To make it easy to visualize, wrangle, and feature engineer time series data for forecasting and machine learning prediction.
Download the development version with latest features:
remotes::install_github("business-science/timetk")
Or, download CRAN approved version:
install.packages("timetk")
Full Time Series Machine Learning and Feature Engineering
Tutorial:
Showcases the (NEW) step_timeseries_signature()
for building
200+ time series features using parsnip
, recipes
, and
workflows
.
Visit the timetk website documentation for tutorials and a complete list of function references.
There are many R packages for working with Time Series data. Here’s
how timetk
compares to the “tidy” time series R packages for data
visualization, wrangling, and feature engineeering (those that leverage
data frames or tibbles).
Task | timetk | tsibble | feasts | tibbletime |
---|---|---|---|---|
Structure | ||||
Data Structure | tibble (tbl) | tsibble (tbl_ts) | tsibble (tbl_ts) | tibbletime (tbl_time) |
Visualization | ||||
Interactive Plots (plotly) | ✅ | ❌ | ❌ | ❌ |
Static Plots (ggplot) | ✅ | ❌ | ✅ | ❌ |
Time Series | ✅ | ❌ | ✅ | ❌ |
Correlation, Seasonality | ✅ | ❌ | ✅ | ❌ |
Anomaly Detection | ✅ | ❌ | ❌ | ❌ |
Data Wrangling | ||||
Time-Based Summarization | ✅ | ❌ | ❌ | ✅ |
Time-Based Filtering | ✅ | ❌ | ❌ | ✅ |
Padding Gaps | ✅ | ✅ | ❌ | ❌ |
Low to High Frequency | ✅ | ❌ | ❌ | ❌ |
Imputation | ✅ | ✅ | ❌ | ❌ |
Sliding / Rolling | ✅ | ✅ | ❌ | ✅ |
Feature Engineering (recipes) | ||||
Date Feature Engineering | ✅ | ❌ | ❌ | ❌ |
Holiday Feature Engineering | ✅ | ❌ | ❌ | ❌ |
Fourier Series | ✅ | ❌ | ❌ | ❌ |
Smoothing & Rolling | ✅ | ❌ | ❌ | ❌ |
Padding | ✅ | ❌ | ❌ | ❌ |
Imputation | ✅ | ❌ | ❌ | ❌ |
Cross Validation (rsample) | ||||
Time Series Cross Validation | ✅ | ❌ | ❌ | ❌ |
Time Series CV Plan Visualization | ✅ | ❌ | ❌ | ❌ |
More Awesomeness | ||||
Making Time Series (Intelligently) | ✅ | ✅ | ❌ | ✅ |
Handling Holidays & Weekends | ✅ | ❌ | ❌ | ❌ |
Class Conversion | ✅ | ✅ | ❌ | ❌ |
Automatic Frequency & Trend | ✅ | ❌ | ❌ | ❌ |
Investigate a time series…
taylor_30_min %>%
plot_time_series(date, value, .color_var = week(date),
.interactive = FALSE, .color_lab = "Week")
Visualize anomalies…
walmart_sales_weekly %>%
group_by(Store, Dept) %>%
plot_anomaly_diagnostics(Date, Weekly_Sales,
.facet_ncol = 3, .interactive = FALSE)
Make a seasonality plot…
taylor_30_min %>%
plot_seasonal_diagnostics(date, value, .interactive = FALSE)
Inspect autocorrelation, partial autocorrelation (and cross correlations too)…
taylor_30_min %>%
plot_acf_diagnostics(date, value, .lags = "1 week", .interactive = FALSE)
The timetk
package wouldn’t be possible without other amazing time
series packages.
timetk
function that uses a period (frequency) argument owes
it to ts()
.
plot_acf_diagnostics()
: Leverages stats::acf()
,
stats::pacf()
& stats::ccf()
plot_stl_diagnostics()
: Leverages stats::stl()
timetk
makes heavy
use of floor_date()
, ceiling_date()
, and duration()
for
“time-based phrases”.
%+time%
& %-time%
): "2012-01-01" %+time% "1 month 4 days"
uses lubridate
to intelligently
offset the dayts
, and
it’s predecessor is the tidyverts
(fable
, tsibble
, feasts
,
and fabletools
).
ts_impute_vec()
function for low-level vectorized
imputation using STL + Linear Interpolation uses na.interp()
under the hood.ts_clean_vec()
function for low-level vectorized
imputation using STL + Linear Interpolation uses tsclean()
under the hood.auto_lambda()
uses BoxCox.Lambda()
.timetk
does not import tibbletime
, it uses much of the
innovative functionality to interpret time-based phrases:
tk_make_timeseries()
- Extends seq.Date()
and seq.POSIXt()
using a simple phase like “2012-02” to populate the entire time
series from start to finish in February 2012.filter_by_time()
, between_time()
- Uses innovative endpoint
detection from phrases like “2012”slidify()
is basically rollify()
using slider
(see below).purrr
-syntax for complex rolling (sliding)
calculations.
slidify()
uses slider::pslide
under the hood.slidify_vec()
uses slider::slide_vec()
for simple vectorized
rolls (slides).pad_by_time()
function is a wrapper for padr::pad()
.step_ts_pad()
to apply padding as a preprocessing
recipe!ts
system, which is the same system the forecast
R
package uses. A ton of inspiration for visuals came from using
TSstudio
.My Talk on High-Performance Time Series Forecasting
Time series is changing. Businesses now need 10,000+ time series forecasts every day. This is what I call a High-Performance Time Series Forecasting System (HPTSF) - Accurate, Robust, and Scalable Forecasting.
High-Performance Forecasting Systems will save companies MILLIONS of dollars. Imagine what will happen to your career if you can provide your organization a “High-Performance Time Series Forecasting System” (HPTSF System).
I teach how to build a HPTFS System in my High-Performance Time Series Forecasting Course. If interested in learning Scalable High-Performance Forecasting Strategies then take my course. You will learn:
Modeltime
- 30+
Models (Prophet, ARIMA, XGBoost, Random Forest, & many more)GluonTS
(Competition Winners)