CausalQueries (version 0.0.3)

set_priors: Set prior distribution

Description

A flexible function to add priors to a model.

Usage

set_priors(
  model,
  priors = NULL,
  distribution = NA,
  alphas = NA,
  node = NA,
  label = NA,
  statement = NA,
  confound = NA,
  nodal_type = NA,
  param_names = NA,
  param_set = NA
)

Arguments

model

A model created with make_model

priors

A optional vector of positive reals indicating priors over all parameters. These are interpreted as arguments for Dirichlet distributions---one for each parameter set. To see the structure of parameter sets examine model$parameters_df

distribution

String (or list of strings) indicating a common prior distribution (uniform, jeffreys or certainty)

alphas

Real positive numbers giving hyperparameters of the Dirichlet distribution

node

A string (or list of strings) indicating nodes for which priors are to be altered

label

String. Label for nodal type indicating nodal types for which priors are to be altered

statement

A causal query (or list of queries) that determines nodal types for which priors are to be altered

confound

A confound statement (or list of statements) that restricts nodal types for which priors are to be altered

nodal_type

String. Label for nodal type indicating nodal types for which priors are to be altered

param_names

String. The name of specific parameter in the form of, for example, 'X.1', 'Y.01'

param_set

String. Indicates the name of the set of parameters to be modified (useful when setting confounds)

Value

An object of class causal_model. It essentially returns a list containing the elements comprising a model (e.g. 'statement', 'nodal_types' and 'DAG') with the `priors` attached to it.

Details

Four arguments govern *which* parameters should be altered. The default is 'all' but this can be reduced by specifying

* label The label of a particular nodal type, written either in the form Y0000 or Y.Y0000

* node, which restricts for example to parameters associated with node 'X'

* statement, which restricts for example to nodal types that satisfy the statement 'Y[X=1] > Y[X=0]'

* confound, which restricts for example to nodal types that satisfy the statement 'Y[X=1] > Y[X=0]'

Two arguments govern what values to apply:

* alphas is one or more non negative numbers and

* distribution indicates one of a common class: uniform, jeffreys, or 'certain'

Any arguments entered as lists or vectors of size > 1 should be of the same length as each other.

For more examples and details see make_priors

See Also

Other priors: get_priors(), make_par_values_multiple(), make_par_values(), make_priors(), make_values_task_list()

Examples

# NOT RUN {
library(dplyr)
# Set priors to the model
model <- make_model('X -> Y') %>%
  set_priors(alphas = 3)
get_priors(model)
model <- make_model('X -> Y') %>%
  set_priors(distribution = 'jeffreys')
get_priors(model)

# Pass all nodal types
model <- make_model("Y <- X") %>%
  set_priors(.4)
get_priors(model)
model <- make_model("Y <- X") %>%
  set_priors(.7)
get_priors(model)
model <- make_model("Y <- X") %>%
  set_priors(distribution = "jeffreys")
get_priors(model)

# Passing by names of node, parameter set or label
model <- make_model('X -> M -> Y')
model_new_priors <- set_priors(model, param_name = "X.1", alphas = 2)
get_priors(model_new_priors)
model_new_priors <- set_priors(model, node = 'X', alphas = 3)
get_priors(model_new_priors)
model_new_priors <- set_priors(model, param_set = 'Y', alphas = 5)
get_priors(model_new_priors)
model_new_priors <- set_priors(model, node = c('X', 'Y'), alphas = 3)
get_priors(model_new_priors)
model_new_priors <- set_priors(model, param_set = c('X', 'Y'), alphas = 5)
get_priors(model_new_priors)
model_new_priors <- set_priors(model, node = list('X', 'Y'), alphas = list(3, 6))
get_priors(model_new_priors)
model_new_priors <- set_priors(model, param_set = list('X', 'Y'), alphas = list(4, 6))
get_priors(model_new_priors)
model_new_priors <- set_priors(model,
                               node = c('X', 'Y'),
                               distribution = c('certainty', 'jeffreys'))
get_priors(model_new_priors)
model_new_priors <- set_priors(model,
                               param_set = c('X', 'Y'),
                               distribution = c('jeffreys', 'certainty'))
get_priors(model_new_priors)
model_new_priors <- set_priors(model, label = '01', alphas = 5)
get_priors(model_new_priors)
model_new_priors <- set_priors(model, node = 'Y', label = '00', alphas = 2)
get_priors(model_new_priors)
model_new_priors <- set_priors(model, node =c('M', 'Y'), label = '11', alphas = 4)
get_priors(model_new_priors)

# Passing a causal statement
model_new_priors <- set_priors(model, statement = 'Y[M=1] > Y[M=0]', alphas = 3)
get_priors(model_new_priors)
model_new_priors <- set_priors(model,
                               statement = c('Y[M=1] > Y[M=0]', 'M[X=1]== M[X=0]'),
                               alphas = c(3, 2))
get_priors(model_new_priors)

# Passing a confound statement
model <- make_model('X->Y') %>%
 set_confound(list(X = 'Y[X=1] > Y[X=0]', X = 'Y[X=1] < Y[X=0]'))

model_new_priors <- set_priors(model,
            confound = list(X='Y[X=1] > Y[X=0]',
                            X='Y[X=1] < Y[X=0]'),
            alphas = c(3, 6))
get_priors(model_new_priors)
model_new_priors <- set_priors(model, confound= list(X='Y[X=1] > Y[X=0]'), alphas = 4)
get_priors(model_new_priors)
model_new_priors <- set_priors(model, param_set='X_1', alphas = 5)
get_priors(model_new_priors)
model_new_priors <-  set_priors(model, param_names='X_2.1', alphas = .75)
get_priors(model_new_priors)

# A more complex example
model <- make_model('X -> Y') %>%
  set_confound(list(X = 'Y[X=1]>Y[X=0]'))%>%
  set_priors(statement = 'X[]==1',
              confound = list(X = 'Y[X=1]>Y[X=0]', X = 'Y[X=1]<Y[X=0]'),
              alphas = c(2, .5))
get_priors(model)
# }