Browse Source

Merge pull request 'update benchmark' (#2) from add_done into master

master
Xingyuan 3 years ago
parent
commit
373caf7b74
24 changed files with 1025 additions and 215 deletions
  1. +6
    -0
      .gitmodules
  2. +201
    -0
      LICENSE
  3. +14
    -7
      README.md
  4. +1
    -0
      benchmark/OfflineRL
  5. +34
    -0
      benchmark/README.md
  6. +1
    -0
      benchmark/d3pe
  7. +8
    -0
      benchmark/download_datasets.py
  8. +40
    -0
      benchmark/export_bc.py
  9. +30
    -0
      benchmark/launch_algo.py
  10. +119
    -0
      benchmark/launch_ope.py
  11. +156
    -0
      benchmark/launch_task.py
  12. +138
    -0
      benchmark/pretrain_dynamics.py
  13. +1
    -1
      demo.py
  14. +10
    -2
      neorl/__init__.py
  15. +16
    -4
      neorl/core.py
  16. +142
    -142
      neorl/data_map.json
  17. +2
    -2
      neorl/md5_util.py
  18. +43
    -0
      neorl/neorl_envs/Hopper-v3/Hopper-v3_done.py
  19. +43
    -0
      neorl/neorl_envs/Walker2d-v3/Walker2d-v3_done.py
  20. +2
    -2
      neorl/neorl_envs/ib/ib_reward.py
  21. +5
    -47
      neorl/neorl_envs/ib/industrial_benchmark_python/IBGym.py
  22. +4
    -0
      neorl/neorl_envs/mujoco.py
  23. +4
    -4
      neorl/tests/test_envdata.py
  24. +5
    -4
      neorl/utils.py

+ 6
- 0
.gitmodules View File

@ -0,0 +1,6 @@
[submodule "benchmark/OfflineRL"]
path = benchmark/OfflineRL
url = https://agit.ai/Polixir/OfflineRL.git
[submodule "benchmark/d3pe"]
path = benchmark/d3pe
url = https://agit.ai/Polixir/d3pe.git

+ 201
- 0
LICENSE View File

@ -0,0 +1,201 @@
Apache License
Version 2.0, January 2004
http://www.apache.org/licenses/
TERMS AND CONDITIONS FOR USE, REPRODUCTION, AND DISTRIBUTION
1. Definitions.
"License" shall mean the terms and conditions for use, reproduction,
and distribution as defined by Sections 1 through 9 of this document.
"Licensor" shall mean the copyright owner or entity authorized by
the copyright owner that is granting the License.
"Legal Entity" shall mean the union of the acting entity and all
other entities that control, are controlled by, or are under common
control with that entity. For the purposes of this definition,
"control" means (i) the power, direct or indirect, to cause the
direction or management of such entity, whether by contract or
otherwise, or (ii) ownership of fifty percent (50%) or more of the
outstanding shares, or (iii) beneficial ownership of such entity.
"You" (or "Your") shall mean an individual or Legal Entity
exercising permissions granted by this License.
"Source" form shall mean the preferred form for making modifications,
including but not limited to software source code, documentation
source, and configuration files.
"Object" form shall mean any form resulting from mechanical
transformation or translation of a Source form, including but
not limited to compiled object code, generated documentation,
and conversions to other media types.
"Work" shall mean the work of authorship, whether in Source or
Object form, made available under the License, as indicated by a
copyright notice that is included in or attached to the work
(an example is provided in the Appendix below).
"Derivative Works" shall mean any work, whether in Source or Object
form, that is based on (or derived from) the Work and for which the
editorial revisions, annotations, elaborations, or other modifications
represent, as a whole, an original work of authorship. For the purposes
of this License, Derivative Works shall not include works that remain
separable from, or merely link (or bind by name) to the interfaces of,
the Work and Derivative Works thereof.
"Contribution" shall mean any work of authorship, including
the original version of the Work and any modifications or additions
to that Work or Derivative Works thereof, that is intentionally
submitted to Licensor for inclusion in the Work by the copyright owner
or by an individual or Legal Entity authorized to submit on behalf of
the copyright owner. For the purposes of this definition, "submitted"
means any form of electronic, verbal, or written communication sent
to the Licensor or its representatives, including but not limited to
communication on electronic mailing lists, source code control systems,
and issue tracking systems that are managed by, or on behalf of, the
Licensor for the purpose of discussing and improving the Work, but
excluding communication that is conspicuously marked or otherwise
designated in writing by the copyright owner as "Not a Contribution."
"Contributor" shall mean Licensor and any individual or Legal Entity
on behalf of whom a Contribution has been received by Licensor and
subsequently incorporated within the Work.
2. Grant of Copyright License. Subject to the terms and conditions of
this License, each Contributor hereby grants to You a perpetual,
worldwide, non-exclusive, no-charge, royalty-free, irrevocable
copyright license to reproduce, prepare Derivative Works of,
publicly display, publicly perform, sublicense, and distribute the
Work and such Derivative Works in Source or Object form.
3. Grant of Patent License. Subject to the terms and conditions of
this License, each Contributor hereby grants to You a perpetual,
worldwide, non-exclusive, no-charge, royalty-free, irrevocable
(except as stated in this section) patent license to make, have made,
use, offer to sell, sell, import, and otherwise transfer the Work,
where such license applies only to those patent claims licensable
by such Contributor that are necessarily infringed by their
Contribution(s) alone or by combination of their Contribution(s)
with the Work to which such Contribution(s) was submitted. If You
institute patent litigation against any entity (including a
cross-claim or counterclaim in a lawsuit) alleging that the Work
or a Contribution incorporated within the Work constitutes direct
or contributory patent infringement, then any patent licenses
granted to You under this License for that Work shall terminate
as of the date such litigation is filed.
4. Redistribution. You may reproduce and distribute copies of the
Work or Derivative Works thereof in any medium, with or without
modifications, and in Source or Object form, provided that You
meet the following conditions:
(a) You must give any other recipients of the Work or
Derivative Works a copy of this License; and
(b) You must cause any modified files to carry prominent notices
stating that You changed the files; and
(c) You must retain, in the Source form of any Derivative Works
that You distribute, all copyright, patent, trademark, and
attribution notices from the Source form of the Work,
excluding those notices that do not pertain to any part of
the Derivative Works; and
(d) If the Work includes a "NOTICE" text file as part of its
distribution, then any Derivative Works that You distribute must
include a readable copy of the attribution notices contained
within such NOTICE file, excluding those notices that do not
pertain to any part of the Derivative Works, in at least one
of the following places: within a NOTICE text file distributed
as part of the Derivative Works; within the Source form or
documentation, if provided along with the Derivative Works; or,
within a display generated by the Derivative Works, if and
wherever such third-party notices normally appear. The contents
of the NOTICE file are for informational purposes only and
do not modify the License. You may add Your own attribution
notices within Derivative Works that You distribute, alongside
or as an addendum to the NOTICE text from the Work, provided
that such additional attribution notices cannot be construed
as modifying the License.
You may add Your own copyright statement to Your modifications and
may provide additional or different license terms and conditions
for use, reproduction, or distribution of Your modifications, or
for any such Derivative Works as a whole, provided Your use,
reproduction, and distribution of the Work otherwise complies with
the conditions stated in this License.
5. Submission of Contributions. Unless You explicitly state otherwise,
any Contribution intentionally submitted for inclusion in the Work
by You to the Licensor shall be under the terms and conditions of
this License, without any additional terms or conditions.
Notwithstanding the above, nothing herein shall supersede or modify
the terms of any separate license agreement you may have executed
with Licensor regarding such Contributions.
6. Trademarks. This License does not grant permission to use the trade
names, trademarks, service marks, or product names of the Licensor,
except as required for reasonable and customary use in describing the
origin of the Work and reproducing the content of the NOTICE file.
7. Disclaimer of Warranty. Unless required by applicable law or
agreed to in writing, Licensor provides the Work (and each
Contributor provides its Contributions) on an "AS IS" BASIS,
WITHOUT WARRANTIES OR CONDITIONS OF ANY KIND, either express or
implied, including, without limitation, any warranties or conditions
of TITLE, NON-INFRINGEMENT, MERCHANTABILITY, or FITNESS FOR A
PARTICULAR PURPOSE. You are solely responsible for determining the
appropriateness of using or redistributing the Work and assume any
risks associated with Your exercise of permissions under this License.
8. Limitation of Liability. In no event and under no legal theory,
whether in tort (including negligence), contract, or otherwise,
unless required by applicable law (such as deliberate and grossly
negligent acts) or agreed to in writing, shall any Contributor be
liable to You for damages, including any direct, indirect, special,
incidental, or consequential damages of any character arising as a
result of this License or out of the use or inability to use the
Work (including but not limited to damages for loss of goodwill,
work stoppage, computer failure or malfunction, or any and all
other commercial damages or losses), even if such Contributor
has been advised of the possibility of such damages.
9. Accepting Warranty or Additional Liability. While redistributing
the Work or Derivative Works thereof, You may choose to offer,
and charge a fee for, acceptance of support, warranty, indemnity,
or other liability obligations and/or rights consistent with this
License. However, in accepting such obligations, You may act only
on Your own behalf and on Your sole responsibility, not on behalf
of any other Contributor, and only if You agree to indemnify,
defend, and hold each Contributor harmless for any liability
incurred by, or claims asserted against, such Contributor by reason
of your accepting any such warranty or additional liability.
END OF TERMS AND CONDITIONS
APPENDIX: How to apply the Apache License to your work.
To apply the Apache License to your work, attach the following
boilerplate notice, with the fields enclosed by brackets "[]"
replaced with your own identifying information. (Don't include
the brackets!) The text should be enclosed in the appropriate
comment syntax for the file format. We also recommend that a
file or class name and description of purpose be included on the
same "printed page" as the copyright notice for easier
identification within third-party archives.
Copyright [yyyy] [name of copyright owner]
Licensed under the Apache License, Version 2.0 (the "License");
you may not use this file except in compliance with the License.
You may obtain a copy of the License at
http://www.apache.org/licenses/LICENSE-2.0
Unless required by applicable law or agreed to in writing, software
distributed under the License is distributed on an "AS IS" BASIS,
WITHOUT WARRANTIES OR CONDITIONS OF ANY KIND, either express or implied.
See the License for the specific language governing permissions and
limitations under the License.

+ 14
- 7
README.md View File

@ -1,12 +1,16 @@
# NeoRL
[![License](https://img.shields.io/badge/License-Apache%202.0-blue.svg)](https://opensource.org/licenses/Apache-2.0)
[![License](https://licensebuttons.net/l/by/3.0/88x31.png)](https://creativecommons.org/licenses/by/4.0/)
This repository is the interface for the offline reinforcement learning benchmark NeoRL: NEar real-world benchmarks for Offline Reinforcement Learning.
This repository is the interface for the offline reinforcement learning benchmark NeoRL: A Near Real-World Benchmark for Offline Reinforcement Learning.
