R/prepare-numeric.R

In mladenjovanovic/athletemonitoring: Athlete Monitoring Data Analysis and Visualization

prepare_numeric <- function(data,
                            athlete,
                            date,
                            variable,
                            value,
                            day_aggregate,
                            NA_session,
                            NA_day,
                            acute,
                            chronic,
                            partial,
                            rolling_fill,
                            rolling_estimators,
                            posthoc_estimators,
                            group_summary_estimators,
                            extend,
                            extend_fill,
                            start_date,
                            stop_date,
                            iter) {

  # +++++++++++++++++++++++++++++++++++++++++++
  # Code chunk for dealing with R CMD check note
  entries <- NULL
  missing_day <- NULL
  missing_entries <- NULL
  extended_day <- NULL
  proportion <- NULL
  estimator <- NULL
  acute.missing_entries <- NULL
  chronic.missing_entries <- NULL
  acute.missing_day <- NULL
  chronic.extended_day <- NULL
  acute.extended_day <- NULL
  chronic.missing_day <- NULL
  # +++++++++++++++++++++++++++++++++++++++++++

  if (iter) {
    message("Preparing data...")
  }

  # Extract the data
  data <- data.frame(
    athlete = data[[athlete]],
    date = data[[date]],
    variable = data[[variable]],
    value = data[[value]]
  )

  # ------------------------------------
  data <- data %>%
    dplyr::group_by(athlete, date, variable) %>%
    # Fill in missing entry
    dplyr::mutate(
      missing_entries = is.na(value),
      value = ifelse(missing_entries, NA_session, value)
    ) %>%
    # Aggregate to day value
    dplyr::summarise(
      entries = dplyr::n(),
      missing_entries = sum(missing_entries),
      value = day_aggregate(value)
    ) %>%
    # Get start and stop dates for every athlete and variable
    dplyr::group_by(athlete, variable) %>%
    dplyr::mutate(
      start_date = min(date),
      stop_date = max(date)
    )

  # Get the overall start and stop date to impute missing days
  if (is.null(start_date)) {
    overall_start_date <- min(data$date)
  } else {
    overall_start_date <- start_date
  }

  if (is.null(stop_date)) {
    overall_stop_date <- max(data$date)
  } else {
    overall_stop_date <- stop_date
  }

  # Create a sequence of day
  if (is.numeric(data$date)) {
    date_seq <- seq(overall_start_date, overall_stop_date)
  } else {
    date_seq <- seq(overall_start_date, overall_stop_date, "days")
  }

  # Generate a continuous df so that there are no missing days for every athlete
  data <- dplyr::left_join(
    expand.grid(
      athlete = unique(data$athlete),
      date = date_seq,
      variable = unique(data$variable)
    ),
    data,
    by = c("athlete", "date", "variable")
  )

  # Fill missing_entries that has NA values now
  data$missing_entries <- ifelse(
    is.na(data$missing_entries),
    0,
    data$missing_entries
  )

  # Rolling function
  roll_func <- function(value, date) {
    # Acute
    acute_df <- data.frame(
      zoo::rollapply(
        value,
        FUN = rolling_estimators,
        width = acute,
        partial = partial,
        fill = rolling_fill,
        align = "right"
      )
    )


    # Check is amount of data in 'value' is smaller than rolling window
    # Since zoo::rollapply will not return names estimators
    # ++++++++++++++++++++++++++++++++++++++++++++++++++++
    if (length(value) < acute & partial == FALSE) {
      acute_df <- rolling_estimators(rep(rolling_fill, length(value)))
      acute_df <- as.data.frame(t(acute_df))
    }

    # When there is only one estimator, the zoo::rollapply screws up
    # the name
    colnames(acute_df) <- paste0("acute.", names(rolling_estimators(1:100)))

    # Chronic
    chronic_df <- data.frame(
      zoo::rollapply(
        value,
        FUN = rolling_estimators,
        width = chronic,
        partial = partial,
        fill = rolling_fill,
        align = "right"
      )
    )

    # Check is amount of data in 'value' is smaller than rolling window
    # Since zoo::rollapply will not return names estimators
    # ++++++++++++++++++++++++++++++++++++++++++++++++++++
    if (length(value) < chronic & partial == FALSE) {
      chronic_df <- rolling_estimators(rep(rolling_fill, length(value)))
      chronic_df <- as.data.frame(t(chronic_df))
    }

    # When there is only one estimator, the zoo::rollapply screws up
    # the name
    colnames(chronic_df) <- paste0("chronic.", names(rolling_estimators(1:100)))

    # merge together
    data.frame(date = date, variable.value = value, acute_df, chronic_df)
  }

  if (iter) {
    message("Rolling...")
  }

  data <- data %>%
    # fill in the individual/variable start and stop days and remove excess
    dplyr::group_by(athlete, variable) %>%
    # Arrange/Sort
    dplyr::arrange(date) %>%
    # Tag missing day
    dplyr::mutate(missing_day = is.na(start_date)) %>%
    tidyr::fill(start_date, stop_date, .direction = "updown")

