Function compare_latent_plot

experiments/RL/visualization.py:127–144

(pop, pop_raw, pop_cmaes, random_map, pop_large, random_map_large, ax=None)

    ax.scatter(z[:, 0], z[:, 1], zorder=zorder, marker='o', color=color, alpha=alpha, label=label, edgecolors='none')

def compare_latent_plot(pop, pop_raw, pop_cmaes, random_map, pop_large, random_map_large, ax=None):
    if ax is None:
        ax = plt.gca()
    latent_plot(random_map(pop).detach(), ax, color='#E93A01', label='latent diffusion evolution', alpha=0.5)
    latent_plot(random_map(pop_raw).detach(), ax, color='#46B3D5', label='DiffEvo', alpha=0.25)
    latent_plot(random_map(pop_cmaes).detach(), ax, color='#6F6E6E', alpha=0.5, label='CMA-ES')
    latent_plot(random_map_large(pop_large).detach(), ax, color='#F5851E', alpha=0.25, label='latent DiffEvo (high-d)')

    # calculate the range of the data
    x = torch.cat([random_map(pop).detach(), random_map(pop_raw).detach(), random_map_large(pop_large).detach()], dim=0) # not include cmaes
    x_mean = x.mean(dim=0)
    x_std = x.std(dim=0)
    n = 3.0

    ax.set_xlabel('$z_1$')
    ax.set_ylabel('$z_2$')
    ax.set_xlim(x_mean[0]-n*x_std[0], x_mean[0]+n*x_std[0])
    ax.set_ylim(x_mean[1]-n*x_std[1], x_mean[1]+n*x_std[1])

def draw_cartpole_demo(ax):
    ax.axhline(y=0, color='gray', linestyle='--')