In opoyc/autoforecast: Forecast Development Kit

library("autoforecast")
library("ggrepel")
library("tsibble")
library("dplyr")

Regression models' helpers

Get Time Weights

Regression estimations can be adjusted by increasing the importance of the observations. Particularly, in time series data it could happen that recent data is considered a better signal to predict future values.

# ap <- AirPassengers %>% 
#   as_tsibble() %>% 
#   as_tibble() %>% 
#   mutate(reg_name = "0", reg_value = 0, key = "airpassengers", index = as.Date(index)) %>% 
#   prescribe_ts(key = "key", y_var = "value", date_var = "index", reg_name = "reg_name", reg_value = "reg_value", freq = 12)

# ap %>% 
#   feature_engineering_ts() %>% 
#   autoforecast::fit_ts()

# c(0, .3, 0.91, 0.95, .99, .999, .9999, .99999, 1) %>% 
#   enframe() %>% 
#   mutate(time_weight = as.character(value)
#          , tw = map(value, ~get_time_weights(1:100, time_weight = .x) %>% 
#                       enframe)) %>%
#   select(time_weight, tw) %>% 
#   unnest(tw) %>% 
#   mutate(time_weight_label = ifelse(name == 75, time_weight, NA)) %>% 
#   ggplot(aes(name, value, col = time_weight))+
#   theme(legend.position = "none")+
#   geom_line()+
#   labs(x = "Time series index", y = "Observation weight")+
#   geom_text_repel(aes(label = time_weight_label))