M4: M4 Competition data

In carlanetto/Mcomp2018: Data from the M4-Competition

Description

Featuring 100000 time series from different domains

The 100000 time series from the M4-competition. Includes the submitted forecasts (https://www.m4.unic.ac.cy/)

Usage

data(M4)

Format

M4 is a list of 100000 series. Each series within M4 is an object with the following structure:

st

Series number and period. For example "Y1" denotes first yearly series, "Q20" denotes 20th quarterly series and so on.

n

The number of observations in the time series

h

The number of required forecasts

period

Interval of the time series. Possible values are "Yearly", "Quarterly", "Monthly", "Weekly", "Daily" & "Hourly".

type

The type of series. Possible values are "Demographic", "Finance", "Industry", "Macro", "Micro" & "Other".

x

A time series of length n (the historical data)

xx

A time series of length h (the future data)

pt_ff

A 25xh matrix with the top 25 submissions to the M4 Competition, one on each row. Information can be found in the submission_info data.frame included in the package. Includes some benchmark methods such as Theta of ARIMA. The order of the rows is the rank in competition, e.g. best method is in row 1.

up_ff

A matrix containing the upper bound for the prediction intervals of the submissions. The row are ordered is the same as in pt_ff, not by their prediction interval performance. Partipants that submitted point forecasts but not prediction intervals have NAs in their row.

low_ff

A matrix similar to up_ff but with the lower bounds of the prediction intervals.

Note

Series start is provided in year format for Yearly, Monthly and Quarterly series. For Daily and Hourly series, it is given in the number of days after 1970-01-01 and for Weekly data it is given in number of weeks since 1970-01-01.

Source

M4 Competition Dataset

References

(M4 Competition Web) (M4 Competition GitHub Repository)

Examples

data(M4)
names(M4[[1]])
#extract yearly series
yearly_M4 <- Filter(function(l) l$period == "Yearly", M4)

carlanetto/Mcomp2018 documentation built on June 13, 2019, 3:44 p.m.