In business-science/timekit: A Tool Kit for Working with Time Series

knitr::opts_chunk$set(
    message = F, 
    warning = F,
    collapse = TRUE,
    comment = "#>",
    fig.path = "README-"
)

library(timetk)

timetk

A collection of tools for working with time series in R

Benefits

The timetk package enables a user to more easily work with time series objects in R. The package has tools for inspecting and manipulating the time-based index, expanding the time features for data mining and machine learning, and converting time-based objects to and from the many time series classes. The following are key benefits:

• Index extraction: get the time series index from any time series object.
• Understand time series: create a signature decomposition and summary from a time series index.
• Build future time series: create a future time series from an index.
• Coerce between time-based tibbles (tbl) and the major time series data types xts, zoo, zooreg, and ts: Simplifies coercion and maximizes time-based data retention during coercion to regularized time series (e.g. ts).

An example of the forecasting capabilities as shown in vignette TK03 - Forecasting Using a Time Series Signature with timetk.

knitr::include_graphics("README_1_bikes_forecast.png")

Tools

The package contains the following functions:

1. Get an index: tk_index returns the time series index of time series objects, models. The argument timetk_idx can be used to return a special timetk "index" attribute for regularized ts objects that returns a non-regularized date / date-time index if present.

2. Get critical timeseries information: tk_get_timeseries_signature and tk_get_timeseries_summary takes an index and provides a time series decomposition and key summary attributes of the index, respectively. The tk_augment_timeseries_signature expedites adding the time series decomposition to the time series object.

3. Make a future timeseries: tk_make_future_timeseries models a future time series after an existing time series index.

4. Coercion functions: tk_tbl, tk_ts, tk_xts, tk_zoo, and tk_zooreg coerce time-based tibbles tbl to and from each of the main time-series data types xts, zoo, zooreg, ts, maintaining the time-based index.

Getting started

Load libraries and start with some time series data

library(timetk)
library(lubridate)

Use the FB time series.

FB_tbl <- FANG %>%
    filter(symbol == "FB")
FB_tbl

Extract a time series index

Get the timeseries index.

idx <- tk_index(FB_tbl)
head(idx)

Expand the time series signature

Get the time series signature from the index, a tibble of decomposed features that are useful for data mining and machine learning.

tk_get_timeseries_signature(idx)

Get a summary of the time series

Get the time series summary from the index, a single-row tibble of key summary information from the time series.

# General summary
tk_get_timeseries_summary(idx)[1:6]

# Frequency summary
tk_get_timeseries_summary(idx)[6:12]

Make a future time series

Use an index to make a future time series.

holidays <- c("2017-01-02", "2017-01-16", "2017-02-20",
              "2017-04-14", "2017-05-29", "2017-07-04",
              "2017-09-04", "2017-11-23", "2017-12-25") %>%
    ymd()

idx_future <- tk_make_future_timeseries(
    idx, 
    n_future         = 366, 
    skip_values      = holidays, 
    inspect_weekdays = TRUE) 

head(idx_future)

tail(idx_future)

Coerce time series without specifying order.by or worrying about coercion issues

Coercion to xts, zoo, or ts is simplified. The data is ordered correctly automatically using the column containing the date or datetime information. Non-numeric columns are automatically dropped with a warning to the user (the silent = TRUE hides the warnings).

# xts
FB_xts <- tk_xts(FB_tbl, silent = TRUE)

# zoo
FB_zoo <- tk_zoo(FB_tbl, silent = TRUE)

# ts
FB_ts <- tk_ts(FB_tbl, start = 2013, freq = 252, silent = TRUE)

This covers the basics of the timetk package capabilities. Here's how to get started.

Installation

Download development version with latest features:

``` {r, eval = FALSE}

install.packages("devtools")

devtools::install_github("business-science/timetk")

_Or, download CRAN approved version_:

```r
install.packages("timetk")

Acknowledgements: Standing On Shoulders

A lot of innovative time series and forecasting work is going on that ultimately benefits the community. We'd like to thank the following people and packages that came before timetk in time series analysis and machine learning.

• maltese: Similar in respect to timetk in that it enables machine learning-friendly data frame generation exposing a number of critical features that can be used for forecasting.
• lubridate: Contains an excellent set of functions to extract components of the date and datetime index.
• xts and zoo: Fundamental packages for working with time series enabling creation of a time series index for ts class and calculating periodicity.

Further Information

The timetk package includes a vignette to help users get up to speed quickly:

• TK00 - Time Series Coercion Using timetk
• TK01 - Working with the Time Series Index using timetk
• TK02 - Making a Future Time Series Index using timetk
• TK03 - Forecasting Using a Time Series Signature with timetk

business-science/timekit documentation built on Feb. 2, 2024, 2:51 a.m.