The NeoRL benchmarks contains environments, datasets, and reward functions for training and benchmarking offline reinforcement learning algorithms. Current benchmarks contains environments of CityLearn, FinRL, IB, and three MuJoCo tasks.
The NeoRL benchmark contains environments, datasets, and reward functions for training and benchmarking offline reinforcement learning algorithms. Current benchmark contains environments of CityLearn, FinRL, IB, and three Gym-MuJoCo tasks.
More about the NeoRL benchmarks can be found at http://polixir.ai/research/neorl and the following paper
More about the NeoRL benchmark can be found at http://polixir.ai/research/neorl and the following paper
> Rongjun Qin, Songyi Gao, Xingyuan Zhang, Zhen Xu, Shengkai Huang, Zewen Li, Weinan Zhang, Yang Yu. Near Real-World Benchmarks for Offline Reinforcement Learning. https://arxiv.org/abs/2102.00714
> Rongjun Qin, Songyi Gao, Xingyuan Zhang, Zhen Xu, Shengkai Huang, Zewen Li, Weinan Zhang, Yang Yu. NeoRL: A Near Real-World Benchmark for Offline Reinforcement Learning. https://arxiv.org/abs/2102.00714
The benchmark is supported by two addtional repos, i.e. [OfflineRL](https://agit.ai/Polixir/OfflineRL) for training offline RL algorithms and [d3pe](https://agit.ai/Polixir/d3pe) for offline evaluation. Details for reproducing the benchmark can be found at [here](benchmark/).
## Install NeoRL interface
@ -38,8 +42,8 @@ env = neorl.make("citylearn")
env.reset()
env.step(env.action_space.sample())
# Get 99 trajectories of low level policy collection on citylearn task
train_data, val_data = env.get_dataset(data_type = "low", train_num = 99)
# Get 100 trajectories of low level policy collection on citylearn task
train_data, val_data = env.get_dataset(data_type = "low", train_num = 100)
```
To facilitate setting different goals, users can provide custom reward function to `neorl.make()` while creating an env. See [usage and examples of `neorl.make()`](https://agit.ai/Polixir/neorl/wiki/Parameter-usage#user-content-usage-of-neorl-make) for more details.
@ -48,7 +52,7 @@ As a benchmark, in order to test algorithms conveniently and quickly, each task
with a small training dataset and a validation dataset by default. They can be obtained by
`env.get_dataset()`. Meanwhile, for flexibility, extra parameters can be passed into `get_dataset()`
to get multiple pairs of datasets for benchmarking. Each task collects data using a low, medium,
or high level policy; for each task, we provide training data for a maximum of 9999 trajectories.
or high level policy; for each task, we provide training data for a maximum of 10000 trajectories.
See [usage of `get_dataset()`](https://agit.ai/Polixir/neorl/wiki/Parameter-usage#user-content-usage-of-get-dataset) for more details about parameter usage.
## Data in NeoRL
@ -75,3 +79,6 @@ In NeoRL, training data and validation data returned by `get_dataset()` function
- **FinRL**: Liu X Y, Yang H, Chen Q, et al. "FinRL: A Deep Reinforcement Learning Library for Automated Stock Trading in Quantitative Finance." arXiv preprint arXiv:2011.09607, 2020. [paper](https://arxiv.org/abs/2011.09607) [code](https://github.com/AI4Finance-LLC/FinRL-Library)
- **Industrial Benchmark**: Hein D, Depeweg S, Tokic M, et al. "A Benchmark Environment Motivated by Industrial Control Problems." Proceedings of the 2017 IEEE Symposium Series on Computational Intelligence, pp. 1-8, 2017. [paper](https://arxiv.org/abs/1709.09480) [code](https://github.com/siemens/industrialbenchmark)
- **MuJoCo**: Todorov E, Erez T, Tassa Y. "Mujoco: A Physics Engine for Model-based Control." Proceedings of the 2012 IEEE/RSJ International Conference on Intelligent Robots and Systems, pp. 5026-5033, 2012. [paper](https://ieeexplore.ieee.org/abstract/document/6386109) [website](https://gym.openai.com/envs/#mujoco)
## Licenses
All datasets are licensed under the [Creative Commons Attribution 4.0 License (CC BY)](https://creativecommons.org/licenses/by/4.0/), and code is licensed under the [Apache 2.0 License](https://www.apache.org/licenses/LICENSE-2.0.html).

+ 1
- 0
benchmark/OfflineRL

@ -0,0 +1 @@
Subproject commit 807933a87f77529f17bd81ac64d717aad89f5cdf

+ 34
- 0
benchmark/README.md View File

@ -0,0 +1,34 @@
## Instruction for reproducing the benchmark
Before you start, make sure you have followed the instruction [here](../README.md) to install NeoRL. All the results of the benchmark will be stored in the `results` folder.
### Step 1: Install dependencies
```bash
git submodule init
git submodule update
cd OfflineRL/
pip install -e .
mkdir offlinerl_tmp
cd offlinerl_tmp
mkdir offlinerl_datasets
aim init
cd ../..
cd d3pe/
pip install -e .
cd ..
```
### Step 2: Download the datasets
Download all the datasets before training the algorithms by `python download_datasets.py`.
### Step 3: Train policies by Offline RL Algorithms
You can use `launch_algo.py` for this part. The script will automatically launch an algorithm for 51 tasks provided in NeoRL, and training them in parallel based on `ray`. For example, you can use `python launch_algo.py --algo bc` to launch benchmark for BC algorithm. The trained policies will be stored in `OfflineRL/offlinerl_tmp/.aim`.
Note that, to speed up the benchmark, for model-based algorithms, i.e. BREMEN and MOPO, we used pretrained dynamic models. These models can be obtained by `python pretrain_dynamics.py`.
For BREMEN, we also used the pretrained behavior policy obtained by BC algorithm. Thus, before run benchmark on BREMEN, make sure the BC benchmark is done, and export the policies by `python export_bc.py`.
### Step 4: Evaluate the trained policies by OPEs
You can use `launch_ope.py` for this part. The script will automatically launch an OPE algorithms to evaluate all the policies trained by the designated offline RL algorithm. For example, you can use `python launch_ope.py --algo bc --ope fqe` to launch FQE algorithm to evaluate all the policies trained by BC.

+ 1
- 0
benchmark/d3pe

@ -0,0 +1 @@
Subproject commit 7afe06a749d0ec50a05213eee91db12335432dcb

+ 8
- 0
benchmark/download_datasets.py View File

@ -0,0 +1,8 @@
import neorl
if __name__ == '__main__':
for domain in ['HalfCheetah-v3', 'Hopper-v3', 'Walker2d-v3', 'ib', 'finance', 'citylearn']:
for level in ['low', 'medium', 'high']:
for amount in [100, 1000, 10000] if not domain == 'finance' else [100, 1000]:
env = neorl.make(domain)
env.get_dataset(data_type=level, train_num=amount)

+ 40
- 0
benchmark/export_bc.py View File

@ -0,0 +1,40 @@
import os
import json
import offlinerl
aim_folder = os.path.abspath(os.path.join(os.path.dirname(offlinerl.__file__), '..', 'offlinerl_tmp', '.aim'))
target_folder = 'behaviors'
if not os.path.exists(target_folder): os.makedirs(target_folder)
for task_name in sorted(filter(lambda x: 'bc' in x, os.listdir(aim_folder))):
task_folder = os.path.join(aim_folder, task_name)
exp_names = list(filter(lambda x: not x == 'index' and not 'json' in x, os.listdir(task_folder)))
split_name = task_name.split('-')
if len(split_name) == 5:
domain = split_name[0] + '-' + split_name[1]
level = split_name[2]
amount = int(split_name[3])
else:
domain = split_name[0]
level = split_name[1]
amount = int(split_name[2])
for exp_name in exp_names:
exp_folder = os.path.join(task_folder, exp_name)
with open(os.path.join(exp_folder, 'objects', 'map', 'dictionary.log'), 'r') as f:
data = json.load(f)
seed = data['hparams']['seed']
with open(os.path.join(exp_folder, 'metric_logs.json'), 'r') as f:
metrics = json.load(f)
max_step = max([int(name.split('.')[0]) for name in os.listdir(os.path.join(exp_folder, 'models'))])
policy_file = os.path.join(exp_folder, 'models', f'{max_step}.pt')
target_policy_file = os.path.join(target_folder, f'{domain}-{level}-{amount}-{seed}.pt')
os.system(f"cp {policy_file} {target_policy_file}")

+ 30
- 0
benchmark/launch_algo.py View File

@ -0,0 +1,30 @@
import os
import time
import argparse
ResultDir = os.path.abspath(os.path.join(os.path.dirname(__file__), 'results'))
def check_file(domain, level, amount, algo):
''' check if the result is already exist '''
json_file = f'{domain}-{level}-{amount},{algo}.json'
return json_file in os.listdir(ResultDir)
if __name__ == '__main__':
parser = argparse.ArgumentParser()
parser.add_argument('--algo', type=str, help='select from `bc`, `bcq`, `plas`, `cql`, `crr`, `bremen`, `mopo`')
parser.add_argument('--address', type=str, default=None, help='address of the ray cluster')
args = parser.parse_args()
if not os.path.exists(ResultDir): os.makedirs(ResultDir)
''' run a single algorithm on all the tasks '''
