adding readxdsl

REQUIRE

@@ -2,3 +2,4 @@ julia 0.3-
 Graphs
 DataFrames
 TikzGraphs
+LightXML

src/BayesNets.jl

@@ -278,5 +278,6 @@ function estimateConvergence(df::DataFrame, a::Assignment)
 end
 
 include("learning.jl")
+include("io.jl")
 
 end # module

src/io.jl

@@ -0,0 +1,118 @@
+using LightXML
+
+export readxdsl
+export writelatex
+
+function assignment_dicts(
+ bn :: BayesNet,
+ nodes :: Vector{Symbol}
+ )
+
+ # constructs an array of assignment dictionaries [Dict{Symbol, Assignment}]
+ # assignments are created such that the first node's instantiations change most quickly
+
+ n_nodes = length(nodes)
+
+ @assert(n_nodes > 0)
+
+ domains = map(pa->domain(bn, pa), nodes)
+ nbins = map(dom->length(dom.elements), domains)
+ n_inst = prod(nbins)
+ retval = Array(Dict{Symbol, Any}, n_inst)
+ inst = ones(Int32, n_nodes)
+ inst2assignment = instantiation->[nodes[i]=>domains[i].elements[instantiation[i]] for i in 1:n_nodes]
+ retval[1] = inst2assignment(inst)
+ for perm = 2 : n_inst
+ # get the next instantiation
+ i = 1
+ while true
+ if inst[i] < nbins[i]
+ inst[i] += 1
+ inst[1:i-1] = 1
+ break
+ else
+ i += 1
+ end
+ end
+ retval[perm] = inst2assignment(inst)
+ end
+
+ retval
+end
+function discrete_parameter_function(
+ assignments :: Vector{Dict{Symbol, Any}}, # assumed to be in order
+ probs :: Vector{Float64}, # length ninst_target * length(assignments)
+ ninst_target :: Int # number of values in target variable domain
+ )
+
+ # returns a function mapping an assignment to the list of probabilities
+ n_assignments = length(assignments)
+ @assert(length(probs) == n_assignments * ninst_target)
+
+ ind = 1
+ dict = Dict{Dict{Symbol, Any}, Vector{Float64}}()
+ for a in assignments
+ dict[a] = probs[ind:ind+ninst_target-1]
+ ind += ninst_target
+ end
+
+ const names = keys(assignments[1])
+ return (a)->begin
+ a_extracted = [sym=>a[sym] for sym in names]
+ dict[a_extracted]
+ end
+
+ # return (a)->dict[a]
+end
+function readxdsl( filename::String )
+
+ # Loads a discrete Bayesian Net from XDSL format (SMILE / GeNIe)
+
+ splitext(filename)[2] == ".xdsl" || error("readxdsl only supports .xdsl format")
+
+ xdoc = parse_file(filename)
+ xroot = root(xdoc) 
+ ces = get_elements_by_tagname(xroot, "nodes")[1]
+ cpts = collect(child_elements(ces))
+
+ names = Array(Symbol, length(cpts))
+ for (i,e) in enumerate(cpts)
+ id = attribute(e, "id")
+ names[i] = symbol(id)
+ end
+
+ BN = BayesNet(names)
+
+ for (i,e) in enumerate(cpts)
+
+ node_sym = names[i]
+
+ # set the node's domain
+ states = [convert(ASCIIString, attribute(s, "id")) for s in get_elements_by_tagname(e, "state")]
+ n_states = length(states)::Int
+ BN.domains[i] = DiscreteDomain(states)
+
+ probs = map(s->float64(s), split(content(find_element(e, "probabilities"))))
+
+ # set any parents & populate probability table
+ parents_elem = get_elements_by_tagname(e, "parents")
+ if !isempty(parents_elem)
+ parents = map(s->symbol(s), split(content(parents_elem[1])))
+ println(parents)
+ for pa in parents
+ addEdge!(BN, pa, node_sym)
+ end
+
+ # populate probability table
+ reverse!(parents) # because SMILE varies first parent least quickly
+ assigments = assignment_dicts(BN, parents)
+ parameterFunction = discrete_parameter_function(assigments, probs, n_states)
+ setCPD!(BN, node_sym, CPDs.Discrete(states, parameterFunction))
+ else
+ # no parents
+ setCPD!(BN, node_sym, CPDs.Discrete(states, probs))
+ end
+ end
+
+ BN
+end