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)
plt.gca().xaxis.set_major_formatter(plt.FuncFormatter(format_func))
plt.xlabel("Timesteps")
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)
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),
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'
})
her = plot.load_exps(
[ashvin_base_dir + "s3doodad/share/reward-reaching-sweep"],
f,
import matplotlib.pyplot as plt
from railrl.visualization import plot_util as plot
########## PUSHER PLOT
dirs = [
'/home/ashvin/data/rail-khazatsky/sasha/cond-rig/hyp-tuning/tuning/batch_size/',
]
#f = plot.filter_by_flat_params({'grill_variant.algo_kwargs.batch_size': 128,
# 'grill_variant.algo_kwargs.num_trains_per_train_loop': 1000})
normal = plot.load_exps(dirs,
suppress_output=True,
progress_filename="tensorboard_log.csv")
#f = plot.filter_by_flat_params({'grill_variant.algo_kwargs.num_trains_per_train_loop': 4000})
#4000_updates = plot.load_exps(dirs, f, suppress_output=True, progress_filename="tensorboard_log.csv")
#f = plot.filter_by_flat_params({'grill_variant.algo_kwargs.batch_size': 1024})
#batch_1024 = plot.load_exps(dirs, f, suppress_output=True, progress_filename="tensorboard_log.csv")
plot.comparison(
normal,
[
"evaluation/env_infos/final/current_object_distance_Mean",
], # "AverageReturn", "Final puck_distance Mean", "Final hand_and_puck_distance Mean"],
# [
# 'train_vae_variant.algo_kwargs.batch_size',
# 'train_vae_variant.latent_sizes'
# ],
[
import matplotlib
from visualization.grill.config import (
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)
# f = plot.filter_by_flat_params({'algorithm': 'Ours'})
ours = plot.load_exps([ashvin_base_dir + "s3doodad/share/real-reacher/ours"],
suppress_output=True)
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"],
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(
# [ashvin_base_dir + 's3doodad/share/steven/pushing-multipushing/multipusher-reward-variants-spatial'],
# plot.filter_by_flat_params({'training_mode': 'test'}),
import matplotlib
from visualization.grill.config import (
output_dir,
format_func,
configure_matplotlib,
)
import matplotlib.pyplot as plt
from railrl.visualization import plot_util as plot
configure_matplotlib(matplotlib)
dirs = [
'/home/khazatsky/rail/data/rail-khazatsky/sasha/cond-rig/pointmass/baseline',
]
VAE = plot.load_exps(dirs, suppress_output=True)
plot.tag_exps(VAE, "name", "VAE")
dirs = [
'/home/khazatsky/rail/data/rail-khazatsky/sasha/cond-rig/pointmass/standard',
]
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")
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":
0.5,
"replay_kwargs.fraction_goals_are_rollout_goals":
0.2
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)
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))
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/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")
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':
1.0,
'replay_kwargs.fraction_goals_are_env_goals':
0.0
})
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",
figsize=(6, 4),
# figsize=(7.5, 4),