tasks = []
for domain in ['HalfCheetah-v3', 'Hopper-v3', 'Walker2d-v3', 'ib', 'finance', 'citylearn']:
for level in ['low', 'medium', 'high']:
for amount in [100, 1000, 10000] if not domain == 'finance' else [100, 1000]:
if not check_file(domain, level, amount, args.algo):
if args.address is not None:
os.system(f'python launch_task.py --domain {domain} --level {level} --amount {amount} --algo {args.algo} --address {args.address}')
else:
os.system(f'python launch_task.py --domain {domain} --level {level} --amount {amount} --algo {args.algo}')
time.sleep(20) # wait ray to release the resource

+ 119
- 0
benchmark/launch_ope.py View File

@ -0,0 +1,119 @@
import os
import ray
import json
import time
import torch
import argparse
from ray import tune
import neorl
import offlinerl
from offlinerl.utils.exp import setup_seed
from d3pe.evaluator.fqe import FQEEvaluator
from d3pe.evaluator.IS import ISEvaluator
from d3pe.utils.data import get_neorl_datasets
ResultDir = os.path.abspath(os.path.join(os.path.dirname(__file__), 'results'))
aim_folder = os.path.abspath(os.path.join(os.path.dirname(offlinerl.__file__), '..', 'offlinerl_tmp', '.aim'))
SEEDS = [7, 42, 210]
def check_file(domain, level, amount, algo, ope):
''' check if the result is already exist '''
json_file = f'{domain}-{level}-{amount},{algo},{ope}.json'
return json_file in os.listdir(ResultDir)
def launch_ope(config):
''' run on a seed '''
setup_seed(config['seed'])
if config['ope'] == 'fqe':
evaluator = FQEEvaluator()
elif config['ope'] == 'is':
evaluator = ISEvaluator()
train_dataset, val_dataset = get_neorl_datasets(config["domain"], config['level'], config['amount'])
evaluator.initialize(train_dataset=train_dataset, val_dataset=val_dataset)
exp_folder = os.path.join(config['task_folder'], config['exp_name'])
with open(os.path.join(exp_folder, 'metric_logs.json'), 'r') as f:
metrics = json.load(f)
max_step = str(max(map(int, metrics.keys())))
gt = metrics[max_step]['Reward_Mean_Env']
policy_file = os.path.join(exp_folder, 'models', f'{max_step}.pt')
policy = torch.load(policy_file)
ope = evaluator(policy)
return {
'gt' : gt,
'ope' : ope,
'policy_file' : policy_file,
'exp_name' : config['exp_name'],
'seed' : config['seed'],
}
if __name__ == '__main__':
parser = argparse.ArgumentParser()
parser.add_argument('--algo', type=str, help='select from `bc`, `bcq`, `plas`, `cql`, `crr`, `bremen`, `mopo`')
parser.add_argument('--ope', type=str, help='select from `fqe` and `is`')
parser.add_argument('--address', type=str, default=None, help='address of the ray cluster')
args = parser.parse_args()
ray.init(address=args.address)
for task_name in sorted(filter(lambda x: args.algo in x, os.listdir(aim_folder))):
task_folder = os.path.join(aim_folder, task_name)
exp_names = list(filter(lambda x: not x == 'index' and not 'json' in x, os.listdir(task_folder)))
split_name = task_name.split('-')
if len(split_name) == 5:
domain = split_name[0] + '-' + split_name[1]
level = split_name[2]
amount = int(split_name[3])
else:
domain = split_name[0]
level = split_name[1]
amount = int(split_name[2])
if (not args.overwrite) and check_file(domain, level, amount, args.algo, args.ope): continue
config = {
'seed' : tune.grid_search(SEEDS),
'ope' : args.ope,
'domain' : domain,
'level' : level,
'amount' : amount,
'task_folder' : task_folder,
'exp_name' : tune.grid_search(exp_names),
}
analysis = tune.run(
launch_ope,
name=f'{domain}-{level}-{amount}-{args.algo}-{args.ope}',
config=config,
queue_trials=True,
metric='ope',
mode='max',
resources_per_trial={
"cpu": 1,
"gpu": 1.0,
}
)
''' process result '''
df = analysis.results_df
results = {seed : {} for seed in SEEDS}
for i in range(len(df)):
results[df['seed'][i]][df['exp_name'][i]] = {'gt' : df['gt'][i], 'ope' : df['ope'][i]}
local_file = os.path.join(task_folder, f'{args.ope}.json')
remote_file = os.path.join(ResultDir, f'{domain}-{level}-{amount},{args.algo},{args.ope}.json')
with open(local_file, 'w') as f:
json.dump(results, f, indent=4)
os.system(f'cp {local_file} {remote_file}')
time.sleep(20) # wait ray to release the resource

+ 156
- 0
benchmark/launch_task.py View File

@ -0,0 +1,156 @@
import os
import ray
import json
import time
import argparse
import numpy as np
from ray import tune
from offlinerl.algo import algo_select
from offlinerl.data import load_data_from_neorl
from offlinerl.evaluation import OnlineCallBackFunction, PeriodicCallBack
SEEDS = [7, 42, 210]
ResultDir = os.path.abspath(os.path.join(os.path.dirname(__file__), 'results'))
def training_function(config):
''' run on a seed '''
config["kwargs"]['seed'] = config['seed']
algo_init_fn, algo_trainer_obj, algo_config = algo_select(config["kwargs"])
train_buffer, val_buffer = load_data_from_neorl(algo_config["task"], algo_config["task_data_type"], algo_config["task_train_num"])
algo_config.update(config)
algo_config["device"] = "cuda"
algo_config['dynamics_path'] = os.path.join(config['dynamics_root'],
f'{algo_config["task"]}-{algo_config["task_data_type"]}-{algo_config["task_train_num"]}-{config["seed"]}.pt')
algo_config['behavior_path'] = os.path.join(config['behavior_root'],
f'{algo_config["task"]}-{algo_config["task_data_type"]}-{algo_config["task_train_num"]}-{config["seed"]}.pt')
algo_init = algo_init_fn(algo_config)
algo_trainer = algo_trainer_obj(algo_init, algo_config)
callback = PeriodicCallBack(OnlineCallBackFunction(), 50)
callback.initialize(train_buffer=train_buffer, val_buffer=val_buffer, task=algo_config["task"], number_of_runs=1000)
algo_trainer.train(train_buffer, val_buffer, callback_fn=callback)
algo_trainer.exp_logger.flush()
time.sleep(10) # sleep ensure the log is flushed even if the disks or cpus are busy
result, parameter = find_result(algo_trainer.index_path)
return {
'reward' : result,
'parameter' : parameter,
'seed' : config['seed'],
}
def upload_result(task_name : str, algo_name : str, results : list):
''' upload the result '''
# upload txt
file_name = task_name + ',' + algo_name + '.txt'
reward_means = [result['reward_mean'] for result in results]
max_reward_mean = max(reward_means)
best_index = reward_means.index(max_reward_mean)
best_result = results[best_index]
with open(os.path.join(ResultDir, file_name), 'w') as f:
f.write(str(best_result['reward_mean']) + '+-' + str(best_result['reward_std']))
for k, v in best_result['parameter'].items():
f.write('\n')
f.write(f'{k} : {v}')
# upload json
file_name = task_name + ',' + algo_name + '.json'
with open(os.path.join(ResultDir, file_name), 'w') as f:
json.dump(results, f, indent=4)
def find_result(exp_dir : str):
''' return the online performance of last epoch and the hyperparameter '''
data_file = os.path.join(exp_dir, 'objects', 'map', 'dictionary.log')
with open(data_file, 'r') as f:
data = json.load(f)
result = data['__METRICS__']['Reward_Mean_Env'][0]['values']['last']
grid_search_keys = list(data['hparams']['grid_tune'].keys())
parameter = {k : data['hparams'][k] for k in grid_search_keys}
return result, parameter
if __name__ == '__main__':
parser = argparse.ArgumentParser()
parser.add_argument('--domain', type=str)
parser.add_argument('--level', type=str)
parser.add_argument('--amount', type=int)
parser.add_argument('--algo', type=str, help='select from `bc`, `bcq`, `plas`, `cql`, `crr`, `bremen`, `mopo`')
parser.add_argument('--address', type=str, default=None, help='address of the ray cluster')
args = parser.parse_args()
ray.init(args.address)
domain = args.domain
level = args.level
amount = args.amount
algo = args.algo
''' run and upload result '''
config = {}
config["kwargs"] = {
"exp_name" : f'{domain}-{level}-{amount}-{algo}',
"algo_name" : algo,
"task" : domain,
"task_data_type" : level,
"task_train_num" : amount,
}
_, _, algo_config = algo_select({"algo_name" : algo})
parameter_names = []
grid_tune = algo_config["grid_tune"]
for k, v in grid_tune.items():
parameter_names.append(k)
config[k] = tune.grid_search(v)
config['seed'] = tune.grid_search(SEEDS)
config['dynamics_root'] = os.path.abspath('dynamics')
config['behavior_root'] = os.path.abspath('behaviors')
analysis = tune.run(
training_function,
name=f'{domain}-{level}-{amount}-{algo}',
config=config,
queue_trials=True,
metric='reward',
mode='max',
resources_per_trial={
"cpu": 1,
"gpu": 1.0,
}
)
df = analysis.results_df
''' process result '''
results = {}
for i in range(len(df)):
parameter = {}
for pn in parameter_names:
parameter[pn] = df[f'parameter.{pn}'][i]
if type(parameter[pn]) == np.int64:
parameter[pn] = int(parameter[pn]) # covert to python type
parameter_string = str(parameter)
