In jangorecki/data.cube: OLAP cube data type

Subset multidimensional data with cube

Jan Gorecki, 2015-11-19

R cube class defined in data.cube package.

Please note there is a new updated oop cube class called data.cube. See Subset and aggregate multidimensional data with data.cube vignette instead.

characteristic of [.cube

what's the same in [.array

• subsetting dimensions by keys
• handling drop argument
• returns same class object, so is chainable

what's more than [.array

• can subset on hierarchy of a dimension
• scales with dimensions and their cardinality
• stores multiple measures
• high performance thanks to data.table

what's different to [.array

• array can be subsetted by dimension keys but also by integer position
• in cube subset directly by integer position is not possible, but you can have integer surrogate keys which can refer to any custom integer positions

start session

if(!"data.cube" %in% rownames(installed.packages())) install.packages(
    "data.cube", repos = paste0("https://", c("jangorecki.github.io/data.cube","cran.rstudio.com"))
)
library(data.table)
library(data.cube)

tiny example

set.seed(1)
# array
ar = array(rnorm(8,10,5), 
           rep(2,3), 
           dimnames = list(color = c("green","red"), 
                           year = c("2014","2015"), 
                           country = c("UK","IN")))
# cube normalized to star schema
cb = as.cube(ar)

ar["green","2015",]
cb["green","2015",]

ar["green",c("2014","2015"),]
cb["green",c("2014","2015"),]

# tabular representation of array is just a formatting on the cube
format(cb["green",c("2014","2015"),], 
       dcast = TRUE, 
       formula = year ~ country)

ar[,"2015",c("UK","IN")]
cb[,"2015",c("UK","IN")]
format(cb[,"2015",c("UK","IN")], 
       dcast = TRUE, 
       formula = color ~ country)

hierarchies example

# as.cube.list - investigate X to see structure
X = populate_star(N=1e5)
lapply(X, sapply, ncol)
lapply(X, sapply, nrow)
cb = as.cube(X)
print(cb)

# slice
cb["Mazda RX4"]
cb["Mazda RX4",,"BTC"]
# dice
cb[,, c("CNY","BTC"), c("GA","IA","AD")]

# check dimensions
cb$dims

# use dimensions hierarchy attributes for slice and dice, mix filters from various levels in hierarchy
cb["Mazda RX4",, .(curr_type = "crypto"),, .(time_year = 2014L, time_quarter_name = c("Q1","Q2"))]
# same as above but more verbose
cb[product = "Mazda RX4",
   customer = .(),
   currency = .(curr_type = "crypto"),
   geography = .(),
   time = .(time_year = 2014L, time_quarter_name = c("Q1","Q2"))]

# cube `[` operator returns another cube so queries can be chained
cb[,,, .(geog_region_name = "North Central")
   ][,,, .(geog_abb = c("IA","NV","MO")), .(time_year = 2014L)
     ]

scalability

# ~1e5 facts for 5 dims of cardinalities: 32, 32, 49, 50, 1826
cb = as.cube(populate_star(N=1e5))
## estimated size of memory required to store an base R `array` for single numeric measure
sprintf("array: %.2f GB", (prod(dim(cb)) * 8)/(1024^3))
## fact table of *cube* object having multiple measures
sprintf("cube: %.2f GB", as.numeric(object.size(cb$env$fact$sales))/(1024^3))

# ~1e6 facts for 5 dims of cardinalities: 32, 32, 49, 50, 1826
cb = as.cube(populate_star(N=1e6))
## estimated size of memory required to store an base R `array` for single numeric measure
sprintf("array: %.2f GB", (prod(dim(cb)) * 8)/(1024^3))
## fact table of *cube* object having multiple measures
sprintf("cube: %.2f GB", as.numeric(object.size(cb$env$fact$sales))/(1024^3))

# ~1e6 facts for 5 dims of cardinalities: 32, 32, 49, 50, 3652
cb = as.cube(populate_star(N=1e6, Y = c(2005L,2014L))) # bigger time dimension
## estimated size of memory required to store an base R `array` for single numeric measure
sprintf("array: %.2f GB", (prod(dim(cb)) * 8)/(1024^3))
## fact table of *cube* object having multiple measures
sprintf("cube: %.2f GB", as.numeric(object.size(cb$env$fact$sales))/(1024^3))

examples

Lots of examples can be found in tests: tests/tests-sub-.cube.R.
Feel free to PR your use case for future regression testing.

jangorecki/data.cube documentation built on Aug. 22, 2019, 4:15 p.m.