ptable makes it easy to create perturbation tables that can be used for applying noise to statistical tables with any cell-key method approach - among others either the cellKey()-package or the standalone tool TauArgus.
The package provides four main functions to create the perturbation tables:
create_ptable(): generic function that creates a ptable, either for
frequency count or magnitude tables with a various set of options.
create_cnt_ptable(): creates a ptable suitable for frequency count
tables.
create_num_ptable(): creates a ptable suitable for magnitude tables
(i.e. with numerical variables).
modify_cnt_ptable(): modifies the ptable for a higher level of protection
create_ptable(
D,
V,
js = 0,
pstay = NULL,
optim = 1,
mono = TRUE,
step = 1,
icat = NULL,
table = "cnts",
type = "all",
label = paste0("D", D, "V", V * 100),
monitoring = FALSE,
debugging = FALSE,
create = TRUE,
params = NULL
)create_cnt_ptable(
D,
V,
js = 0,
pstay = NULL,
optim = 1,
mono = TRUE,
label = paste0("D", D, "V", V * 100),
monitoring = FALSE,
create = TRUE
)
create_num_ptable(
D,
V,
pstay = NULL,
optim = 1,
mono = TRUE,
step = 2,
icat = NULL,
type = "all",
label = paste0("D", D, "V", V * 100),
monitoring = FALSE,
create = TRUE
)
Returns ptable object including the
created perturbation table by default. If the argument create = FALSE,
a ptable_params object is returned.
perturbation parameter for maximum noise (scalar integer)
perturbation parameter for variance (scalar double)
threshold value for blocking of small frequencies (i.e. the perturbation will not produce positive cell values that are equal to or smaller than the threshold value). (scalar integer)
optional parameter to set the probability (0 < p < 1) of an original frequency to remain unperturbed: NA (default) no preset probability (i.e. produces the maximum entropy solution)
optimization parameter: 1 standard approach (default) with
regular constraints, 4 alternative approach with simplified constraints
(may work if constraints using the standard approach are violated)
(logical) vector specifying optimization parameter for monotony condition
(integer) number of steps for the noise (between two integer
values). Whereas the cell-key approach
for frequency count tables only allows to have noise values that are
integers (step = 1)
$$-D, 1-D, 2-D, ..., -1, 0, 1, ..., D-2, D-1, D$$
the noise distribution for magnitude values does not have
to be integer valued:
$$-D, (1/step)-D, (2/step)-D, ..., 0, ..., D-(2/step), D-(1/step), D$$
The reciprocal of step (='step width') is computed and used internally
for the perturbation table.
(integer) categorized original frequencies i
(character) type of the table: frequency count (cnts)
or magnitude table (nums)
(character) type indicator for the extra column 'type' used for
magnitude tables: 'even', 'odd' or 'all' (default)
(character) label of the Output
(logical) output monitoring on/off
(logical) debug monitoring on/off
(logical) scalar specifying to create just the
input parameters of class ptable_params (FALSE) or also to
create the perturbation table object of class ptable
(default: TRUE)
object of class ptable_params can be used as input instead of the remaining parameters
The perturbation probabilities are constructed given the following constraints:
Maximum noise
Zero mean (unbiased noise)
Fixed noise variance
Transition probabilities are between zero and one and the sum up to 1
Perturbations will not produce negative cell values or positive cell values equal to or less than a specific threshold value
# create ptable for frequency count tables
create_cnt_ptable(D = 3, V = 1.08, js = 1, label = "ptable_frequency_tab")
# create ptable for magnitude tables
create_num_ptable(D = 5, V = 2, step = 4, icat = c(1, 3, 5))
# \donttest{
# create ptable for frequency or magnitude tables
create_ptable(D = 3, V = 1.08, js = 1, table="cnts")
create_ptable(D = 5, V = 2, step = 4, icat = c(1, 4, 5), table="nums")
# }
Run the code above in your browser using DataLab