ts_nn_preparation: Timeseries data preparation for neural network Keras models

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

Description

The following steps are proceeded:

• Split in train, validation and test split

• Transformation to 3D arrays for Keras

Usage

ts_nn_preparation(
  data,
  tsteps,
  length_val = 16L,
  length_test = 8L,
  value_col = "value",
  value_lag_col = "value_lag"
)

Arguments

data	balanced univariate time series as data.table object. Lagged columns in the form value_lagX required, where X represents number of lag
tsteps	number of lagged time steps
length_val	length of validation set
length_test	length of test set
value_col	name of column with value to prepare
value_lag_col	name of lagged value column(s) to prepare, searched by starting pattern. E.g. value_lag_col = "value_lag" will catch column "value_lag" and "value_lag1" but not "2_value_lag"

Value

list of "X" and "Y" each containing "train", "val" and "test" arrays

Examples

data <- tsRNN::DT_apple

data[, value_lag1 := data.table::shift(value, type = "lag", n = 1)]
data[, value_lag2 := data.table::shift(value, type = "lag", n = 2)]
data <- data[!is.na(get(paste0("value_lag2")))]

ts_nn_preparation(data, tsteps = 2L, length_val = 6L, length_test = 6L)

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