if not parameter_string in results:
results[parameter_string] = {
'parameter' : parameter,
'rewards' : [0, 0, 0],
}
results[parameter_string]['rewards'][SEEDS.index(df['seed'][i])] = df['reward'][i]
def summary_result(single_result):
single_result.update({
'reward_mean' : np.mean(single_result['rewards']),
'reward_std' : np.std(single_result['rewards']),
})
return single_result
results = [summary_result(single_result) for single_result in results.values()]
''' upload result '''
upload_result(f'{domain}-{level}-{amount}', algo, results)

+ 138
- 0
benchmark/pretrain_dynamics.py View File

@ -0,0 +1,138 @@
import os
import ray
import torch
import numpy as np
from ray import tune
from offlinerl.utils.exp import setup_seed
from offlinerl.data import load_data_from_neorl
from offlinerl.utils.net.model.ensemble import EnsembleTransition
SEEDS = [7, 42, 210]
def _select_best_indexes(metrics, n):
pairs = [(metric, index) for metric, index in zip(metrics, range(len(metrics)))]
pairs = sorted(pairs, key=lambda x: x[0])
selected_indexes = [pairs[i][1] for i in range(n)]
return selected_indexes
def _train_transition(transition, data, optim, device='cuda'):
data.to_torch(device=device)
dist = transition(torch.cat([data['obs'], data['act']], dim=-1))
loss = - dist.log_prob(torch.cat([data['obs_next'], data['rew']], dim=-1))
loss = loss.mean()
loss = loss + 0.01 * transition.max_logstd.mean() - 0.01 * transition.min_logstd.mean()
optim.zero_grad()
loss.backward()
optim.step()
def _eval_transition(transition, valdata, device='cuda'):
with torch.no_grad():
valdata.to_torch(device=device)
dist = transition(torch.cat([valdata['obs'], valdata['act']], dim=-1))
loss = ((dist.mean - torch.cat([valdata['obs_next'], valdata['rew']], dim=-1)) ** 2).mean(dim=(1,2))
return list(loss.cpu().numpy())
def training_dynamics(config):
if config["task"] == 'finance' and config["amount"] == 10000:
return {
'performance' : [],
'path' : '',
}
seed = config['seed']
setup_seed(seed)
train_buffer, val_buffer = load_data_from_neorl(config["task"], config["level"], config["amount"])
obs_shape = train_buffer['obs'].shape[-1]
action_shape = train_buffer['act'].shape[-1]
device = 'cuda'
hidden_units = 1024 if config["task"] in ['ib', 'finance', 'citylearn'] else 256
transition = EnsembleTransition(obs_shape, action_shape, hidden_units, 4, 7).to(device)
transition_optim = torch.optim.AdamW(transition.parameters(), lr=1e-3, weight_decay=0.000075)
data_size = len(train_buffer)
val_size = min(int(data_size * 0.2) + 1, 1000)
train_size = data_size - val_size
train_splits, val_splits = torch.utils.data.random_split(range(data_size), (train_size, val_size))
valdata = train_buffer[val_splits.indices]
train_buffer = train_buffer[train_splits.indices]
batch_size = 256
val_losses = [float('inf') for i in range(7)]
epoch = 0
cnt = 0
while True:
epoch += 1
idxs = np.random.randint(train_buffer.shape[0], size=[7, train_buffer.shape[0]])
for batch_num in range(int(np.ceil(idxs.shape[-1] / batch_size))):
batch_idxs = idxs[:, batch_num * batch_size:(batch_num + 1) * batch_size]
batch = train_buffer[batch_idxs]
_train_transition(transition, batch, transition_optim, device)
new_val_losses = _eval_transition(transition, valdata, device)
indexes = []
for i, new_loss, old_loss in zip(range(len(val_losses)), new_val_losses, val_losses):
if new_loss < old_loss:
indexes.append(i)
val_losses[i] = new_loss
if len(indexes) > 0:
transition.update_save(indexes)
cnt = 0
else:
cnt += 1
if cnt >= 5:
break
indexes = _select_best_indexes(val_losses, n=5)
transition.set_select(indexes)
performance = _eval_transition(transition, valdata, device)
transition_path = os.path.join(config['dynamics_path'], f'{config["task"]}-{config["level"]}-{config["amount"]}-{seed}.pt')
torch.save(transition, transition_path)
return {
'performance' : performance,
'path' : transition_path,
}
if __name__ == '__main__':
if not os.path.exists('dynamics'):
os.makedirs('dynamics')
ray.init()
abs_path = os.path.abspath('dynamics')
config = {}
config['task'] = tune.grid_search(['HalfCheetah-v3', 'Hopper-v3', 'Walker2d-v3', 'ib', 'finance', 'citylearn'])
config['level'] = tune.grid_search(['low', 'medium', 'high'])
config['amount'] = tune.grid_search([100, 1000, 10000])
config['seed'] = tune.grid_search(SEEDS)
config['dynamics_path'] = abs_path
analysis = tune.run(
training_dynamics,
name='dynamics',
config=config,
queue_trials=True,
resources_per_trial={
"cpu": 1,
"gpu": 1.0,
}
)
df = analysis.results_df
df.to_pickle(os.path.join(abs_path, 'summary.pkl'))

+ 1
- 1
demo.py View File

@ -2,7 +2,7 @@ import neorl
env = neorl.make("citylearn")
train_data, val_data = env.get_dataset(data_type="medium", train_num=99, need_val=True)
train_data, val_data = env.get_dataset(data_type="medium", train_num=100, need_val=True)
print("citylearn:", train_data, val_data)
reward_func = env.get_reward_func()


+ 10
- 2
neorl/__init__.py View File

@ -1,7 +1,7 @@
import importlib
def make(task: str, reward_func=None):
def make(task: str, reward_func=None, done_func=None):
try:
if task == "ib" or task == "Ib" or task == "industrial-benchmark" or task == "Industrial-Benchmark":
from neorl.neorl_envs.ib import ib_envs, get_env
@ -40,7 +40,15 @@ def make(task: str, reward_func=None):
default_reward_func = importlib.import_module(f"neorl.neorl_envs.{task}.{task}_reward").get_reward
except ModuleNotFoundError:
default_reward_func = None
env.set_reward_func(default_reward_func if reward_func is None else reward_func)
try:
default_done_func = importlib.import_module(f"neorl.neorl_envs.{task}.{task}_done").get_done
except ModuleNotFoundError:
default_done_func = None
env.set_done_func(default_done_func if done_func is None else done_func)
return env

+ 16
- 4
neorl/core.py View File

@ -3,7 +3,7 @@ from .utils import get_json, sample_dataset, LOCAL_JSON_FILE_PATH, DATA_PATH
class EnvData(gym.Env):
def get_dataset(self, task_name_version: str = None, data_type: str = "high", train_num: int = 99,
def get_dataset(self, task_name_version: str = None, data_type: str = "high", train_num: int = 100,
need_val: bool = True, val_ratio: float = 0.1, path: str = DATA_PATH, use_data_reward: bool = True):
"""
Get dataset from given env.
@ -13,7 +13,7 @@ class EnvData(gym.Env):
:param data_type: Which type of policy is used to collect data. It should
be one of ["high", "medium", "low"], default to `high`
:param train_num: The num of trajectory of training data. Note that the num
should be less than 10,000, default to `99`
should be less than 10,000, default to `100`
:param need_val: Whether needs to download validation data, default to `True`
:param val_ratio: The ratio of validation data to training data, default to `0.1`
:param path: The directory of data to load from or download to `./data/`
@ -53,13 +53,13 @@ class EnvData(gym.Env):
def set_reward_func(self, reward_func):
"""
Users can call this func to set customized reward.
Users can call this func to set customized reward func.
"""
self._reward_func = reward_func
def get_reward_func(self):
"""
Users can call this func to set customized reward.
Users can call this func to get customized reward func.
"""
return self._reward_func
@ -68,3 +68,15 @@ class EnvData(gym.Env):
Set name for envs.
"""
self._name = name
def set_done_func(self, done_func):
"""
Users can call this func to set done func.
"""
self._done_func = done_func
def get_done_func(self):
"""
Users can call this func to get done func.
"""
return self._done_func

+ 142
- 142
neorl/data_map.json View File

@ -1,144 +1,144 @@
{
"HalfCheetah-v3-expert-10-val-noise.npz": {"url": "http://datasets.polixir.site/HalfCheetah-v3/HalfCheetah-v3-expert-10-val-noise.npz", "md5": "efd6c418f1c2aa53a99150fad8005bfa"},
"HalfCheetah-v3-expert-100-val-noise.npz": {"url": "http://datasets.polixir.site/HalfCheetah-v3/HalfCheetah-v3-expert-100-val-noise.npz", "md5": "156537accc30926c4038e2e6720ac372"},
"HalfCheetah-v3-expert-1000-val-noise.npz": {"url": "http://datasets.polixir.site/HalfCheetah-v3/HalfCheetah-v3-expert-1000-val-noise.npz", "md5": "27b57ba06ec2de27b00159e8d469e98f"},
"HalfCheetah-v3-expert-99-train-noise.npz": {"url": "http://datasets.polixir.site/HalfCheetah-v3/HalfCheetah-v3-expert-99-train-noise.npz", "md5": "026f5bf1b4d0d43276d86cfbb596462a"},
"HalfCheetah-v3-expert-999-train-noise.npz": {"url": "http://datasets.polixir.site/HalfCheetah-v3/HalfCheetah-v3-expert-999-train-noise.npz", "md5": "599ec025a39d5c83db95cd94668cba73"},
"HalfCheetah-v3-expert-9999-train-noise.npz": {"url": "http://datasets.polixir.site/HalfCheetah-v3/HalfCheetah-v3-expert-9999-train-noise.npz", "md5": "34507aea46b6d5108179b23705652c09"},
