In gouthaman87/av-gt-food_demand_forecast: Food demand forecast

knitr::opts_chunk$set(echo = TRUE)
targets::tar_make(script = here::here("pipelines", "_targets.R"))

r_functions <- list.files(path = "R", full.names = TRUE)
sapply(r_functions, source)

# targets::tar_load("data_list")
# targets::tar_load("complete_data")
# targets::tar_load("test_results")
# 
# df_rec <- test_results |> 
#   recipes::prep() |> 
#   recipes::juice()
# 
# test_results

This R Markdown document is create to check Food Demand Forecast.

Read Data & Transform

targets::tar_load(c("complete_data", "test_results"))

master_data <- complete_data[["master_data"]]
modeltime_table <- test_results[["modeltime_table"]]
vip_dt <- test_results[["variable_importance"]]
test_results <- test_results[["test_results"]]

statistics_fcast <- 
  list.files(
  "./inst/extdata/statistics",
  pattern = "_tbl.csv",
  full.names = TRUE
  ) |>
  purrr::map_dfr(~readr::read_csv(.x, show_col_types = FALSE))

rm(complete_data)
gc()

master_data <- 
  master_data |>
  dplyr::mutate(year = lubridate::year(date), week = lubridate::week(date))

statistics_fcast <- 
  statistics_fcast |>
  dplyr::mutate(
    year = lubridate::year(week_date),
    week = lubridate::week(week_date)
  ) |>

  # Join Actual for Statistics Forecast
  dplyr::left_join(
    master_data |>
      dplyr::select(week, year, meal_id, center_id, num_orders, date),
    by = c("week", "year", "meal_id", "center_id")
  ) |>
  dplyr::filter(!is.na(fc_qty)) |>

  dplyr::transmute(
    date,
    week,
    year,
    meal_id,
    center_id,
    .model_desc = .model,
    .value = fc_qty,
    num_orders
  )

statistics_accuracy_tbl <- 
  statistics_fcast |>
  accuracy_metric() |>
  dplyr::group_by(meal_id, center_id) |>
  dplyr::filter(accuracy_rmsle == min(accuracy_rmsle)) |>
  dplyr::select(-accuracy_rmsle)

test_results <- 
  statistics_fcast |>
  dplyr::inner_join(statistics_accuracy_tbl, by = colnames(statistics_accuracy_tbl)) |>
  dplyr::mutate(.model_desc = "Statistics") |>
  dplyr::bind_rows(test_results) |>
  dplyr::mutate(ts_id = paste0(center_id, "_", meal_id))

report_dt <- create_report(test_results)

accuracy_tbl <- 
  test_results |>
  accuracy_metric() |>
  dplyr::mutate(ts_id = paste0(center_id, "_", meal_id)) |>
  dplyr::inner_join(report_dt[["common_item"]], "ts_id") |>
  dplyr::mutate(
    .model_desc = stringr::str_replace(.model_desc, "RECIPE_", ""),
     accuracy_rmsle = round(accuracy_rmsle, 2)
  )

model_accuracy <- 
  accuracy_tbl |>
  dplyr::left_join(
    master_data |>
      dplyr::distinct(center_id, meal_id, base_price),
    by = c("center_id", "meal_id")
  ) |>
  dplyr::with_groups(
    .model_desc,
    dplyr::summarise,
    WRMSLE = round(weighted.mean(accuracy_rmsle, base_price), 2)
  )

master_data <- 
  master_data |>
  dplyr::mutate(
    .model_desc = "Actual",
    .value = num_orders,
    ts_id = paste0(center_id, "_", meal_id)
  ) |>
  dplyr::select(dplyr::all_of(colnames(test_results))) |>
  dplyr::bind_rows(test_results)

Report:

shiny::renderText(glue::glue("Total Number of Items: {report_dt[['N']]}"))
shiny::renderText(glue::glue("Statistics Items: {report_dt[['S']]}"))
shiny::renderText(glue::glue("ML Items: {report_dt[['M']]}"))
shiny::renderText(glue::glue("Common Items: {nrow(report_dt[['common_item']])}"))

Accuracy Table

DT::renderDataTable(
  accuracy_tbl |>
    tidyr::pivot_wider(
      dplyr::ends_with("_id"),
      values_from = accuracy_rmsle,
      names_from = .model_desc
    ) |>
    DT::datatable(
      class = 'cell-border stripe',
      rownames = FALSE
    )
)

TS Plot

items <- report_dt[['common_item']]$ts_id

shiny::selectInput("ts_id", label = "TS ID", choices = items)

plotly::renderPlotly({
  master_data |>
    dplyr::filter(ts_id == input$ts_id) |>
    # dplyr::filter(center_id == input$center_id, meal_id == input$meal_id) |>
    timetk::plot_time_series(
      .date_var = date,
      .value = .value,
      .smooth = FALSE,
      .color_var = .model_desc
    )
})

WRMSLE: Model Accuracy

DT::renderDataTable(
  model_accuracy |>
    DT::datatable(
      class = 'cell-border stripe',
      rownames = FALSE
    )
)

Variable Importance

shiny::selectInput("mod", label = NULL, choices = unique(vip_dt$Model))

shiny::renderPlot({
  vip_dt |>
    dplyr::filter(Model == input$mod) |>
    dplyr::mutate(Variable = forcats::fct_reorder(Variable, Importance)) |>

    ggplot2::ggplot(ggplot2::aes(Variable, Importance)) +
    ggplot2::geom_col() +
    ggplot2::coord_flip() +
    ggplot2::theme_minimal()
}, res = 96)