Assembles features for a single forecast step by combining target-based features (lags, MAs, rolling stats) with calendar and exogenous variable features.
Usage
.prepare_feature_row(
y_hist,
next_date,
target_col,
p = NULL,
q = NULL,
roll_windows = NULL,
roll_stats = c("sum", "sd", "min", "max"),
trend_windows = NULL,
trend_degrees = NULL,
CAL_row = NULL,
XF_row = NULL,
groups_chr = NULL,
date_col = "date"
)Arguments
- y_hist
Numeric vector of historical target values
- next_date
Date for this forecast step (used for calendar/xreg lookup)
- target_col
Character, name of target column
- p
Integer, number of lags (NULL if none)
- q
Integer vector of MA windows (NULL if none)
- roll_windows
Integer vector of rolling stat windows (NULL if none)
- roll_stats
Character vector of rolling statistics to compute
- trend_windows
Integer vector for trend slopes (NULL if none)
- trend_degrees
Integer vector for polynomial trends (NULL if none)
- CAL_row
Single-row tibble of calendar features for this date (NULL if none)
- XF_row
Single-row tibble of xreg features for this date (NULL if none)
- groups_chr
Character vector of group column names (NULL if ungrouped)
- date_col
Character, name of date column