"HalfCheetah-v3-high-10-val-noise.npz": {"url": "http://datasets.polixir.site/HalfCheetah-v3/HalfCheetah-v3-high-10-val-noise.npz", "md5": "bf0b14233334a46d8741960c5af4281a"},
"HalfCheetah-v3-high-100-val-noise.npz": {"url": "http://datasets.polixir.site/HalfCheetah-v3/HalfCheetah-v3-high-100-val-noise.npz", "md5": "041bf5d267d14e7c238776cf9d7a4402"},
"HalfCheetah-v3-high-1000-val-noise.npz": {"url": "http://datasets.polixir.site/HalfCheetah-v3/HalfCheetah-v3-high-1000-val-noise.npz", "md5": "b8703ac4e2182b4aaba6bf55d2a8d368"},
"HalfCheetah-v3-high-99-train-noise.npz": {"url": "http://datasets.polixir.site/HalfCheetah-v3/HalfCheetah-v3-high-99-train-noise.npz", "md5": "f5792dbe56f838be02dc76c824902e6f"},
"HalfCheetah-v3-high-999-train-noise.npz": {"url": "http://datasets.polixir.site/HalfCheetah-v3/HalfCheetah-v3-high-999-train-noise.npz", "md5": "a94518465dd714f2be828561522d8d3d"},
"HalfCheetah-v3-high-9999-train-noise.npz": {"url": "http://datasets.polixir.site/HalfCheetah-v3/HalfCheetah-v3-high-9999-train-noise.npz", "md5": "4a61afc2a4879342cf95e6fac3528474"},
"HalfCheetah-v3-low-10-val-noise.npz": {"url": "http://datasets.polixir.site/HalfCheetah-v3/HalfCheetah-v3-low-10-val-noise.npz", "md5": "5c114b0b8e73ea45b78511e608271b18"},
"HalfCheetah-v3-low-100-val-noise.npz": {"url": "http://datasets.polixir.site/HalfCheetah-v3/HalfCheetah-v3-low-100-val-noise.npz", "md5": "83853e69f81cca959eb315e399bd5df1"},
"HalfCheetah-v3-low-1000-val-noise.npz": {"url": "http://datasets.polixir.site/HalfCheetah-v3/HalfCheetah-v3-low-1000-val-noise.npz", "md5": "59509b399bdad9e0586a3a0690d45ac5"},
"HalfCheetah-v3-low-99-train-noise.npz": {"url": "http://datasets.polixir.site/HalfCheetah-v3/HalfCheetah-v3-low-99-train-noise.npz", "md5": "54cd79dc696c5eb626990db948508ba0"},
"HalfCheetah-v3-low-999-train-noise.npz": {"url": "http://datasets.polixir.site/HalfCheetah-v3/HalfCheetah-v3-low-999-train-noise.npz", "md5": "2e893193b6b8bd879eee5524dfc1a826"},
"HalfCheetah-v3-low-9999-train-noise.npz": {"url": "http://datasets.polixir.site/HalfCheetah-v3/HalfCheetah-v3-low-9999-train-noise.npz", "md5": "55fa767eb11b8edd8db035e841552495"},
"HalfCheetah-v3-medium-10-val-noise.npz": {"url": "http://datasets.polixir.site/HalfCheetah-v3/HalfCheetah-v3-medium-10-val-noise.npz", "md5": "2664eda2cdb71bea4030d02a59c3d150"},
"HalfCheetah-v3-medium-100-val-noise.npz": {"url": "http://datasets.polixir.site/HalfCheetah-v3/HalfCheetah-v3-medium-100-val-noise.npz", "md5": "d9cadeb60d5af33bea6998e2408441b5"},
"HalfCheetah-v3-medium-1000-val-noise.npz": {"url": "http://datasets.polixir.site/HalfCheetah-v3/HalfCheetah-v3-medium-1000-val-noise.npz", "md5": "62bde111637bfd481b4a1b249d4b63bf"},
"HalfCheetah-v3-medium-99-train-noise.npz": {"url": "http://datasets.polixir.site/HalfCheetah-v3/HalfCheetah-v3-medium-99-train-noise.npz", "md5": "224ff5ace1b921f3941894d53cbf9486"},
"HalfCheetah-v3-medium-999-train-noise.npz": {"url": "http://datasets.polixir.site/HalfCheetah-v3/HalfCheetah-v3-medium-999-train-noise.npz", "md5": "1a2471c9913759baf3642d76cf188890"},
"HalfCheetah-v3-medium-9999-train-noise.npz": {"url": "http://datasets.polixir.site/HalfCheetah-v3/HalfCheetah-v3-medium-9999-train-noise.npz", "md5": "21552b26e2e97d41a94fefd744fa9d90"},
"HalfCheetah-v3.yaml": {"url": "http://datasets.polixir.site/HalfCheetah-v3/HalfCheetah-v3.yaml", "md5": "56dc498a29083875b2c0517eb2ca8e9f"},
"HalfCheetah-v3_reward.py": {"url": "http://datasets.polixir.site/HalfCheetah-v3/HalfCheetah-v3_reward.py", "md5": "b5a6f5ff578b8d3d09ea46cab69a7b22"},
"Hopper-v3-expert-10-val-noise.npz": {"url": "http://datasets.polixir.site/Hopper-v3/Hopper-v3-expert-10-val-noise.npz", "md5": "82df84d20ab9d245b2963beb15efc05a"},
"Hopper-v3-expert-100-val-noise.npz": {"url": "http://datasets.polixir.site/Hopper-v3/Hopper-v3-expert-100-val-noise.npz", "md5": "aba50230c21dac0fc284924e3f12b229"},
"Hopper-v3-expert-1000-val-noise.npz": {"url": "http://datasets.polixir.site/Hopper-v3/Hopper-v3-expert-1000-val-noise.npz", "md5": "4df321cbae2d6931e722678ec5b3c7d3"},
"Hopper-v3-expert-99-train-noise.npz": {"url": "http://datasets.polixir.site/Hopper-v3/Hopper-v3-expert-99-train-noise.npz", "md5": "b8f4e311223b8aaaae4759773fedb02f"},
"Hopper-v3-expert-999-train-noise.npz": {"url": "http://datasets.polixir.site/Hopper-v3/Hopper-v3-expert-999-train-noise.npz", "md5": "14896a72bf01b57a32592c510e7cdd53"},
"Hopper-v3-expert-9999-train-noise.npz": {"url": "http://datasets.polixir.site/Hopper-v3/Hopper-v3-expert-9999-train-noise.npz", "md5": "82ce02d8ad6307b14c04cc17aa63440c"},
"Hopper-v3-high-10-val-noise.npz": {"url": "http://datasets.polixir.site/Hopper-v3/Hopper-v3-high-10-val-noise.npz", "md5": "0d3373f4c1021a837a0b335d56f4cd3b"},
"Hopper-v3-high-100-val-noise.npz": {"url": "http://datasets.polixir.site/Hopper-v3/Hopper-v3-high-100-val-noise.npz", "md5": "246295e60cc797ef184e7952a836502a"},
"Hopper-v3-high-1000-val-noise.npz": {"url": "http://datasets.polixir.site/Hopper-v3/Hopper-v3-high-1000-val-noise.npz", "md5": "3a295e49b2568ee946d05e7eefdc9337"},
"Hopper-v3-high-99-train-noise.npz": {"url": "http://datasets.polixir.site/Hopper-v3/Hopper-v3-high-99-train-noise.npz", "md5": "1480a6e517bd1ae5e947782ea75a4e55"},
"Hopper-v3-high-999-train-noise.npz": {"url": "http://datasets.polixir.site/Hopper-v3/Hopper-v3-high-999-train-noise.npz", "md5": "240a4b9cc3ae97a1969311c628c6cce9"},
"Hopper-v3-high-9999-train-noise.npz": {"url": "http://datasets.polixir.site/Hopper-v3/Hopper-v3-high-9999-train-noise.npz", "md5": "45ab80512b7b288e4ad73bc142b1d9fb"},
"Hopper-v3-low-10-val-noise.npz": {"url": "http://datasets.polixir.site/Hopper-v3/Hopper-v3-low-10-val-noise.npz", "md5": "47240e2ec6e077d592d34f3749ee734e"},
"Hopper-v3-low-100-val-noise.npz": {"url": "http://datasets.polixir.site/Hopper-v3/Hopper-v3-low-100-val-noise.npz", "md5": "5343d5dd355c3fc912ddfb58f3ba0e6e"},
"Hopper-v3-low-1000-val-noise.npz": {"url": "http://datasets.polixir.site/Hopper-v3/Hopper-v3-low-1000-val-noise.npz", "md5": "d843ca00665e4624b30e7fa235e627e3"},
"Hopper-v3-low-99-train-noise.npz": {"url": "http://datasets.polixir.site/Hopper-v3/Hopper-v3-low-99-train-noise.npz", "md5": "2670a41d517aa45a6e69b154d18743dd"},
"Hopper-v3-low-999-train-noise.npz": {"url": "http://datasets.polixir.site/Hopper-v3/Hopper-v3-low-999-train-noise.npz", "md5": "b7d71de1f8a34750aa3833707c20c515"},
"Hopper-v3-low-9999-train-noise.npz": {"url": "http://datasets.polixir.site/Hopper-v3/Hopper-v3-low-9999-train-noise.npz", "md5": "cb51dc8c229336fded170efd7bb95ed3"},
"Hopper-v3-medium-10-val-noise.npz": {"url": "http://datasets.polixir.site/Hopper-v3/Hopper-v3-medium-10-val-noise.npz", "md5": "264ffb9bafc7781a90ac3de733f8ca04"},
"Hopper-v3-medium-100-val-noise.npz": {"url": "http://datasets.polixir.site/Hopper-v3/Hopper-v3-medium-100-val-noise.npz", "md5": "cebff154828a047a5b76af176e4c8dce"},
"Hopper-v3-medium-1000-val-noise.npz": {"url": "http://datasets.polixir.site/Hopper-v3/Hopper-v3-medium-1000-val-noise.npz", "md5": "d0147c65b58c165b09634ddd33171ef1"},
"Hopper-v3-medium-99-train-noise.npz": {"url": "http://datasets.polixir.site/Hopper-v3/Hopper-v3-medium-99-train-noise.npz", "md5": "517b9f083a36ce7aab300657335d84fe"},
"Hopper-v3-medium-999-train-noise.npz": {"url": "http://datasets.polixir.site/Hopper-v3/Hopper-v3-medium-999-train-noise.npz", "md5": "c85b0f4ec328cc9cd0752f4c88fc2696"},
"Hopper-v3-medium-9999-train-noise.npz": {"url": "http://datasets.polixir.site/Hopper-v3/Hopper-v3-medium-9999-train-noise.npz", "md5": "360bdf34c5f546bce7905c4b200b3c23"},
"Hopper-v3.yaml": {"url": "http://datasets.polixir.site/Hopper-v3/Hopper-v3.yaml", "md5": "f5c1806ff8d6b664b9f60b718c4acbb5"},
"Hopper-v3_reward.py": {"url": "http://datasets.polixir.site/Hopper-v3/Hopper-v3_reward.py", "md5": "afb3081431da107993adb6d30c9c63c5"},
"Walker2d-v3-expert-10-val-noise.npz": {"url": "http://datasets.polixir.site/Walker2d-v3/Walker2d-v3-expert-10-val-noise.npz", "md5": "14c9601454105187e7edcbd8ba594859"},
"Walker2d-v3-expert-100-val-noise.npz": {"url": "http://datasets.polixir.site/Walker2d-v3/Walker2d-v3-expert-100-val-noise.npz", "md5": "4195a1745d749d2bc62aabf127545eba"},
"Walker2d-v3-expert-1000-val-noise.npz": {"url": "http://datasets.polixir.site/Walker2d-v3/Walker2d-v3-expert-1000-val-noise.npz", "md5": "580a224f2a8fb6322245b21877311c6f"},
"Walker2d-v3-expert-99-train-noise.npz": {"url": "http://datasets.polixir.site/Walker2d-v3/Walker2d-v3-expert-99-train-noise.npz", "md5": "cc7077d55d3d39c9dc9bf33762fe530c"},
