Logic/diffusion_policy
1
0
Fork 0
Code Issues Pull Requests Actions Packages Projects Releases Wiki Activity

Merge pull request #7 from columbia-ai-robotics/cchi/bug_fix_eval_sample

Browse Source 
fixed bug where only n_envs samples of metrics are used
This commit is contained in:
Cheng Chi
2023-06-01 11:09:50 -04:00
committed by GitHub
parent 27395b7500 5e36d50603
commit 74b6391737
6 changed files with 54 additions and 6 deletions
Download Patch File Download Diff File Expand all files Collapse all files

10
diffusion_policy/env_runner/blockpush_lowdim_runner.py
View File

@@ -235,7 +235,15 @@ class BlockPushLowdimRunner(BaseLowdimRunner):
prefix_counts = collections.defaultdict(lambda : 0)
log_data = dict()
for i in range(len(self.env_fns)):
# results reported in the paper are generated using the commented out line below
# which will only report and average metrics from first n_envs initial condition and seeds
# fortunately this won't invalidate our conclusion since
# 1. This bug only affects the variance of metrics, not their mean
# 2. All baseline methods are evaluated using the same code
# to completely reproduce reported numbers, uncomment this line:
# for i in range(len(self.env_fns)):
# and comment out this line
for i in range(n_inits):
seed = self.env_seeds[i]
prefix = self.env_prefixs[i]
this_rewards = all_rewards[i]

10
diffusion_policy/env_runner/kitchen_lowdim_runner.py
View File

@@ -279,7 +279,15 @@ class KitchenLowdimRunner(BaseLowdimRunner):
log_data = dict()
prefix_total_reward_map = collections.defaultdict(list)
prefix_n_completed_map = collections.defaultdict(list)
for i in range(len(self.env_fns)):
# results reported in the paper are generated using the commented out line below
# which will only report and average metrics from first n_envs initial condition and seeds
# fortunately this won't invalidate our conclusion since
# 1. This bug only affects the variance of metrics, not their mean
# 2. All baseline methods are evaluated using the same code
# to completely reproduce reported numbers, uncomment this line:
# for i in range(len(self.env_fns)):
# and comment out this line
for i in range(n_inits):
seed = self.env_seeds[i]
prefix = self.env_prefixs[i]
this_rewards = all_rewards[i]

10
diffusion_policy/env_runner/pusht_image_runner.py
View File

@@ -221,7 +221,15 @@ class PushTImageRunner(BaseImageRunner):
# log
max_rewards = collections.defaultdict(list)
log_data = dict()
for i in range(len(self.env_fns)):
# results reported in the paper are generated using the commented out line below
# which will only report and average metrics from first n_envs initial condition and seeds
# fortunately this won't invalidate our conclusion since
# 1. This bug only affects the variance of metrics, not their mean
# 2. All baseline methods are evaluated using the same code
# to completely reproduce reported numbers, uncomment this line:
# for i in range(len(self.env_fns)):
# and comment out this line
for i in range(n_inits):
seed = self.env_seeds[i]
prefix = self.env_prefixs[i]
max_reward = np.max(all_rewards[i])

10
diffusion_policy/env_runner/pusht_keypoints_runner.py
View File

@@ -243,7 +243,15 @@ class PushTKeypointsRunner(BaseLowdimRunner):
# log
max_rewards = collections.defaultdict(list)
log_data = dict()
for i in range(len(self.env_fns)):
# results reported in the paper are generated using the commented out line below
# which will only report and average metrics from first n_envs initial condition and seeds
# fortunately this won't invalidate our conclusion since
# 1. This bug only affects the variance of metrics, not their mean
# 2. All baseline methods are evaluated using the same code
# to completely reproduce reported numbers, uncomment this line:
# for i in range(len(self.env_fns)):
# and comment out this line
for i in range(n_inits):
seed = self.env_seeds[i]
prefix = self.env_prefixs[i]
max_reward = np.max(all_rewards[i])

10
diffusion_policy/env_runner/robomimic_image_runner.py
View File

@@ -324,7 +324,15 @@ class RobomimicImageRunner(BaseImageRunner):
# log
max_rewards = collections.defaultdict(list)
log_data = dict()
for i in range(len(self.env_fns)):
# results reported in the paper are generated using the commented out line below
# which will only report and average metrics from first n_envs initial condition and seeds
# fortunately this won't invalidate our conclusion since
# 1. This bug only affects the variance of metrics, not their mean
# 2. All baseline methods are evaluated using the same code
# to completely reproduce reported numbers, uncomment this line:
# for i in range(len(self.env_fns)):
# and comment out this line
for i in range(n_inits):
seed = self.env_seeds[i]
prefix = self.env_prefixs[i]
max_reward = np.max(all_rewards[i])

10
diffusion_policy/env_runner/robomimic_lowdim_runner.py
View File

@@ -317,7 +317,15 @@ class RobomimicLowdimRunner(BaseLowdimRunner):
# log
max_rewards = collections.defaultdict(list)
log_data = dict()
for i in range(len(self.env_fns)):
# results reported in the paper are generated using the commented out line below
# which will only report and average metrics from first n_envs initial condition and seeds
# fortunately this won't invalidate our conclusion since
# 1. This bug only affects the variance of metrics, not their mean
# 2. All baseline methods are evaluated using the same code
# to completely reproduce reported numbers, uncomment this line:
# for i in range(len(self.env_fns)):
# and comment out this line
for i in range(n_inits):
seed = self.env_seeds[i]
prefix = self.env_prefixs[i]
max_reward = np.max(all_rewards[i])