choose_int.recursive: Get n-step-ahead intervention values
In neilbramley/acl_source: Functions for active causal learning

Computes the most useful intervention recursively according to information, probability and utility gain. Requires a dataframe called li in the global workspace that encodes the likelihoods of any outcome of any intervention created with likelihood() Requires a dataframe called o which incodes the outcome space (see example) Requires a matrix of interventions called ints, 1 is on, 0 is free, -1 is off.

1 2	choose_int.recursive(ints, p_dist.old, Z.old = 1, li_c.old = as.matrix(1), depth = 0, n_steps = 1, compute_value = 1)

`compute_value`	defaults to 0=no. If 1 a value matrix C must be provided with a value for each judgement i given each true network j Additional arguments can then be passed to this objective function
`g`	matrix of hypothesis graphs, one line is one graph to be written by row to a matrix.
`pDist`	is the prior distribution over these graphs.

ints<-as.matrix(expand.grid(rep(list(0:2), 3)))
ints[ints==2]<--1
o<-expand.grid(rep(list(0:1 ), sqrt(dim(g)[2])))
dist<-prior('flat',g)
li<-likelihood(g,.1,.8)
out<-choose_int.recursive(ints, dist, n_steps=2, compute_value=0)