"Walker2d-v3-expert-999-train-noise.npz": {"url": "http://datasets.polixir.site/Walker2d-v3/Walker2d-v3-expert-999-train-noise.npz", "md5": "1595eb1f0306cabfea44d5e0a6b31646"},
"Walker2d-v3-expert-9999-train-noise.npz": {"url": "http://datasets.polixir.site/Walker2d-v3/Walker2d-v3-expert-9999-train-noise.npz", "md5": "ce95b88dfee371da602c0cc787853fe0"},
"Walker2d-v3-high-10-val-noise.npz": {"url": "http://datasets.polixir.site/Walker2d-v3/Walker2d-v3-high-10-val-noise.npz", "md5": "7f493ceb445c8bd5f4d3dbe04f5b5321"},
"Walker2d-v3-high-100-val-noise.npz": {"url": "http://datasets.polixir.site/Walker2d-v3/Walker2d-v3-high-100-val-noise.npz", "md5": "eaa1561773fa5bb697239b4179f308ad"},
"Walker2d-v3-high-1000-val-noise.npz": {"url": "http://datasets.polixir.site/Walker2d-v3/Walker2d-v3-high-1000-val-noise.npz", "md5": "7068ab95830b98e81b9ced79608f9218"},
"Walker2d-v3-high-99-train-noise.npz": {"url": "http://datasets.polixir.site/Walker2d-v3/Walker2d-v3-high-99-train-noise.npz", "md5": "749620f874a61979323b1cd85d0c3e7b"},
"Walker2d-v3-high-999-train-noise.npz": {"url": "http://datasets.polixir.site/Walker2d-v3/Walker2d-v3-high-999-train-noise.npz", "md5": "f060c11b8a122921eb9ab3dd88697f05"},
"Walker2d-v3-high-9999-train-noise.npz": {"url": "http://datasets.polixir.site/Walker2d-v3/Walker2d-v3-high-9999-train-noise.npz", "md5": "0456716893e04f1c81a7bd3b58811c4e"},
"Walker2d-v3-low-10-val-noise.npz": {"url": "http://datasets.polixir.site/Walker2d-v3/Walker2d-v3-low-10-val-noise.npz", "md5": "49eec69ae5f2ae176073ff35cfdd429a"},
"Walker2d-v3-low-100-val-noise.npz": {"url": "http://datasets.polixir.site/Walker2d-v3/Walker2d-v3-low-100-val-noise.npz", "md5": "ebb4dd1b4313f546a7514c4f57b0e663"},
"Walker2d-v3-low-1000-val-noise.npz": {"url": "http://datasets.polixir.site/Walker2d-v3/Walker2d-v3-low-1000-val-noise.npz", "md5": "2d506bb17f8d061a894f37597a088e5e"},
"Walker2d-v3-low-99-train-noise.npz": {"url": "http://datasets.polixir.site/Walker2d-v3/Walker2d-v3-low-99-train-noise.npz", "md5": "ba94eaa74b43a44cc154d52c31dff6f5"},
"Walker2d-v3-low-999-train-noise.npz": {"url": "http://datasets.polixir.site/Walker2d-v3/Walker2d-v3-low-999-train-noise.npz", "md5": "c126614e044d5e16eabe1a1a052217bc"},
"Walker2d-v3-low-9999-train-noise.npz": {"url": "http://datasets.polixir.site/Walker2d-v3/Walker2d-v3-low-9999-train-noise.npz", "md5": "7b46fe5dfa9bbafb7cf9eb10c0d1b9cd"},
"Walker2d-v3-medium-10-val-noise.npz": {"url": "http://datasets.polixir.site/Walker2d-v3/Walker2d-v3-medium-10-val-noise.npz", "md5": "953c3336eae683d562881891442b618c"},
"Walker2d-v3-medium-100-val-noise.npz": {"url": "http://datasets.polixir.site/Walker2d-v3/Walker2d-v3-medium-100-val-noise.npz", "md5": "d0d902ec6c97550fd3c696d65277c867"},
"Walker2d-v3-medium-1000-val-noise.npz": {"url": "http://datasets.polixir.site/Walker2d-v3/Walker2d-v3-medium-1000-val-noise.npz", "md5": "4a71647e36989c9a64724f5e46987f20"},
"Walker2d-v3-medium-99-train-noise.npz": {"url": "http://datasets.polixir.site/Walker2d-v3/Walker2d-v3-medium-99-train-noise.npz", "md5": "38a6a8c9f1ee546e8e9aa41d5bfc8793"},
"Walker2d-v3-medium-999-train-noise.npz": {"url": "http://datasets.polixir.site/Walker2d-v3/Walker2d-v3-medium-999-train-noise.npz", "md5": "00ba0fdf89224006fb5ae2ec90142e0b"},
"Walker2d-v3-medium-9999-train-noise.npz": {"url": "http://datasets.polixir.site/Walker2d-v3/Walker2d-v3-medium-9999-train-noise.npz", "md5": "d563a30ba59d0de47f657c479a7f5301"},
"Walker2d-v3.yaml": {"url": "http://datasets.polixir.site/Walker2d-v3/Walker2d-v3.yaml", "md5": "3a8a7a19a96db3d94a56644f8dfbe6f8"},
"Walker2d-v3_reward.py": {"url": "http://datasets.polixir.site/Walker2d-v3/Walker2d-v3_reward.py", "md5": "66a63b9a685374464219ba0d92e2e4aa"},
"citylearn-expert-10-val-noise.npz": {"url": "http://datasets.polixir.site/citylearn/citylearn-expert-10-val-noise.npz", "md5": "76c4a9babad4695011fa5b2f1013b178"},
"citylearn-expert-100-val-noise.npz": {"url": "http://datasets.polixir.site/citylearn/citylearn-expert-100-val-noise.npz", "md5": "3ba1b46f058aae6fb497f7a1bcdfc440"},
"citylearn-expert-1000-val-noise.npz": {"url": "http://datasets.polixir.site/citylearn/citylearn-expert-1000-val-noise.npz", "md5": "e84cab7d8a97cd63a224313096ff2303"},
"citylearn-expert-99-train-noise.npz": {"url": "http://datasets.polixir.site/citylearn/citylearn-expert-99-train-noise.npz", "md5": "1dd8416f898a8b0724f90408ef67015a"},
"citylearn-expert-999-train-noise.npz": {"url": "http://datasets.polixir.site/citylearn/citylearn-expert-999-train-noise.npz", "md5": "acd37f0227d289f06f068bd2986067ec"},
"citylearn-expert-9999-train-noise.npz": {"url": "http://datasets.polixir.site/citylearn/citylearn-expert-9999-train-noise.npz", "md5": "7a34824df15083e0a75f4fca91ffb5d4"},
"citylearn-high-10-val-noise.npz": {"url": "http://datasets.polixir.site/citylearn/citylearn-high-10-val-noise.npz", "md5": "4ed4dbb1021695758388fd8861b3c5d9"},
"citylearn-high-100-val-noise.npz": {"url": "http://datasets.polixir.site/citylearn/citylearn-high-100-val-noise.npz", "md5": "979bcc8eebb17ebbf0c75e524065cc3a"},
"citylearn-high-1000-val-noise.npz": {"url": "http://datasets.polixir.site/citylearn/citylearn-high-1000-val-noise.npz", "md5": "36d61c8b8bc43f3e8ca92f1fe6f468cd"},
"citylearn-high-99-train-noise.npz": {"url": "http://datasets.polixir.site/citylearn/citylearn-high-99-train-noise.npz", "md5": "407238377d840a3d4683f7d4d8377ccc"},
"citylearn-high-999-train-noise.npz": {"url": "http://datasets.polixir.site/citylearn/citylearn-high-999-train-noise.npz", "md5": "191bf2388973a7c4742b79d411ed6e00"},
"citylearn-high-9999-train-noise.npz": {"url": "http://datasets.polixir.site/citylearn/citylearn-high-9999-train-noise.npz", "md5": "337f21bf033ccc9ded20257386dcf8a3"},
"citylearn-low-10-val-noise.npz": {"url": "http://datasets.polixir.site/citylearn/citylearn-low-10-val-noise.npz", "md5": "d52ad7284835dfa3fd467616b35c23c1"},
"citylearn-low-100-val-noise.npz": {"url": "http://datasets.polixir.site/citylearn/citylearn-low-100-val-noise.npz", "md5": "e15e34479c8eaa2fb9f14fafe592e0bb"},
"citylearn-low-1000-val-noise.npz": {"url": "http://datasets.polixir.site/citylearn/citylearn-low-1000-val-noise.npz", "md5": "c47352dac7f5243c1b2cbdaabc69f3b9"},
"citylearn-low-99-train-noise.npz": {"url": "http://datasets.polixir.site/citylearn/citylearn-low-99-train-noise.npz", "md5": "b59c92e298bd38057e8f7d3a33805fd6"},
"citylearn-low-999-train-noise.npz": {"url": "http://datasets.polixir.site/citylearn/citylearn-low-999-train-noise.npz", "md5": "b859b923d0007e522c234cfe23c01dfc"},
"citylearn-low-9999-train-noise.npz": {"url": "http://datasets.polixir.site/citylearn/citylearn-low-9999-train-noise.npz", "md5": "3176fd40f4eba3bb4f6696dc9ff74c94"},
"citylearn-medium-10-val-noise.npz": {"url": "http://datasets.polixir.site/citylearn/citylearn-medium-10-val-noise.npz", "md5": "346c9cbc8205102ec6d226680c6636b2"},
"citylearn-medium-100-val-noise.npz": {"url": "http://datasets.polixir.site/citylearn/citylearn-medium-100-val-noise.npz", "md5": "46c821efc87f6a1d273862d626f9bcbc"},
"citylearn-medium-1000-val-noise.npz": {"url": "http://datasets.polixir.site/citylearn/citylearn-medium-1000-val-noise.npz", "md5": "7555ec64d1f7f2587f7bc81de1114a46"},
"citylearn-medium-99-train-noise.npz": {"url": "http://datasets.polixir.site/citylearn/citylearn-medium-99-train-noise.npz", "md5": "77fbf8dbea04af6a7a4027cebbc39ad7"},
"citylearn-medium-999-train-noise.npz": {"url": "http://datasets.polixir.site/citylearn/citylearn-medium-999-train-noise.npz", "md5": "7f60dae0c6a0f3f6597f41605e4c04b2"},
"citylearn-medium-9999-train-noise.npz": {"url": "http://datasets.polixir.site/citylearn/citylearn-medium-9999-train-noise.npz", "md5": "d29ed7220b8ec79562f9289b873d4fa0"},
"citylearn.yaml": {"url": "http://datasets.polixir.site/citylearn/citylearn.yaml", "md5": "da63379d53ad05221f089afc5b44f673"},
"citylearn_reward.py": {"url": "http://datasets.polixir.site/citylearn/citylearn_reward.py", "md5": "d147bd875461e83132352291afcb4431"},
"finance-expert-10-val-noise.npz": {"url": "http://datasets.polixir.site/finance/finance-expert-10-val-noise.npz", "md5": "2f84951ba2082f64c76c18e409f3d798"},
"finance-expert-100-val-noise.npz": {"url": "http://datasets.polixir.site/finance/finance-expert-100-val-noise.npz", "md5": "4a25d252852997d19b632ff02c87cba2"},
"finance-expert-99-train-noise.npz": {"url": "http://datasets.polixir.site/finance/finance-expert-99-train-noise.npz", "md5": "179757ed62cff4fbecdad343c0ccae26"},
