lr = plot.load_exps( [vitchyr_base_dir + "papers/nips2018/06-06-lr-baseline-pusher-0.2-range/"], plot.filter_by_flat_params({ 'rdim': '16--lr-1e-3', }), suppress_output=True, ) plot.tag_exps(lr, "name", "l&r") plot.comparison( #lr + ours + oracle + her + dsae, ours + oracle + her + lr + dsae, ["Final puck_distance Mean", "Final hand_distance Mean"], vary=["name"], smooth=plot.padded_ma_filter(10), #method_order=[4, 0, 1, 3, 2], ylim=(0.1, 0.28), # xlim=(0, 250000), xlim=(0, 500000), figsize=(6, 4), ) plt.gca().xaxis.set_major_formatter(plt.FuncFormatter(format_func)) plt.xlabel("Timesteps") plt.ylabel("") # "Final Distance to Goal") plt.title("Visual Pusher Baselines") plt.legend([]) # [our_method_name, "DSAE", "HER", "Oracle", "L&R", ]) plt.tight_layout() plt.savefig(output_dir + "pusher_baselines.pdf") print("File saved to", output_dir + "pusher_baselines.pdf")
configure_matplotlib(matplotlib) dirs = [ ashvin_base_dir + 's3doodad/share/camera_ready_door', ] exps = plot.load_exps(dirs, suppress_output=True) plot.comparison( exps, ["Final angle_difference Mean"], [ # "seed", "exp_prefix", ], default_vary={"env_kwargs.randomize_position_on_reset": True}, smooth=plot.padded_ma_filter(10), print_final=False, print_min=False, print_plot=True, xlim=(0, 200000), # ylim=(0, 0.35), figsize=(6, 4), method_order=(3, 2, 4, 0, 1), ) plt.gca().xaxis.set_major_formatter(plt.FuncFormatter(format_func)) plt.xlabel("Timesteps") plt.ylabel("Final Distance to Goal") plt.legend([]) # [], # [ # "RIG",
f = plot.filter_by_flat_params({ 'algo_kwargs.num_updates_per_env_step': 4, "replay_kwargs.fraction_goals_are_env_goals": 0.5, "replay_kwargs.fraction_goals_are_rollout_goals": 0.2 }) exps = plot.load_exps(dirs, suppress_output=True) plot.comparison( exps, ["Final puck_distance Mean", "Final hand_distance Mean"], vary=[ "replay_kwargs.fraction_goals_are_env_goals", "replay_kwargs.fraction_goals_are_rollout_goals" ], default_vary={"replay_strategy": "future"}, smooth=plot.padded_ma_filter(10), xlim=(0, 500000), ylim=(0.14, 0.26), figsize=(6, 5), method_order=[2, 1, 0, 3], ) plt.gca().xaxis.set_major_formatter(plt.FuncFormatter(format_func)) # plt.ylabel("") plt.xlabel("Timesteps") plt.ylabel("Final Distance to Goal") plt.title("Visual Pusher") plt.legend( [ our_method_name, "None",
import matplotlib.pyplot as plt from rlkit.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/reward-reaching-sweep"], f, suppress_output=True) plot.comparison( exps, "Final distance Mean", vary=["reward_params.type"], # smooth=plot.padded_ma_filter(10), ylim=(0.0, 0.2), xlim=(0, 10000), # method_order=[1, 0, 2]), figsize=(6, 4), ) plt.gca().xaxis.set_major_formatter(plt.FuncFormatter(format_func)) plt.xlabel("Timesteps") plt.ylabel("Final Distance to Goal") plt.title("Visual Reacher") plt.legend([]) plt.tight_layout() plt.savefig(output_dir + "reacher_reward_type_ablation.pdf") print("File saved to", output_dir + "reacher_reward_type_ablation.pdf")
dsae = plot.load_exps( [ashvin_base_dir + 's3doodad/share/steven/no-relabeling-test'], suppress_output=True) plot.tag_exps(dsae, "name", "dsae") lr = plot.load_exps([ vitchyr_base_dir + "papers/nips2018/autoencoder_result/05-25-sawyer-reacher-autoencoder-ablation-final/" ], suppress_output=True) plot.tag_exps(lr, "name", "l&r") plot.comparison( ours + oracle + her + lr + dsae, "Final distance Mean", vary=["name"], figsize=(6, 4), # figsize=(7.5, 4), method_order=[4, 0, 1, 3, 2], ylim=(0.0, 0.25), xlim=(0, 10000), ) plt.gca().xaxis.set_major_formatter(plt.FuncFormatter(format_func)) plt.xlabel("Timesteps") plt.ylabel("Final Distance to Goal") plt.title("Visual Reacher Baselines") plt.legend([]) # plt.tight_layout() # plt.savefig(output_dir + "reacher_baselines.pdf") # print("File saved to", output_dir + "reacher_baselines.pdf") # plt.legend(
from rlkit.visualization import plot_util as plot configure_matplotlib(matplotlib) exps = plot.load_exps([ ashvin_base_dir + "s3doodad/share/reacher/reacher-abalation-resample-strategy" ], suppress_output=True) # plot.tag_exps(exps, "name", "oracle") plot.comparison( exps, "Final distance Mean", vary=[ "replay_kwargs.fraction_goals_are_env_goals", "replay_kwargs.fraction_goals_are_rollout_goals" ], # smooth=plot.padded_ma_filter(10), ylim=(0.04, 0.2), xlim=(0, 10000), method_order=[2, 1, 0, 3], figsize=(6, 4)) plt.gca().xaxis.set_major_formatter(plt.FuncFormatter(format_func)) plt.xlabel("Timesteps") plt.ylabel("Final Distance to Goal") plt.title("Visual Reacher") # plt.legend([]) # ["Ours", "No Relabeling", "HER", "VAE Only", ]) plt.legend([]) plt.tight_layout() plt.savefig(output_dir + "reacher_relabeling_ablation.pdf") print("File saved to", output_dir + "reacher_relabeling_ablation.pdf")
