maltese: machine learning for time series

Installing

# install.packages("devtools")
devtools::install_github("bearloga/maltese")

Example

Data

The included dataset is a tidy time series of pageviews for R's article on English Wikipedia from 2015-10-01 to 2017-01-30.

library(maltese)
head(r_enwiki)

date	pageviews
2015-10-01	3072
2015-10-02	2575
2015-10-03	1431
2015-10-04	1540
2015-10-05	3041
2015-10-06	3695

We can use mlts_transform to convert the data into a machine learning-friendly format with a 7-day lag:

mlts <- mlts_transform(
  r_enwiki, date, pageviews,
  p = 7, # how many previous points of data to use as features
  granularity = "day", # optional, can be automatically detected,
  extras = TRUE, extrasAsFactors = TRUE # FALSE by default :D
)
head(mlts)

dt	y	mlts_extras_monthday	mlts_extras_weekday	mlts_extras_week	mlts_extras_month	mlts_extras_year	mlts_lag_1	mlts_lag_2	mlts_lag_3	mlts_lag_4	mlts_lag_5	mlts_lag_6	mlts_lag_7
2015-10-08	3278	8	Thursday	41	October	2015	3385	3695	3041	1540	1431	2575	3072
2015-10-09	2886	9	Friday	41	October	2015	3278	3385	3695	3041	1540	1431	2575
2015-10-10	1692	10	Saturday	41	October	2015	2886	3278	3385	3695	3041	1540	1431
2015-10-11	1902	11	Sunday	41	October	2015	1692	2886	3278	3385	3695	3041	1540
2015-10-12	3030	12	Monday	41	October	2015	1902	1692	2886	3278	3385	3695	3041
2015-10-13	3245	13	Tuesday	41	October	2015	3030	1902	1692	2886	3278	3385	3695

Results

See the vignette for a detailed walkthrough.

Additional Information

Users of maltese may also be interested in timetk (available on CRAN) which provides several utility functions for working with and manipulating time series data into a ML-friendly form.