Arguments
x
the index of the treatment variable.
y
the index of the outcome variable.
latent_idx
an array with the indices of variables which should be considered latent
dat
a matrix of binary data, can be ignored if a model is provided.
g
a binary matrix encoding a causal graph, where g[i, j] == 1 if
a directed edge from vertex $j$ to $i$ should exist, 0 otherwise. This is only required
if a ground truth model exists.
model
if g is specified, this needs to be specified too. This argument should be
a list of conditional probability tables, each encoding the conditional probability of
each vertex in g given its parents. Entry model[[i]] is
an array of non-negative numbers, describing the probability of random variable/vertex
$i$ being equal to 1. In particular, model[[i]][j] is the conditional
probability of this event given that the parents of $i$ are in state $j$.
States are indexed as follows. If $S$ is the binary string corresponding to the
binary values of the parents of $i$ in g, sorted by their index, then
$j$ is given by $1 + bin2dec(S)$, where $bin2dec$ is the transformation of
a binary string into a decimal number.
num_v_max
the maximum dimensionality in which the joint distribution implied by a model is pre-computed.
Having this pre-computed can speed up some computations for methods that use the provided
ground truth model. Because the space required to store a joint distribution
grows exponentially with the dimensionality, this quantity cannot be too large.