plot.tag_exps(ours, "name", "ours") # f = plot.filter_by_flat_params({'algorithm': 'Sparse-HER', 'reward_params.epsilon': 0.1}) her = plot.load_exps([ashvin_base_dir + "s3doodad/share/real-reacher/her"], suppress_output=True) plot.tag_exps(her, "name", "her") # f = plot.filter_by_flat_params({'algorithm': 'TDM'}) oracle = plot.load_exps( [ashvin_base_dir + "s3doodad/share/real-reacher/oracle"], suppress_output=True) plot.tag_exps(oracle, "name", "oracle") plot.comparison( ours + her + oracle, "Test Final End Effector Distance from Target Mean", vary=["name"], ylim=(0.0, 0.35), xlim=(0, 10000), method_order=[2, 0, 1], figsize=(6, 5), ) plt.gca().xaxis.set_major_formatter(plt.FuncFormatter(format_func)) plt.ylabel("Final Distance to Goal") plt.title("Real-World Visual Reacher") leg = plt.legend( [our_method_name, "HER", "Oracle"], bbox_to_anchor=(0.49, -0.2), loc="upper center", ncol=4, handlelength=1, ) leg.get_frame().set_alpha(0.9)
dirs = [ ashvin_base_dir + 's3doodad/ashvin/vae/fixed3/sawyer-pusher/vae-dense-multi3/run1', ] f = plot.filter_by_flat_params({ 'algo_kwargs.num_updates_per_env_step': 4, "replay_kwargs.fraction_goals_are_env_goals": 0.5 }) ours = plot.load_exps(dirs, f, suppress_output=True) plot.tag_exps(ours, "name", "ours") plot.comparison( ours + logprob + exps, "Final total_distance Mean", figsize=(7.5, 4), vary=["name"], default_vary={"reward_params.type": "unknown"}, smooth=plot.padded_ma_filter(10), xlim=(0, 250000), method_order=[1, 0, 2], ) plt.gca().xaxis.set_major_formatter(plt.FuncFormatter(format_func)) plt.xlabel("Timesteps") plt.ylabel("") plt.title("Visual Multi-object Pusher") plt.legend([our_method_name, "Log Prob.", "Pixel MSE", ], bbox_to_anchor=(1.0, 0.5), loc="center left",) plt.tight_layout() plt.savefig(output_dir + "multiobj_pusher_reward_type_ablation.pdf") print("File saved to", output_dir + "multiobj_pusher_reward_type_ablation.pdf")
from rlkit.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")
configure_matplotlib(matplotlib) dirs = [ ashvin_base_dir + 's3doodad/share/camera_ready_pick', ] pick_exps = plot.load_exps(dirs, suppress_output=True) plot.comparison( pick_exps, ["Final hand_and_obj_distance Mean"], [ # "seed", "exp_prefix", "train_vae_variant.vae_type", ], default_vary={"train_vae_variant.vae_type": True}, smooth=plot.padded_ma_filter(10), print_final=False, print_min=False, print_plot=True, xlim=(0, 500000), # ylim=(0, 0.35), figsize=(7.5, 4), method_order=(2, 1, 3, 4, 0), ) plt.gca().xaxis.set_major_formatter(plt.FuncFormatter(format_func)) plt.xlabel("Timesteps") plt.ylabel("") # plt.legend() plt.legend( [ "RIG",
colors = plt.rcParams['axes.prop_cycle'].by_key()['color'] label_to_color = { 'ours': colors[0], 'oracle': colors[3], 'her': colors[1], 'dsae': colors[2], 'l&r': colors[4], } plot.comparison( ours + oracle + dsae + lr + her, ["Final total_distance Mean"], vary=["name"], # default_vary={"replay_strategy": "future"}, smooth=plot.padded_ma_filter(10), # xlim=(0, 250000), xlim=(0, 500000), ylim=(0.15, 0.4), # figsize=(7, 3.5), figsize=(7.5, 4), method_order=[4, 0, 1, 3, 2], # label_to_color=label_to_color, ) plt.gca().xaxis.set_major_formatter(plt.FuncFormatter(format_func)) # leg.get_frame().set_alpha(0.9) plt.xlabel("Timesteps") plt.ylabel("") # "Final Distance to Goal") plt.title("Visual Multi-object Pusher Baselines") plt.legend( [ our_method_name, "DSAE",
0.0 }) norelabel = plot.load_exps(dirs, f, suppress_output=True) dirs = [ ashvin_base_dir + 's3doodad/ashvin/vae/fixed3/sawyer-pusher/vae-dense-multi3-fullrelabel/run1', ] fullrelabel = plot.load_exps(dirs, suppress_output=True) plot.comparison( her + fullrelabel + norelabel, "Final total_distance Mean", [ "replay_kwargs.fraction_goals_are_rollout_goals", "replay_kwargs.fraction_goals_are_env_goals", ], # ["training_mode", "replay_kwargs.fraction_goals_are_env_goals", "replay_kwargs.fraction_goals_are_rollout_goals", "rdim"], default_vary={"replay_strategy": "future"}, smooth=plot.padded_ma_filter(10), figsize=(7.5, 4), xlim=(0, 500000), ylim=(0.15, 0.35), method_order=[1, 2, 0, 3]) plt.gca().xaxis.set_major_formatter(plt.FuncFormatter(format_func)) plt.ylabel("") plt.xlabel("Timesteps") plt.title("Visual Multi-object Pusher") leg = plt.legend( [ our_method_name, "None", "Future",