from railrl.visualization import plot_util as plot
configure_matplotlib(matplotlib)
f = plot.filter_by_flat_params(
{'replay_kwargs.fraction_goals_are_env_goals': 0.5})
exps = plot.load_exps([
ashvin_base_dir +
's3doodad/share/steven/pushing-multipushing/pusher-reward-variants'
],
f,
suppress_output=True)
plot.tag_exps(exps, "name", "dsae")
plot.comparison(exps,
["Final puck_distance Mean", "Final hand_distance Mean"],
figsize=(6, 4),
vary=["vae_wrapped_env_kwargs.reward_params.type"],
default_vary={"reward_params.type": "unknown"},
smooth=plot.padded_ma_filter(10),
xlim=(0, 250000),
ylim=(0.15, 0.22),
method_order=None)
plt.gca().xaxis.set_major_formatter(plt.FuncFormatter(format_func))
plt.xlabel("Timesteps")
plt.ylabel("")
plt.title("Visual Pusher")
plt.legend([])
plt.tight_layout()
plt.savefig(output_dir + "pusher_reward_type_ablation.pdf")
offline_pusher = dp.get_trials(
pusher_dir,
criteria={
'rdim':
250,
'algo_kwargs.should_train_vae.$function':
'railrl.torch.vae.vae_schedules.never_train',
})
plt.figure(figsize=(6, 5))
plot.plot_trials(
OrderedDict([
("Online", online_pusher),
("Offline", offline_pusher),
]),
y_keys="Final sum_distance Mean",
x_key="Number of env steps total",
process_time_series=plot.padded_ma_filter(100),
)
plt.gca().xaxis.set_major_formatter(plt.FuncFormatter(format_func))
plt.xlabel("Timesteps")
plt.ylabel("Final Distance to Goal")
plt.title("Visual Pusher, Online Ablation")
lgnd = plt.legend(["Online", "Offline"],
bbox_to_anchor=(0.49, -0.2),
loc="upper center",
ncol=4,
handlelength=1)
plt.tight_layout()
plt.savefig(output_dir + "pusher_online_ablation.pdf")
print("File saved to", output_dir + "pusher_online_ablation.pdf")
# vae_trials = get_trials(
# '/home/vitchyr/git/railrl/data/doodads3/05-12-sawyer-reach-vae-rl-reproduce-2/',
# criteria={
# 'replay_kwargs.fraction_resampled_goals_are_env_goals': 0.5,
# 'replay_kwargs.fraction_goals_are_rollout_goals': 0.2,
# }
# )
y_keys = [
'Final distance Mean',
]
plot_trials(
{
'State - HER TD3': state_her_td3,
'State - TDM DDPG': state_tdm_ddpg,
'VAE - HER TD3': vae_trials,
# 'VAE - TD3': vae_td3_trials,
},
y_keys=y_keys,
process_time_series=padded_ma_filter(3),
# x_key=x_key,
)
plt.xlabel('Number of Environment Steps Total')
plt.ylabel('Final distance to Goal')
plt.savefig('/home/vitchyr/git/railrl/experiments/vitchyr/nips2018/plots'
'/reach.jpg')
plt.show()
# plt.savefig("/home/ashvin/data/s3doodad/media/plots/pusher2d.pdf")
})
offline_reacher = dp.get_trials(
reacher_dir,
criteria={
'algo_kwargs.should_train_vae.$function':
'railrl.torch.vae.vae_schedules.never_train',
})
plt.figure(figsize=(6, 5))
plot.plot_trials(
OrderedDict([
("Online", online_reacher),
("Offline", offline_reacher),
]),
y_keys="Final distance Mean",
x_key="Number of env steps total",
process_time_series=plot.padded_ma_filter(10, avg_only_from_left=True),
)
plt.gca().xaxis.set_major_formatter(plt.FuncFormatter(format_func))
plt.xlabel("Timesteps")
plt.ylabel("Final Distance to Goal")
plt.title("Visual Reacher Online Ablation")
plt.legend(["Online", "Offline"],
bbox_to_anchor=(0.49, -0.2),
loc="upper center",
ncol=4,
handlelength=1)
plt.tight_layout()
plt.savefig(output_dir + "reacher_online_ablation.pdf")
print("File saved to", output_dir + "reacher_online_ablation.pdf")