"finance-expert-999-train-noise.npz": {"url": "http://datasets.polixir.site/finance/finance-expert-999-train-noise.npz", "md5": "b49e2328726bc3b7af5fabe4e79972d5"},
"finance-high-10-val-noise.npz": {"url": "http://datasets.polixir.site/finance/finance-high-10-val-noise.npz", "md5": "75cef37a50575fb695d9d0ff67261d0d"},
"finance-high-100-val-noise.npz": {"url": "http://datasets.polixir.site/finance/finance-high-100-val-noise.npz", "md5": "c311c59401d2bd6b86292f196ae78070"},
"finance-high-99-train-noise.npz": {"url": "http://datasets.polixir.site/finance/finance-high-99-train-noise.npz", "md5": "80f2924b8a7cf80a2fd5ab1b37e6fca3"},
"finance-high-999-train-noise.npz": {"url": "http://datasets.polixir.site/finance/finance-high-999-train-noise.npz", "md5": "daba743be0479099296050f58a54935d"},
"finance-low-10-val-noise.npz": {"url": "http://datasets.polixir.site/finance/finance-low-10-val-noise.npz", "md5": "4833b468490219f4715b71d369446e4b"},
"finance-low-100-val-noise.npz": {"url": "http://datasets.polixir.site/finance/finance-low-100-val-noise.npz", "md5": "65bd1cf7b60a832488f84f0c57c40acd"},
"finance-low-99-train-noise.npz": {"url": "http://datasets.polixir.site/finance/finance-low-99-train-noise.npz", "md5": "1738478ac066e681e886bda70e9a0df8"},
"finance-low-999-train-noise.npz": {"url": "http://datasets.polixir.site/finance/finance-low-999-train-noise.npz", "md5": "1da1266a3154215a2c213e87e5af05ec"},
"finance-medium-10-val-noise.npz": {"url": "http://datasets.polixir.site/finance/finance-medium-10-val-noise.npz", "md5": "98f199b8725692dd06b030b89fbe7bd0"},
"finance-medium-100-val-noise.npz": {"url": "http://datasets.polixir.site/finance/finance-medium-100-val-noise.npz", "md5": "57e1e80de5b115873decf443269e3e40"},
"finance-medium-99-train-noise.npz": {"url": "http://datasets.polixir.site/finance/finance-medium-99-train-noise.npz", "md5": "587c00ab1810618477843107a9c80fb4"},
"finance-medium-999-train-noise.npz": {"url": "http://datasets.polixir.site/finance/finance-medium-999-train-noise.npz", "md5": "17e84ccab51401755aa4b81c1c6f0ff5"},
"finance.yaml": {"url": "http://datasets.polixir.site/finance/finance.yaml", "md5": "a7ea340f912a2ead1b7b48c1d8f9d140"},
"finance_reward.py": {"url": "http://datasets.polixir.site/finance/finance_reward.py", "md5": "9414a483d6bbf9236e9ff281db285f3e"},
"ib-high-10-val.npz": {"url": "http://datasets.polixir.site/ib/ib-high-10-val.npz", "md5": "e977c29243c522cd936aed151cfe5e3b"},
"ib-high-100-val.npz": {"url": "http://datasets.polixir.site/ib/ib-high-100-val.npz", "md5": "4bb1f5adea82668b85e90e78ffd78b54"},
"ib-high-1000-val.npz": {"url": "http://datasets.polixir.site/ib/ib-high-1000-val.npz", "md5": "8c0afcac59f88998e02065b1b9d0139b"},
"ib-high-99-train.npz": {"url": "http://datasets.polixir.site/ib/ib-high-99-train.npz", "md5": "34e992d4468c41073b7258c8f2520c6a"},
"ib-high-999-train.npz": {"url": "http://datasets.polixir.site/ib/ib-high-999-train.npz", "md5": "cee48994283d5fa53f10f01202db262c"},
"ib-high-9999-train.npz": {"url": "http://datasets.polixir.site/ib/ib-high-9999-train.npz", "md5": "e38804e2886a5baa21c3cdc7fa3d1f48"},
"ib-low-10-val.npz": {"url": "http://datasets.polixir.site/ib/ib-low-10-val.npz", "md5": "9d2003d9a5bd20ee605aedc2d3d0eab3"},
"ib-low-100-val.npz": {"url": "http://datasets.polixir.site/ib/ib-low-100-val.npz", "md5": "8c578828a8a4ffe61547923f79f9e56c"},
"ib-low-1000-val.npz": {"url": "http://datasets.polixir.site/ib/ib-low-1000-val.npz", "md5": "2d545068bbc12b2465c73bfd650421e6"},
"ib-low-99-train.npz": {"url": "http://datasets.polixir.site/ib/ib-low-99-train.npz", "md5": "3789fcf3b4cab98cf6f7a26f3d2c6806"},
"ib-low-999-train.npz": {"url": "http://datasets.polixir.site/ib/ib-low-999-train.npz", "md5": "66976872f72d70a397cd44bf69d1242a"},
"ib-low-9999-train.npz": {"url": "http://datasets.polixir.site/ib/ib-low-9999-train.npz", "md5": "7fabf8df45b3f4fb101ce388482d61ed"},
"ib-medium-10-val.npz": {"url": "http://datasets.polixir.site/ib/ib-medium-10-val.npz", "md5": "6a70e8b62bdf16774b2a5eba53b1132b"},
"ib-medium-100-val.npz": {"url": "http://datasets.polixir.site/ib/ib-medium-100-val.npz", "md5": "79f710e62756ed13ad578a6580405064"},
"ib-medium-1000-val.npz": {"url": "http://datasets.polixir.site/ib/ib-medium-1000-val.npz", "md5": "44128c3305b360b9a9b58c6a01ad248b"},
"ib-medium-99-train.npz": {"url": "http://datasets.polixir.site/ib/ib-medium-99-train.npz", "md5": "7102ca24e62938c2694c776ce905df2f"},
"ib-medium-999-train.npz": {"url": "http://datasets.polixir.site/ib/ib-medium-999-train.npz", "md5": "0b823e85af7376aedbb7c60f01c87d3a"},
"ib-medium-9999-train.npz": {"url": "http://datasets.polixir.site/ib/ib-medium-9999-train.npz", "md5": "32ffecd9a99a2444814de7b4f05bc09f"},
"ib.yaml": {"url": "http://datasets.polixir.site/ib/ib.yaml", "md5": "1ddfbf446e77fbe51ab2704c1632f32a"},
"ib_reward.py": {"url": "http://datasets.polixir.site/ib/ib_reward.py", "md5": "7d6b1693f8ad6d4daf31c14fbda19221"}
"HalfCheetah-v3-expert-10-val-noise.npz": {"url": "http://datasets.polixir.site/v2/HalfCheetah-v3/HalfCheetah-v3-expert-10-val-noise.npz", "md5": "efd6c418f1c2aa53a99150fad8005bfa"},
"HalfCheetah-v3-expert-100-val-noise.npz": {"url": "http://datasets.polixir.site/v2/HalfCheetah-v3/HalfCheetah-v3-expert-100-val-noise.npz", "md5": "156537accc30926c4038e2e6720ac372"},
"HalfCheetah-v3-expert-1000-val-noise.npz": {"url": "http://datasets.polixir.site/v2/HalfCheetah-v3/HalfCheetah-v3-expert-1000-val-noise.npz", "md5": "27b57ba06ec2de27b00159e8d469e98f"},
"HalfCheetah-v3-expert-100-train-noise.npz": {"url": "http://datasets.polixir.site/v2/HalfCheetah-v3/HalfCheetah-v3-expert-100-train-noise.npz", "md5": "25ad13692d66966d3446f0620098ed9f"},
"HalfCheetah-v3-expert-1000-train-noise.npz": {"url": "http://datasets.polixir.site/v2/HalfCheetah-v3/HalfCheetah-v3-expert-1000-train-noise.npz", "md5": "38862beecfa285086164fd8d76f0d3fc"},
"HalfCheetah-v3-expert-10000-train-noise.npz": {"url": "http://datasets.polixir.site/v2/HalfCheetah-v3/HalfCheetah-v3-expert-10000-train-noise.npz", "md5": "8e06d9c4dbb82c329895913ee9efff54"},
"HalfCheetah-v3-high-10-val-noise.npz": {"url": "http://datasets.polixir.site/v2/HalfCheetah-v3/HalfCheetah-v3-high-10-val-noise.npz", "md5": "bf0b14233334a46d8741960c5af4281a"},
"HalfCheetah-v3-high-100-val-noise.npz": {"url": "http://datasets.polixir.site/v2/HalfCheetah-v3/HalfCheetah-v3-high-100-val-noise.npz", "md5": "041bf5d267d14e7c238776cf9d7a4402"},
"HalfCheetah-v3-high-1000-val-noise.npz": {"url": "http://datasets.polixir.site/v2/HalfCheetah-v3/HalfCheetah-v3-high-1000-val-noise.npz", "md5": "b8703ac4e2182b4aaba6bf55d2a8d368"},
"HalfCheetah-v3-high-100-train-noise.npz": {"url": "http://datasets.polixir.site/v2/HalfCheetah-v3/HalfCheetah-v3-high-100-train-noise.npz", "md5": "f0e9fe09344d7ef04765ce1eec6cc18a"},
"HalfCheetah-v3-high-1000-train-noise.npz": {"url": "http://datasets.polixir.site/v2/HalfCheetah-v3/HalfCheetah-v3-high-1000-train-noise.npz", "md5": "e253a2613de6acbbb05ad0441e4949b8"},
"HalfCheetah-v3-high-10000-train-noise.npz": {"url": "http://datasets.polixir.site/v2/HalfCheetah-v3/HalfCheetah-v3-high-10000-train-noise.npz", "md5": "bd681966d9e2147461397cfb3692584d"},
"HalfCheetah-v3-low-10-val-noise.npz": {"url": "http://datasets.polixir.site/v2/HalfCheetah-v3/HalfCheetah-v3-low-10-val-noise.npz", "md5": "5c114b0b8e73ea45b78511e608271b18"},
"HalfCheetah-v3-low-100-val-noise.npz": {"url": "http://datasets.polixir.site/v2/HalfCheetah-v3/HalfCheetah-v3-low-100-val-noise.npz", "md5": "83853e69f81cca959eb315e399bd5df1"},
"HalfCheetah-v3-low-1000-val-noise.npz": {"url": "http://datasets.polixir.site/v2/HalfCheetah-v3/HalfCheetah-v3-low-1000-val-noise.npz", "md5": "59509b399bdad9e0586a3a0690d45ac5"},
"HalfCheetah-v3-low-100-train-noise.npz": {"url": "http://datasets.polixir.site/v2/HalfCheetah-v3/HalfCheetah-v3-low-100-train-noise.npz", "md5": "7df5c5fdf830263db1d7211efcdb17c1"},
"HalfCheetah-v3-low-1000-train-noise.npz": {"url": "http://datasets.polixir.site/v2/HalfCheetah-v3/HalfCheetah-v3-low-1000-train-noise.npz", "md5": "5afc938d022d5ae6afed25397d457e8d"},
"HalfCheetah-v3-low-10000-train-noise.npz": {"url": "http://datasets.polixir.site/v2/HalfCheetah-v3/HalfCheetah-v3-low-10000-train-noise.npz", "md5": "04b222566e3368b9555b8267ba110037"},
"HalfCheetah-v3-medium-10-val-noise.npz": {"url": "http://datasets.polixir.site/v2/HalfCheetah-v3/HalfCheetah-v3-medium-10-val-noise.npz", "md5": "2664eda2cdb71bea4030d02a59c3d150"},
"HalfCheetah-v3-medium-100-val-noise.npz": {"url": "http://datasets.polixir.site/v2/HalfCheetah-v3/HalfCheetah-v3-medium-100-val-noise.npz", "md5": "d9cadeb60d5af33bea6998e2408441b5"},
