nixtlar.R

# nixtlar

#install.packages("nixtlar")
packageVersion("nixtlar") # should return v0.6.2 at least 

library(nixtlar)
library(data.table)
library(tidyverse)
library(lubridate)
library(tictoc)

# 1. Load data ----
df <- fread("data/demo_data.csv")
head(df)

## Data requirements ----
# ds: A timestamp in format YYYY-MM-DD or YYYY-MM-DD hh:mm:ss, either as characters or date objects 
# y: The numerical target variable 
# unique_id: A unique identifier 
df$ds <- ymd(df$ds)

## Separate train and tests sets 
train_df <- df |> 
  filter(ds <= "2017-08-01")

length(unique(train_df$unique_id)) # total number of series 

test_df <- df |> 
  filter(ds > "2017-08-01") 

## Visualize data ----
nixtla_client_plot(train_df)

# 2. Set up your API key ----
# Get yours here: dashboard.nixtla.io

nixtla_client_setup(
  api_key = "Your API key here"
)

options(NIXTLA_API_KEY="Your API key here")

library(usethis)
usethis::edit_r_environ() # Set your API key as NIXTLA_API_KEY='Your API key here'

nixtla_validate_api_key()

# 3. Start forecasting! ----
tic()
fc <- nixtla_client_forecast(train_df, h=14)
toc()

write.table(fc, "output/timegpt.csv" , sep = ",", row.names = FALSE, quote = FALSE)

## Plot results ----
nixtla_client_plot(train_df, fc, max_insample_length = 200)

## Use the long horizon model ---- 
fc_long <- nixtla_client_forecast(train_df, h=14, model = "timegpt-1-long-horizon")

write.table(fc_long, "output/timegpt_longh.csv" , sep = ",", row.names = FALSE, quote = FALSE)

nixtla_client_plot(train_df, fc, max_insample_length = 200)

# 4. Compute accuracy ----
## Here we will use MAE and MASE

# 5. Additional features 
fc <- nixtla_client_forecast(train_df, h = 7, level = c(80,95))

nixtla_client_plot(train_df, fc, max_insample_length = 200)

# 6. Anomaly detection ---- 
anomalies <- nixtla_client_detect_anomalies(train_df, level = c(95))
head(anomalies) 

anomalies |> 
  filter(anomaly == TRUE)

nixtla_client_plot(train_df, anomalies, plot_anomalies = TRUE, max_insample_length = 100)

# 7. Cross validation ---- 
cv <- nixtla_client_cross_validation(train_df, h = 7, n_windows = 3, step_size = 7)
head(cv)

nixtla_client_plot(df, cv, max_insample_length = 200)