flow_series_from_dataframe: Setup a time series data generator specification

In bagasbgy/kerasgenerator: Data generator for deep learning using Keras interface to R

Description

Setup a time series data generator specification

Usage

flow_series_from_dataframe(data, x, y, length_out, stride, lookback,
  timesteps, batch_size = 32, mode = "training")

## S3 method for class 'series_generator'
tidy_prediction(generator, prediction, ...)

Arguments

data	A tibble or data.frame.
x	A character vectors containing the column names for x variables
y	A character vectors containing the column names for y variables
length_out	An integer specifying length out size.
stride	An integer specifying stride size.
lookback	An integer specifying lookback period.
timesteps	An integer specifying timesteps length.
batch_size	An integer specifying batch size.
mode	A character vector of length one which specify the generator behaviour: either "training" or "prediction". See details for further information.
generator	A reference data generator generated from kerasgenerator package.
prediction	A matrix or an array containing the prediction results
...	Further argument for specific generators. See details.

Details

soon updated

Value

A function which could generate batches of x and/or y arrays

Examples

# import library
library(kerasgenerator)

# create example data
sunspots_df <- as.data.frame(sunspots)

# setup a generator
data_gen <- flow_series_from_dataframe(
  data = sunspots_df,
  x = "x",
  y = "x",
  length_out = 1,
  stride = 1,
  lookback = 1,
  timesteps = 12,
  batch_size = 32,
  mode = "training"
)

# quick check
arrays <- data_gen()

str(arrays[[1]])
str(arrays[[2]])

bagasbgy/kerasgenerator documentation built on Dec. 25, 2019, 8:52 p.m.