import matplotlib from visualization.grill.config import ( output_dir, ashvin_base_dir, format_func, configure_matplotlib, ) import matplotlib.pyplot as plt 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)
) import matplotlib.pyplot as plt from railrl.visualization import plot_util as plot configure_matplotlib(matplotlib) dirs = [ ashvin_base_dir + 's3doodad/ashvin/vae/fixed3/sawyer-pusher/vae-dense-wider3/run1', ashvin_base_dir + 's3doodad/ashvin/vae/fixed3/sawyer-pusher/vae-dense-wider3-relabeling/run1', ] 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, 250000), ylim=(0.14, 0.24),
import matplotlib from visualization.grill.config import ( output_dir, ashvin_base_dir, vitchyr_base_dir, format_func, configure_matplotlib, ) import matplotlib.pyplot as plt from railrl.visualization import plot_util as plot configure_matplotlib(matplotlib) f = plot.filter_by_flat_params({ 'algo_kwargs.num_updates_per_env_step': 4, 'replay_kwargs.fraction_goals_are_env_goals': 0.5 }) oracle = plot.load_exps( [ashvin_base_dir + "s3doodad/share/reacher/reacher-baseline-oracle"], suppress_output=True) plot.tag_exps(oracle, "name", "oracle") ours = plot.load_exps( [ashvin_base_dir + "s3doodad/share/reacher/reacher-main-results-ours"], suppress_output=True) plot.tag_exps(ours, "name", "ours") f = plot.filter_by_flat_params({ 'replay_kwargs.fraction_goals_are_env_goals': 0.0, 'reward_params.type': 'latent_distance'
] f = plot.exclude_by_flat_params({'grill_variant.vae_path': '/home/khazatsky/rail/data/rail-khazatsky/sasha/PCVAE/standard/run1002/id0/vae.pkl',}) CVAE = plot.load_exps(dirs, f, suppress_output=True) plot.tag_exps(CVAE, "name", "CVAE") dirs = [ '/home/khazatsky/rail/data/rail-khazatsky/sasha/cond-rig/pointmass/dynamics/run12', ] CDVAE = plot.load_exps(dirs, suppress_output=True) plot.tag_exps(CDVAE, "name", "Dynamics CVAE") dirs = [ '/home/khazatsky/rail/data/rail-khazatsky/sasha/cond-rig/pointmass/adversarial/run10/', ] f = plot.filter_by_flat_params({'grill_variant.vae_path': '/home/khazatsky/rail/data/rail-khazatsky/sasha/PCVAE/ACE/run1000/id1/vae.pkl',}) ACE = plot.load_exps(dirs, f, suppress_output=True) plot.tag_exps(ACE, "name", "Adversarially Conditioned VAE") dirs = [ '/home/khazatsky/rail/data/rail-khazatsky/sasha/cond-rig/pointmass/CADVAE/run10/', ] CADVAE = plot.load_exps(dirs, suppress_output=True) plot.tag_exps(CADVAE, "name", "Adversarially Conditioned Dynamics VAE") dirs = [ '/home/khazatsky/rail/data/rail-khazatsky/sasha/cond-rig/pointmass/standard/', ] f = plot.filter_by_flat_params({'grill_variant.vae_path': '/home/khazatsky/rail/data/rail-khazatsky/sasha/PCVAE/standard/run1001/id0/vae.pkl',}) DCVAE = plot.load_exps(dirs, f, suppress_output=True)
output_dir, ashvin_base_dir, format_func, our_method_name, configure_matplotlib, ) import matplotlib.pyplot as plt from railrl.visualization import plot_util as plot configure_matplotlib(matplotlib) 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") dirs = [ ashvin_base_dir + 's3doodad/ashvin/vae/fixed3/sawyer-pusher/state-dense-multi1/run1', ] f = plot.filter_by_flat_params({ 'replay_kwargs.fraction_goals_are_env_goals': 0.5, 'algo_kwargs.reward_scale': 1e-4 }) oracle = plot.load_exps(dirs, f, suppress_output=True) plot.tag_exps(oracle, "name", "oracle") # dsae = plot.load_exps(
vitchyr_base_dir, format_func, configure_matplotlib, ) import matplotlib.pyplot as plt from railrl.visualization import plot_util as plot configure_matplotlib(matplotlib) dirs = [ ashvin_base_dir + 's3doodad/ashvin/vae/fixed3/sawyer-pusher/state-dense-wider2/run1', ] f = plot.filter_by_flat_params({ 'algo_kwargs.num_updates_per_env_step': 4, 'replay_kwargs.fraction_goals_are_env_goals': 0.0 }) oracle = plot.load_exps(dirs, f, suppress_output=True) plot.tag_exps(oracle, "name", "oracle") dirs = [ ashvin_base_dir + 's3doodad/ashvin/vae/fixed3/sawyer-pusher/vae-dense-wider3/run1', ashvin_base_dir + 's3doodad/ashvin/vae/fixed3/sawyer-pusher/vae-dense-wider3-relabeling/run1', ] f = plot.filter_by_flat_params({ 'algo_kwargs.num_updates_per_env_step': 4, "replay_kwargs.fraction_goals_are_env_goals":
our_method_name, configure_matplotlib, ) import matplotlib.pyplot as plt from railrl.visualization import plot_util as plot configure_matplotlib(matplotlib) 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, 'rdim': 16, 'replay_kwargs.fraction_goals_are_rollout_goals': 0.2 }) her = plot.load_exps(dirs, f, suppress_output=True) 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, 'rdim': 16, 'replay_kwargs.fraction_goals_are_rollout_goals':
import matplotlib from visualization.grill.config import ( output_dir, ashvin_base_dir, configure_matplotlib, ) import matplotlib.pyplot as plt from railrl.visualization import plot_util as plot configure_matplotlib(matplotlib) f = plot.filter_by_flat_params({ 'grill_variant.algo_kwargs.rl_offpolicy_num_training_steps': 10000, }) ours = plot.load_exps([ ashvin_base_dir + "s3doodad/ashvin/corl2019/offpolicy/rig-dcvae-offpolicy2-sweep1/run3", ashvin_base_dir + "s3doodad/ashvin/corl2019/offpolicy/rig-dcvae-offpolicy2-sweep1/run4", ashvin_base_dir + "s3doodad/ashvin/corl2019/offpolicy/rig-dcvae-offpolicy2-sweep1/run5", ], f, suppress_output=True) plot.tag_exps(ours, "name", "off-policy") baseline = plot.load_exps([ashvin_base_dir + "s3doodad/ashvin/corl2019/offpolicy/rig-dcvae-offpolicy2-baseline", ], suppress_output=True) plot.tag_exps(baseline, "name", "on-policy") plot.comparison( ours + baseline, "evaluation/distance_to_target Mean", vary=["name"], # xlabel="evaluation/num steps total",