  # Extend features
  if (extend == "none") {
    data <- data %>%
      dplyr::filter(date >= start_date & date <= stop_date)
  } else if (extend == "start") {
    data <- data %>%
      dplyr::group_by(athlete) %>%
      dplyr::filter(date <= my.max(stop_date))
  } else if (extend == "end") {
    data <- data %>%
      dplyr::group_by(athlete) %>%
      dplyr::filter(date >= my.min(start_date))
  }

  data <- data %>%
    dplyr::group_by(athlete, variable) %>%
    # Mark extended days
    dplyr::mutate(
      extended_day = ifelse(date >= start_date & date <= stop_date, FALSE, TRUE),
      extended_day = ifelse(is.na(extended_day), TRUE, extended_day),
      value = ifelse(extended_day, extend_fill, value)
    ) %>%
    dplyr::select(-start_date, -stop_date) %>%
    # Fill in missing days
    dplyr::mutate(value = ifelse(missing_day & !extended_day, NA_day, value)) %>%
    # Count entries
    dplyr::mutate(entries = ifelse(is.na(entries), 0, entries)) %>%
    # Generate rolling estimators
    dplyr::mutate(
      roll_func(value, date)
    ) %>%
    dplyr::ungroup() %>%
    dplyr::select(-value) %>%
    dplyr::relocate(athlete, date, variable, entries, missing_entries, missing_day, extended_day) %>%
    dplyr::arrange(athlete, date, variable) #%>%
    #dplyr::mutate(missing_day = as.numeric(missing_day))

  # Apply post-hoc estimators
  data <- posthoc_estimators(data)

  # Create long  version
  data_long <- tidyr::pivot_longer(
    data,
    cols = -(1:7),
    names_to = "estimator",
    values_to = "value"
  )

  # ===================================
  # Group summaries

  if (iter) {
    message("Group summaries...")
  }

  # Group summary aggregator
  group_func <- function(x) {
    group_df <- data.frame(t(group_summary_estimators(x)))

    colnames(group_df) <- paste0("group.", colnames(group_df))
    return(group_df)
  }

  group_summary <- data_long %>%
    dplyr::group_by(date, variable, estimator) %>%
    dplyr::summarise(
      group_func(value)
    ) %>%
    dplyr::ungroup()

  # Missing data aggregator
  if (iter) {
    message("Missing data summaries...")
  }

  missing_summary_athlete <- data %>%
    dplyr::select(athlete, date, variable, missing_entries, missing_day, extended_day) %>%
    dplyr::group_by(athlete, variable) %>%
    dplyr::mutate(
      acute.missing_entries = zoo::rollapply(
        missing_entries,
        FUN = sum,
        width = acute,
        partial = partial,
        fill = rolling_fill,
        align = "right"
      ),
      chronic.missing_entries = zoo::rollapply(
        missing_entries,
        FUN = sum,
        width = chronic,
        partial = partial,
        fill = rolling_fill,
        align = "right"
      ),
      acute.missing_day = zoo::rollapply(
        missing_day,
        FUN = sum,
        width = acute,
        partial = partial,
        fill = rolling_fill,
        align = "right"
      ),
      chronic.missing_day = zoo::rollapply(
        missing_day,
        FUN = sum,
        width = chronic,
        partial = partial,
        fill = rolling_fill,
        align = "right"
      ),
      acute.extended_day = zoo::rollapply(
        extended_day,
        FUN = sum,
        width = acute,
        partial = partial,
        fill = rolling_fill,
        align = "right"
      ),
      chronic.extended_day = zoo::rollapply(
        extended_day,
        FUN = sum,
        width = chronic,
        partial = partial,
        fill = rolling_fill,
        align = "right"
      )
    ) %>%
    dplyr::ungroup()

  missing_summary_group <- missing_summary_athlete %>%
    dplyr::group_by(date, variable) %>%
    dplyr::summarise(
      missing_entries = sum(missing_entries),
      missing_day = sum(missing_day),
      extended_day = sum(extended_day),
      acute.missing_entries = sum(acute.missing_entries),
      chronic.missing_entries = sum(chronic.missing_entries),
      acute.missing_day = sum(acute.missing_day),
      chronic.missing_day = sum(chronic.missing_day),
      acute.extended_day = sum(acute.extended_day),
      chronic.extended_day = sum(chronic.extended_day)
    ) %>%
    dplyr::ungroup()

  if (iter) {
    message("Done!")
  }

  return(
    new_athletemonitoring(
      type = "numeric",
      data_wide = data,
      data_long = data_long,
      group_summary = group_summary,
      missing_summary_athlete = missing_summary_athlete,
      missing_summary_group = missing_summary_group,
      day_aggregate = day_aggregate,
      NA_session = NA_session,
      NA_day = NA_day,
      acute = acute,
      chronic = chronic,
      partial = partial,
      rolling_fill = rolling_fill,
      rolling_estimators = rolling_estimators,
      posthoc_estimators = posthoc_estimators,
      group_summary_estimators = group_summary_estimators,
      extend = extend,
      extend_fill = extend_fill
    )
  )
}