def setup_factor_graph(self): self.vecfac = dai.VecFactor() elem_map = {} for elem in self.variable_map: if elem.variable_name not in elem_map: elem_map[elem.variable_name] = { elem.variable_type: elem.variable_id } else: elem_map[elem.variable_name][ elem.variable_type] = elem.variable_id #dai_var_map[elem.variable] = dai.Var(elem.variable, elem.dim) for cpt in sorted(self.cpt_map.keys(), key=lambda x: x.factor_id): cpt_value = self.cpt_map[cpt] var_list = dai.VarSet() v_list = list(cpt_value.variables) v_list.sort() for v_id in v_list: v = self.variable_map.get_variable_by_id(v_id) if v not in self.dai_variables: self.dai_variables[v] = dai.Var(v.variable_id, v.dim) var_list.append(self.dai_variables[v]) dai_factor = dai.Factor(var_list) for i, v in enumerate(cpt_value.cpt_linear_table()): dai_factor[i] = v self.vecfac.append(dai_factor)
def build_libdaiFactorGraph_from_SATproblem(clauses, N): ''' Inputs: - clauses: (list of list of ints) variables should be numbered 1 to N with no gaps - N: (int) the number of variables in the sat problem. This should be equal the largest variable name (otherwise could have problems with implicit variables, e.g. a missing variable x_i is equivalent to the clause (x_i or not x_i), doubling the solution count) Outputs: - sat_FactorGraph: (dai.FactorGraph) ''' #make sure variables are numbered 1 to N with no gaps all_literals = {} max_literal = 0 for clause in clauses: for var in clause: lit = abs(var) all_literals[lit] = True if lit > max_literal: max_literal = lit assert (len(all_literals) == max_literal) assert (max_literal == N) for i in range(1, max_literal + 1): assert (i in all_literals) # Define binary variables in the factor graph variables = [] for var_idx in range(max_literal): variables.append(dai.Var(var_idx, 2)) #variable can take 2 values factors = [] for clause in clauses: factors.append(build_libdaiFactor_fromClause(clause, variables)) assert (len(factors) == len(clauses)) assert (len(variables) == max_literal) # Build factor graph sat_Factors = dai.VecFactor() for factor in factors: sat_Factors.append(factor) sat_FactorGraph = dai.FactorGraph(sat_Factors) return sat_FactorGraph
def build_libdaiFactorGraph_from_SpinGlassModel(sg_model, fixed_variables={}): ''' copied from https://github.com/jkuck/mrf_nesting_ub/blob/master/Factor_Graphs/libdai_ising_model.py Inputs: - sg_model: (SG_model) - fixed_variables: (dictionary) key: (int) 0 to N-1 variable index value: (int) -1 or 1, the value the variable is fixed to Outputs: - sg_FactorGraph: (dai.FactorGraph) ''' N = sg_model.lcl_fld_params.shape[0] assert (N == sg_model.lcl_fld_params.shape[1]) # accumulator_var_idx = None # if len(fixed_variables) > 0: # assert(len(fixed_variables) < N**2) # for var_idx in range(N**2): # if var_idx not in fixed_variables: # accumulator_var_idx = var_idx #this variable gets all the fixed factors multiplied into its single node factor # break # assert(accumulator_var_idx is not None) # Define binary variables in the factor graph variables = [] for var_idx in range(N**2): if var_idx in fixed_variables: variables.append(dai.Var(var_idx, 1)) #variable can take 1 values if var_idx not in fixed_variables: variables.append(dai.Var(var_idx, 2)) #variable can take 2 values factors = [] # Define factors for each single variable factor for var_idx in range(N**2): r = var_idx // N c = var_idx % N factors.append( build_single_node_factor(variables, fixed_variables, var_idx, f=sg_model.lcl_fld_params[r, c])) # Define pairwise factors for var_idx in range(N**2): r = var_idx // N c = var_idx % N if r < N - 1: factors.append( build_pairwise_factor(variables, fixed_variables, var_idx1=var_idx, var_idx2=var_idx + N, c=sg_model.cpl_params_v[r, c])) if c < N - 1: factors.append( build_pairwise_factor(variables, fixed_variables, var_idx1=var_idx, var_idx2=var_idx + 1, c=sg_model.cpl_params_h[r, c])) #Define higher order factors if sg_model.contains_higher_order_potentials: # Add higher order factors for potential_idx in range(sg_model.ho_potential_count): factor_potentials_list.append( sg_model.higher_order_potentials[potential_idx]) masks_list.append( torch.zeros_like( sg_model.higher_order_potentials[potential_idx])) assert (len(factors) == N**2 + 2 * N * (N - 1)) # Build factor graph sg_Factors = dai.VecFactor() for factor in factors: sg_Factors.append(factor) sg_FactorGraph = dai.FactorGraph(sg_Factors) return sg_FactorGraph
# Define conditional probability of W given S and R SRW = dai.VarSet(S, R) SRW.append(W) P_W_given_S_R = dai.Factor(SRW) P_W_given_S_R[0] = 1.0 # S = 0, R = 0, W = 0 P_W_given_S_R[1] = 0.1 # S = 1, R = 0, W = 0 P_W_given_S_R[2] = 0.1 # S = 0, R = 1, W = 0 P_W_given_S_R[3] = 0.01 # S = 1, R = 1, W = 0 P_W_given_S_R[4] = 0.0 # S = 0, R = 0, W = 1 P_W_given_S_R[5] = 0.9 # S = 1, R = 0, W = 1 P_W_given_S_R[6] = 0.9 # S = 0, R = 1, W = 1 P_W_given_S_R[7] = 0.99 # S = 1, R = 1, W = 1 # Build factor graph consisting of those four factors SprinklerFactors = dai.VecFactor() SprinklerFactors.append(P_C) SprinklerFactors.append(P_R_given_C) SprinklerFactors.append(P_S_given_C) SprinklerFactors.append(P_W_given_S_R) SprinklerNetwork = dai.FactorGraph(SprinklerFactors) # Write factorgraph to a file SprinklerNetwork.WriteToFile('sprinkler.fg') print 'Sprinkler network written to sprinkler.fg' # Output some information about the factorgraph print SprinklerNetwork.nrVars(), 'variables' print SprinklerNetwork.nrFactors(), 'factors' # Calculate joint probability of all four variables
def build_libdaiFactorGraph_from_SpinGlassModel(sg_model, fixed_variables={}): ''' copied from https://github.com/jkuck/mrf_nesting_ub/blob/master/Factor_Graphs/libdai_ising_model.py Inputs: - sg_model: (SG_model) - fixed_variables: (dictionary) key: (int) 0 to N-1 variable index value: (int) -1 or 1, the value the variable is fixed to Outputs: - sg_FactorGraph: (dai.FactorGraph) ''' N = sg_model.lcl_fld_params.shape[0] assert (N == sg_model.lcl_fld_params.shape[1]) # accumulator_var_idx = None # if len(fixed_variables) > 0: # assert(len(fixed_variables) < N**2) # for var_idx in range(N**2): # if var_idx not in fixed_variables: # accumulator_var_idx = var_idx #this variable gets all the fixed factors multiplied into its single node factor # break # assert(accumulator_var_idx is not None) # Define binary variables in the factor graph variables = [] for var_idx in range(N**2): if var_idx in fixed_variables: variables.append(dai.Var(var_idx, 1)) #variable can take 1 values if var_idx not in fixed_variables: variables.append(dai.Var(var_idx, 2)) #variable can take 2 values factors = [] # Define factors for each single variable factor for var_idx in range(N**2): r = var_idx // N c = var_idx % N factors.append( build_single_node_factor(variables, fixed_variables, var_idx, f=sg_model.lcl_fld_params[r, c])) # Define pairwise factors for var_idx in range(N**2): r = var_idx // N c = var_idx % N if r < N - 1: factors.append( build_pairwise_factor(variables, fixed_variables, var_idx1=var_idx, var_idx2=var_idx + N, c=sg_model.cpl_params_v[r, c])) if (c < N - 1) and (r == 0): factors.append( build_pairwise_factor(variables, fixed_variables, var_idx1=var_idx, var_idx2=var_idx + 1, c=sg_model.cpl_params_h[r, c])) assert (len(factors) == N**2 + (N - 1) + N * (N - 1)) # Build factor graph sg_Factors = dai.VecFactor() for factor in factors: sg_Factors.append(factor) sg_FactorGraph = dai.FactorGraph(sg_Factors) # bp_z_est = run_loopyBP(sg_model, maxiter=1000) # exact_z = libdai_utils.junction_tree(sg_FactorGraph) # print("bp_z_est =", bp_z_est) # print("exact_z =", exact_z) return sg_FactorGraph