Backdoor Criterion

Since version 10

// --------------------------------------------------------------------------------------------------------------------
// <copyright file="BackdoorCriterionExample.cs" company="Bayes Server">
//   Copyright (C) Bayes Server.  All rights reserved.
// </copyright>
// --------------------------------------------------------------------------------------------------------------------


using BayesServer.Causal;
using BayesServer.Causal.Identification;
using BayesServer.Causal.Inference;
using BayesServer.Inference;
using System;
using System.Collections;
using System.Collections.Generic;
using System.Linq;
using System.Text;
using System.Threading.Tasks;

namespace BayesServer.HelpSamples
{
    public static class BackdoorCriterionExample
    {
        public static void Main()
        {
            // This example uses Causal AI (Causal Inference)
            // to calculate potential adjustment sets
            // required to calculate P(Y | Do(X)) in the presence of
            // unobserved confounders, using the 
            // Backdoor Criterion algorithm.


            // ** IMPORTANT **
            // If you want to actually perform the adjustment (inference)
            // the BackdoorInference class can automatically determine 
            // an appropriate adjustment set, making the following code 
            // unecessary.
            // This code is useful when you want to enumerate the possible
            // adjustment sets.

            var network = new Network();

            // TODO download the network from the Bayes Server User Interface (or Bayes Server Online)
            // and adjust the following path
            network.Load(@"Backdoor Collider.bayes");

            var x = network.Variables["X", true];
            var y = network.Variables["Y", true];

            var criterion = new BackdoorCriterion(network);

            // Method 1
            {
                
                var inf = new BackdoorInference(network);
                var query = new Table(y);
                inf.QueryDistributions.Add(query);

                // Set one or more interventions

                inf.Evidence.SetState(x.States[0], null, InterventionType.Do);

                // you could also set non-intervention evidence here

                var options = new BackdoorCriterionOptions();
                var output = (BackdoorCriterionOutput)criterion.Identify(inf.Evidence, query, options);

                Console.WriteLine("Method 1");
                PrintAdjustmentSets(output.Sets);

            }

            Console.WriteLine();

            // Method 2
            {
                var treatments = new List<CausalNode>();
                // add one or more treatments
                treatments.Add(new CausalNode(x.Node));

                var outcomes = new List<CausalNode>();
                // add one or more outcomes
                outcomes.Add(new CausalNode(y.Node));

                // you can add nodes to nonTreatmentEvidence if required, for
                // nodes that have normal (non-intervention) evidence set
                var nonTreatmentEvidence = new List<CausalNode>();

                var options = new BackdoorCriterionOptions();
                var output = (BackdoorCriterionOutput)criterion.Identify(treatments, outcomes, nonTreatmentEvidence, options);

                Console.WriteLine("Method 2");
                PrintAdjustmentSets(output.Sets);
                
            }
        }

        private static void PrintAdjustmentSets(IList<AdjustmentSet> adjustmentSets)
        {
            Console.WriteLine($"Adjustment sets found =  {adjustmentSets.Count}");

            foreach (var adjustmentSet in adjustmentSets)
            {
                Console.WriteLine($"  {{{string.Join(',', adjustmentSet.Nodes.Select(n => n.Node.Name))}}}");
            }
        }
    }
}