tune_keras_rnn_eval: Evaluate (tuned) recurrent neural networks per...

In thfuchs/tsRNN: Time Series Forecasting using Recurrent Neural Networks by Keras

Description

Evaluate (tuned) recurrent neural networks per cross-validation split/sample for time series data

Usage

tune_keras_rnn_eval(
  fc_sample,
  cv_setting,
  bayes_best_par,
  col_id = NULL,
  col_date = "index",
  col_value = "value",
  h = NULL,
  frequency = 4,
  level = 95
)

Arguments

fc_sample	forecasts per rsample split
cv_setting	list of "periods_train", "periods_val", "periods_test" and "skip_span" for rolling_origin
bayes_best_par	tuned hyperparameters, from tune_keras_rnn_bayesoptim()
col_id	Optional ID column in data, default to "ticker"
col_date	Date column in data, default to "index"
col_value	Value column in data, default to "value"
h	NULL if forecast horizon equals cv_setting$n_test, else named list of forecast horizons for accuracy measures
frequency	time series frequency, e.g. 4 for quarters and 12 for months
level	level for prediction interval in percentage

Value

list of time series accuracy measures (MAPE, sMAPE, MASE, sMIS, ACD) optionally for different forecast horizons (if h specified)

See Also

Other RNN tuning with Keras: tune_keras_rnn_bayesoptim(), tune_keras_rnn_predict()

Examples

result_predict <- readRDS(system.file(
"tinytest_data/test_tune_keras_rnn_predict_simple.rds", package = "tsRNN"
))
cv_setting <- list(
  periods_train = 90,
  periods_val = 10,
  periods_test = 10,
  skip_span = 5
)
bayes_best_par <- purrr::map(
  readRDS(system.file("tinytest_data/apple_bayesoptim.rds", package = "tsRNN")),
  "Best_Par"
)

# Result
tune_keras_rnn_eval(
  fc_sample = result_predict,
  cv_setting = cv_setting,
  bayes_best_par = bayes_best_par
)

# Specify multiple forecast horizons
tune_keras_rnn_eval(
  fc_sample = result_predict,
  cv_setting = cv_setting,
  bayes_best_par = bayes_best_par,
  h = list(short = 1:2, long = 3:6)
)

thfuchs/tsRNN documentation built on April 17, 2021, 11:03 p.m.