observe with multivariate gaussian

[Thalia Vrantsidis]

Hello,

I am trying to put an observe statement where the observation is a single 2-d value from a multivariate gaussian distribution. 

When I try what I think should be correct, it doesn't give an error, but seems to get stuck forever on: Initialization warning [4/4]: Trace not initialized after 1000000 attempts.

I think I am just running into a syntax issue that I can't figure out. 

My current code is below.  

Any help on this would be appreciated! 

//observing two pieces of evidence (mu1 and mu2) with certain reliabilities (var 1 and var2)

//the two pieces of evidence have possibly correlated noise

var inference_model = function(mu1,mu2,var1,var2, cor) {

  return Infer({method: 'MCMC', samples: 10000}, function() {

    var prior = Gaussian({mu: 50, sigma: 25})

    var trueX = sample(prior) 

    var cov = cor*Math.sqrt(var1)*Math.sqrt(var2)

    var evidence = MultivariateGaussian({mu: Vector([trueX,trueX]),

                          cov: Matrix([[var1,cov],

                                       [cov,var2]])})

    observe(evs, [mu1,mu2]) 

    //observe(evs,  {"0":mu1,"1":mu2}) 

    

    return {'trueX':trueX}

  })

}

var Posterior = inference_model(40, 60, 10,10, 0.1)

viz.marginals(Posterior)

[null-a]

Hi. I think you just need to pass the observation as a vector rather than an array, i.e. `observe(evs, Vector([mu1,mu2]))`?

[Thalia Vrantsidis]

Yes, passing it as a vector worked! 

Thanks!!