Forecasting & Time Series Plotting at Scale

modeltime_trelliscope.R

# BUSINESS SCIENCE 
# NEW TRELLISCOPE INTEGRATION ----
# * AVAILABLE IN TIMETK 2.8.0 AND MODELTIME 1.2.0

# LIBRARIES ----
library(tidymodels)
library(modeltime)
library(tidyverse)
library(timetk)

# MULTI-TIME SERIES DATASET ----
data_tbl <- walmart_sales_weekly %>%
    select(id, Date, Weekly_Sales) %>%
    set_names(c("id", "date", "value"))

# TRELLISCOPE ----

data_tbl %>%
    group_by(id) %>%
    plot_time_series(
        date, value, 
        .facet_ncol         = 2,
        .facet_nrow         = 2,
        .facet_strip_remove = TRUE, 
        .title              = NULL, 
        .interactive        = TRUE,
        .trelliscope        = TRUE
    )

# TRAIN / TEST SPLIT ----
splits <- data_tbl %>% 
    time_series_split(
        assess     = "3 months", 
        cumulative = TRUE
    )

# PREPROCESS ----
rec_obj <- recipe(value ~ ., training(splits)) %>%
    step_mutate_at(id, fn = droplevels) %>%
    step_timeseries_signature(date) %>%
    step_rm(date) %>%
    step_zv(all_predictors()) %>%
    step_dummy(all_nominal_predictors(), one_hot = TRUE)

# MODEL ----
wflw_xgb <- workflow() %>%
    add_model(
        boost_tree() %>% set_engine("xgboost")
    ) %>%
    add_recipe(rec_obj) %>%
    fit(training(splits))

# ORGANIZE ----
model_tbl <- modeltime_table(
    wflw_xgb
)

# CALIBRATE ON TESTING SET ----
calib_tbl <- model_tbl %>%
    modeltime_calibrate(
        new_data = testing(splits), 
        id       = "id"
    )

# VISUALIZE WITH NEW TRELLISCOPE ----
calib_tbl %>%
    modeltime_forecast(
        new_data    = testing(splits),
        actual_data = data_tbl,
        conf_by_id  = TRUE
    ) %>%
    group_by(id) %>%
    plot_modeltime_forecast(
        .facet_ncol         = 2,
        .facet_nrow         = 2,
        .facet_strip_remove = TRUE,
        .legend_show        = FALSE,  
        .title              = NULL, 
        .interactive        = TRUE,
        .trelliscope        = TRUE
    )

You should get 2 plots.

Plot 1: Time Series Trelliscope

Plot 2: Forecast Trelliscope

New Functionality.

I've just added a new parameter trelliscope_params that exposes the facet_trelliscope() parameters to all of the plotting functions.

Here's an example where I change the width of the trelliscopes.

library(timetk)
library(tidyverse)

walmart_sales_weekly %>%
    select(id, Dept, Date, Weekly_Sales) %>%
    set_names(c("id", "dept", "date", "value")) %>%
    group_by(id, dept) %>%
    
    plot_time_series(
        date, value, 
        .trelliscope        = TRUE, 
        .facet_ncol         = 2, 
        .facet_nrow         = 3,
        .facet_strip_remove = TRUE, 
        .title              = NULL,
        # Access facet_trelliscope params
        .trelliscope_params = list(
            width = 1500
        )
    )