In ahaeusser/echos: Echo State Networks for Time Series Modeling and Forecasting

knitr::opts_chunk$set(
  collapse = TRUE,
  comment = "#>",
  fig.height = 5,
  fig.width = 7
)

Load package

library(echos)
library(tidyverse)
library(tsibble)
library(fable)

Prepare dataset

In this example, we will use the m4_data. The dataset is a monthly tsibble, which is filtered to include only the time series "M21655" and "M2717". The resulting object train_frame contains the training data and is visualized below.

train_frame <- m4_data %>%
  filter(series %in% c("M21655", "M2717"))

train_frame

p <- ggplot()

p <- p + geom_line(
  data = train_frame,
  aes(
    x = index,
    y = value),
  linewidth = 0.5
)

p <- p + facet_wrap(
  vars(series),
  ncol = 1,
  scales = "free")

p

Train ESN model

The function ESN() is used in combination with model() from the fabletools package to train an Echo State Network for the variable value. The trained models are stored as a mable (i.e., model table). Additionally, an ARIMA() model is trained as benchmark.

mable_frame <- train_frame %>%
  model(
    "ESN" = ESN(value),
    "ARIMA" = ARIMA(value)
    )

mable_frame

Forecast ESN model

Forecasts are generated via the function forecast(), where the forecast horizon is set to h = 18 (i.e., 18-month ahead forecasts). The forecasts are stored as fable (i.e., forecast table) and visualized along the historic training data.

fable_frame <- mable_frame %>%
  forecast(h = 18)

fable_frame

fable_frame %>%
  autoplot(train_frame, level = NULL)

ahaeusser/echos documentation built on June 2, 2025, 2:17 a.m.