ewsnet_predict: EWSNet Predict

In EWSmethods: Forecasting Tipping Points at the Community Level

EWSNet Predict

Description

Communicates with EWSNet (https://ewsnet.github.io), a deep learning framework for modelling and anticipating regime shifts in dynamical systems, and returns the model's prediction for the inputted univariate time series.

Usage

ewsnet_predict(
  x,
  scaling = TRUE,
  ensemble = 25,
  envname,
  weights_path = default_weights_path()
)

Arguments

x	A numeric vector of values to be tested.
scaling	Boolean. If TRUE, the time series will be scaled between 1 and 2 and scaled EWSNet model weights will be used. This is the recommended setting.
ensemble	A numeric value stating the number of models to average over. Options range from 1 to 25.
envname	A string naming the Python environment prepared by ewsnet_init().
weights_path	A string naming the path to model weights installed by ewsnet_reset().

Value

A dataframe of EWSNet predictions. Values represent the estimated probability that the quoted event will occur.

Examples

#A dummy dataset of a hedgerow bird population
#monitored over 50 years.

abundance_data <- data.frame(time = seq(1:50),
 abundance = rnorm(50,mean = 20))

#Activate python environment (only necessary
#on first opening of R session).

## Not run: 
ewsnet_init(envname = "EWSNET_env")

## End(Not run)

#Generate EWSNet predictions.

## Not run: 
pred <- ewsnet_predict(
 abundance_data$abundance,
 scaling = TRUE,
 ensemble = 15,
 envname = "EWSNET_env")
 
## End(Not run)

EWSmethods documentation built on May 29, 2024, 5:41 a.m.