Better initialization

[xukai92]

3 runs with the same code for sample(gdemo(x), Gibbs(1000, PG(30, 3, :s), HMC(1, 0.2, 3, :m)))

[Gibbs] Sampling...100% Time: 0:00:08
  1. s ≈ 49/24 (ϵ = 0.2) X
    s = 2.5293348244112015, diff = 0.487668157744535
  2. m ≈ 7/6 (ϵ = 0.2)   X
     m = 0.9544717098265469, diff = 0.2121949568401198
                  
[Gibbs] Sampling...100% Time: 0:00:11
  1. s ≈ 49/24 (ϵ = 0.2) ✓
    s = 2.223441561488678, diff = 0.18177489482201148
  2. m ≈ 7/6 (ϵ = 0.2)   ✓
    m = 0.9683037640417662, diff = 0.19836290262490053

[Gibbs] Sampling...100% Time: 0:00:13
  1. s ≈ 49/24 (ϵ = 0.2) ✓
    s = 1.9990257822648827, diff = 0.04264088440178382
  2. m ≈ 7/6 (ϵ = 0.2)   ✓
    m = 1.1000766092540895, diff = 0.06659005741257729

Why Gibbs is so sensitive to initialization?