Metropolis on Two Humps

The target p(θ) = ½ N(θ; −1, 0.5²) + ½ N(θ; 1, 0.5²) has two humps. We can compute p(θ) anywhere, but we can't sample from it. Start somewhere; propose θ_p from q(θ; θ_c) = N(θ_c, σ²); move with probability p_move = min(1, p(θ_p) / p(θ_c)), otherwise stay. Where the walk has been is the sample, and the width σ is the knob.