.6798702 | 366.93462 | 0.7345418 | 146.74808 | 0.8631650 | 0.0111622 |
| healthyverse | 1 | ARIMA | Test | 0.7509300 | 156.73037 | 0.7956564 | 69.22935 | 0.9180230 | 0.1020942 |
| healthyverse | 2 | LM | Test | 1.2788147 | 269.53868 | 1.3549827 | 158.50920 | 1.4206565 | 0.0000088 |
| healthyverse | 3 | EARTH | Test | 0.8323226 | 364.91823 | 0.8818969 | 61.94034 | 1.0362736 | 0.0360843 |
| healthyverse | 4 | NNAR | Test | 1.0951601 | 300.17317 | 1.1603893 | 136.17314 | 1.2417999 | 0.0038155 |
| healthyR.ai | 1 | ARIMA | Test | 0.4838008 | 191.85899 | 0.6408506 | 103.28068 | 0.7780767 | 0.0591838 |
| healthyR.ai | 2 | LM | Test | 0.5139797 | 163.51725 | 0.6808262 | 122.34592 | 0.7644093 | 0.0651694 |
| healthyR.ai | 3 | EARTH | Test | 0.5466512 | 116.88490 | 0.7241033 | 163.61297 | 0.7510209 | 0.0898351 |
| healthyR.ai | 4 | NNAR | Test | 0.4579734 | 148.30895 | 0.6066393 | 99.20578 | 0.7359135 | 0.0822008 |
| TidyDensity | 1 | ARIMA | Test | 1.2365426 | 146.77995 | 0.6910491 | 149.66380 | 1.3205353 | 0.0099386 |
| TidyDensity | 2 | LM | Test | 1.2155897 | 175.45443 | 0.6793394 | 149.43935 | 1.3075702 | 0.0349318 |
| TidyDensity | 3 | EARTH | Test | 1.2571415 | 169.35012 | 0.7025609 | 141.87097 | 1.3471470 | 0.0281517 |
| TidyDensity | 4 | NNAR | Test | 1.1120456 | 130.96852 | 0.6214732 | 149.50056 | 1.2537058 | 0.0641856 |
| tidyAML | 1 | ARIMA | Test | 0.8386531 | 150.01323 | 0.6811630 | 153.63869 | 1.1068609 | 0.0003378 |
| tidyAML | 2 | LM | Test | 0.8503077 | 247.42513 | 0.6906291 | 156.54244 | 1.0451118 | 0.1321412 |
| tidyAML | 3 | EARTH | Test | 0.9744772 | 241.96285 | 0.7914808 | 148.57084 | 1.2318668 | 0.0986567 |
| tidyAML | 4 | NNAR | Test | 0.8227772 | 237.27805 | 0.6682685 | 137.42697 | 1.0121226 | 0.2015502 |
| RandomWalker | 1 | ARIMA | Test | 0.8423489 | 87.00537 | 0.5085867 | 140.02538 | 0.9685627 | 0.5038078 |
| RandomWalker | 2 | LM | Test | 1.0508404 | 103.68502 | 0.6344681 | 146.24341 | 1.2774097 | 0.0099320 |
| RandomWalker | 3 | EARTH | Test | 1.0189601 | 95.21790 | 0.6152196 | 173.34315 | 1.1879625 | 0.0103766 |
| RandomWalker | 4 | NNAR | Test | 1.0324474 | 113.23750 | 0.6233629 | 142.13449 | 1.2265583 | 0.0040233 |
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.d… 1 ARIMA Test 0.824 121. 0.704 174. 0.999 0.00323
2 healthyR 3 EARTH Test 0.796 2244. 0.765 122. 1.11 0.160
3 healthyR.ts 1 ARIMA Test 0.502 284. 0.542 126. 0.736 0.178
4 healthyver… 1 ARIMA Test 0.751 157. 0.796 69.2 0.918 0.102
5 healthyR.ai 4 NNAR Test 0.458 148. 0.607 99.2 0.736 0.0822
6 TidyDensity 4 NNAR Test 1.11 131. 0.621 150. 1.25 0.0642
7 tidyAML 4 NNAR Test 0.823 237. 0.668 137. 1.01 0.202
8 RandomWalk… 1 ARIMA Test 0.842 87.0 0.509 140. 0.969 0.504
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 [1910|28]> <mdl_tm_t [1 × 5]>
2 healthyR <tibble> <tibble> <split [1904|28]> <mdl_tm_t [1 × 5]>
3 healthyR.ts <tibble> <tibble> <split [1840|28]> <mdl_tm_t [1 × 5]>
4 healthyverse <tibble> <tibble> <split [1786|28]> <mdl_tm_t [1 × 5]>
5 healthyR.ai <tibble> <tibble> <split [1646|28]> <mdl_tm_t [1 × 5]>
6 TidyDensity <tibble> <tibble> <split [1497|28]> <mdl_tm_t [1 × 5]>
7 tidyAML <tibble> <tibble> <split [1103|28]> <mdl_tm_t [1 × 5]>
8 RandomWalker <tibble> <tibble> <split [527|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")