"HalfCheetah-v3-medium-1000-val-noise.npz": {"url": "http://datasets.polixir.site/v2/HalfCheetah-v3/HalfCheetah-v3-medium-1000-val-noise.npz", "md5": "62bde111637bfd481b4a1b249d4b63bf"},
"HalfCheetah-v3-medium-100-train-noise.npz": {"url": "http://datasets.polixir.site/v2/HalfCheetah-v3/HalfCheetah-v3-medium-100-train-noise.npz", "md5": "656f4475c29ec94b4b90cdaef5d08313"},
"HalfCheetah-v3-medium-1000-train-noise.npz": {"url": "http://datasets.polixir.site/v2/HalfCheetah-v3/HalfCheetah-v3-medium-1000-train-noise.npz", "md5": "b13bd54a0466f86455939a1175762a2a"},
"HalfCheetah-v3-medium-10000-train-noise.npz": {"url": "http://datasets.polixir.site/v2/HalfCheetah-v3/HalfCheetah-v3-medium-10000-train-noise.npz", "md5": "5d25fc95cf631947b18fcf879f0646fe"},
"HalfCheetah-v3.yaml": {"url": "http://datasets.polixir.site/v2/HalfCheetah-v3/HalfCheetah-v3.yaml", "md5": "56dc498a29083875b2c0517eb2ca8e9f"},
"HalfCheetah-v3_reward.py": {"url": "http://datasets.polixir.site/v2/HalfCheetah-v3/HalfCheetah-v3_reward.py", "md5": "b5a6f5ff578b8d3d09ea46cab69a7b22"},
"Hopper-v3-expert-10-val-noise.npz": {"url": "http://datasets.polixir.site/v2/Hopper-v3/Hopper-v3-expert-10-val-noise.npz", "md5": "82df84d20ab9d245b2963beb15efc05a"},
"Hopper-v3-expert-100-val-noise.npz": {"url": "http://datasets.polixir.site/v2/Hopper-v3/Hopper-v3-expert-100-val-noise.npz", "md5": "aba50230c21dac0fc284924e3f12b229"},
"Hopper-v3-expert-1000-val-noise.npz": {"url": "http://datasets.polixir.site/v2/Hopper-v3/Hopper-v3-expert-1000-val-noise.npz", "md5": "4df321cbae2d6931e722678ec5b3c7d3"},
"Hopper-v3-expert-100-train-noise.npz": {"url": "http://datasets.polixir.site/v2/Hopper-v3/Hopper-v3-expert-100-train-noise.npz", "md5": "774fd7b5124e199b337093adcdd101ee"},
"Hopper-v3-expert-1000-train-noise.npz": {"url": "http://datasets.polixir.site/v2/Hopper-v3/Hopper-v3-expert-1000-train-noise.npz", "md5": "0aa4670beff3bdc992229188109524aa"},
"Hopper-v3-expert-10000-train-noise.npz": {"url": "http://datasets.polixir.site/v2/Hopper-v3/Hopper-v3-expert-10000-train-noise.npz", "md5": "ccd909ab2f4932e2aa02c490b4b861a6"},
"Hopper-v3-high-10-val-noise.npz": {"url": "http://datasets.polixir.site/v2/Hopper-v3/Hopper-v3-high-10-val-noise.npz", "md5": "0d3373f4c1021a837a0b335d56f4cd3b"},
"Hopper-v3-high-100-val-noise.npz": {"url": "http://datasets.polixir.site/v2/Hopper-v3/Hopper-v3-high-100-val-noise.npz", "md5": "246295e60cc797ef184e7952a836502a"},
"Hopper-v3-high-1000-val-noise.npz": {"url": "http://datasets.polixir.site/v2/Hopper-v3/Hopper-v3-high-1000-val-noise.npz", "md5": "3a295e49b2568ee946d05e7eefdc9337"},
"Hopper-v3-high-100-train-noise.npz": {"url": "http://datasets.polixir.site/v2/Hopper-v3/Hopper-v3-high-100-train-noise.npz", "md5": "73fd09f821a822593dd7a0af72c334f4"},
"Hopper-v3-high-1000-train-noise.npz": {"url": "http://datasets.polixir.site/v2/Hopper-v3/Hopper-v3-high-1000-train-noise.npz", "md5": "91e711f30d4379cd627f63acfcfeb3c7"},
"Hopper-v3-high-10000-train-noise.npz": {"url": "http://datasets.polixir.site/v2/Hopper-v3/Hopper-v3-high-10000-train-noise.npz", "md5": "7da7428bc4d52bdc233b056554519414"},
"Hopper-v3-low-10-val-noise.npz": {"url": "http://datasets.polixir.site/v2/Hopper-v3/Hopper-v3-low-10-val-noise.npz", "md5": "47240e2ec6e077d592d34f3749ee734e"},
"Hopper-v3-low-100-val-noise.npz": {"url": "http://datasets.polixir.site/v2/Hopper-v3/Hopper-v3-low-100-val-noise.npz", "md5": "5343d5dd355c3fc912ddfb58f3ba0e6e"},
"Hopper-v3-low-1000-val-noise.npz": {"url": "http://datasets.polixir.site/v2/Hopper-v3/Hopper-v3-low-1000-val-noise.npz", "md5": "d843ca00665e4624b30e7fa235e627e3"},
"Hopper-v3-low-100-train-noise.npz": {"url": "http://datasets.polixir.site/v2/Hopper-v3/Hopper-v3-low-100-train-noise.npz", "md5": "c1c39051564790a1ac5c9f73160e2744"},
"Hopper-v3-low-1000-train-noise.npz": {"url": "http://datasets.polixir.site/v2/Hopper-v3/Hopper-v3-low-1000-train-noise.npz", "md5": "95a0c17014ba9fc45e739c820f3910ef"},
"Hopper-v3-low-10000-train-noise.npz": {"url": "http://datasets.polixir.site/v2/Hopper-v3/Hopper-v3-low-10000-train-noise.npz", "md5": "6cedfa7716b3fc34bd01f67fc67ea80e"},
"Hopper-v3-medium-10-val-noise.npz": {"url": "http://datasets.polixir.site/v2/Hopper-v3/Hopper-v3-medium-10-val-noise.npz", "md5": "264ffb9bafc7781a90ac3de733f8ca04"},
"Hopper-v3-medium-100-val-noise.npz": {"url": "http://datasets.polixir.site/v2/Hopper-v3/Hopper-v3-medium-100-val-noise.npz", "md5": "cebff154828a047a5b76af176e4c8dce"},
"Hopper-v3-medium-1000-val-noise.npz": {"url": "http://datasets.polixir.site/v2/Hopper-v3/Hopper-v3-medium-1000-val-noise.npz", "md5": "d0147c65b58c165b09634ddd33171ef1"},
"Hopper-v3-medium-100-train-noise.npz": {"url": "http://datasets.polixir.site/v2/Hopper-v3/Hopper-v3-medium-100-train-noise.npz", "md5": "46bf67196e1f42b1f345c1e15c20dd1f"},
"Hopper-v3-medium-1000-train-noise.npz": {"url": "http://datasets.polixir.site/v2/Hopper-v3/Hopper-v3-medium-1000-train-noise.npz", "md5": "c54404ee2f9b3eeb9168b35d48dca341"},
"Hopper-v3-medium-10000-train-noise.npz": {"url": "http://datasets.polixir.site/v2/Hopper-v3/Hopper-v3-medium-10000-train-noise.npz", "md5": "29809d063ff991d34ed5c07e845741d0"},
"Hopper-v3.yaml": {"url": "http://datasets.polixir.site/v2/Hopper-v3/Hopper-v3.yaml", "md5": "f5c1806ff8d6b664b9f60b718c4acbb5"},
"Hopper-v3_reward.py": {"url": "http://datasets.polixir.site/v2/Hopper-v3/Hopper-v3_reward.py", "md5": "afb3081431da107993adb6d30c9c63c5"},
"Walker2d-v3-expert-10-val-noise.npz": {"url": "http://datasets.polixir.site/v2/Walker2d-v3/Walker2d-v3-expert-10-val-noise.npz", "md5": "14c9601454105187e7edcbd8ba594859"},
"Walker2d-v3-expert-100-val-noise.npz": {"url": "http://datasets.polixir.site/v2/Walker2d-v3/Walker2d-v3-expert-100-val-noise.npz", "md5": "4195a1745d749d2bc62aabf127545eba"},
"Walker2d-v3-expert-1000-val-noise.npz": {"url": "http://datasets.polixir.site/v2/Walker2d-v3/Walker2d-v3-expert-1000-val-noise.npz", "md5": "580a224f2a8fb6322245b21877311c6f"},
"Walker2d-v3-expert-100-train-noise.npz": {"url": "http://datasets.polixir.site/v2/Walker2d-v3/Walker2d-v3-expert-100-train-noise.npz", "md5": "4a6ec4f023222cce5786d035b3771056"},
"Walker2d-v3-expert-1000-train-noise.npz": {"url": "http://datasets.polixir.site/v2/Walker2d-v3/Walker2d-v3-expert-1000-train-noise.npz", "md5": "58e8b8cbd2d0f21ed5fd26bc170f06de"},
"Walker2d-v3-expert-10000-train-noise.npz": {"url": "http://datasets.polixir.site/v2/Walker2d-v3/Walker2d-v3-expert-10000-train-noise.npz", "md5": "e2528ca748e63f3c56509833e00e08ec"},
"Walker2d-v3-high-10-val-noise.npz": {"url": "http://datasets.polixir.site/v2/Walker2d-v3/Walker2d-v3-high-10-val-noise.npz", "md5": "7f493ceb445c8bd5f4d3dbe04f5b5321"},
"Walker2d-v3-high-100-val-noise.npz": {"url": "http://datasets.polixir.site/v2/Walker2d-v3/Walker2d-v3-high-100-val-noise.npz", "md5": "eaa1561773fa5bb697239b4179f308ad"},
"Walker2d-v3-high-1000-val-noise.npz": {"url": "http://datasets.polixir.site/v2/Walker2d-v3/Walker2d-v3-high-1000-val-noise.npz", "md5": "7068ab95830b98e81b9ced79608f9218"},
"Walker2d-v3-high-100-train-noise.npz": {"url": "http://datasets.polixir.site/v2/Walker2d-v3/Walker2d-v3-high-100-train-noise.npz", "md5": "defc3ad1f4806179170a3baa4a4578e7"},
"Walker2d-v3-high-1000-train-noise.npz": {"url": "http://datasets.polixir.site/v2/Walker2d-v3/Walker2d-v3-high-1000-train-noise.npz", "md5": "ad15293c78a204fa20bb05e5206d0cb4"},
"Walker2d-v3-high-10000-train-noise.npz": {"url": "http://datasets.polixir.site/v2/Walker2d-v3/Walker2d-v3-high-10000-train-noise.npz", "md5": "f762736cb69fdf22fad8754f6bb8fedd"},
"Walker2d-v3-low-10-val-noise.npz": {"url": "http://datasets.polixir.site/v2/Walker2d-v3/Walker2d-v3-low-10-val-noise.npz", "md5": "49eec69ae5f2ae176073ff35cfdd429a"},
"Walker2d-v3-low-100-val-noise.npz": {"url": "http://datasets.polixir.site/v2/Walker2d-v3/Walker2d-v3-low-100-val-noise.npz", "md5": "ebb4dd1b4313f546a7514c4f57b0e663"},
"Walker2d-v3-low-1000-val-noise.npz": {"url": "http://datasets.polixir.site/v2/Walker2d-v3/Walker2d-v3-low-1000-val-noise.npz", "md5": "2d506bb17f8d061a894f37597a088e5e"},
"Walker2d-v3-low-100-train-noise.npz": {"url": "http://datasets.polixir.site/v2/Walker2d-v3/Walker2d-v3-low-100-train-noise.npz", "md5": "85562d37c73143619e0081b8cb6734c3"},
"Walker2d-v3-low-1000-train-noise.npz": {"url": "http://datasets.polixir.site/v2/Walker2d-v3/Walker2d-v3-low-1000-train-noise.npz", "md5": "3e88129538b30400b5b2b81d7f5a0044"},
"Walker2d-v3-low-10000-train-noise.npz": {"url": "http://datasets.polixir.site/v2/Walker2d-v3/Walker2d-v3-low-10000-train-noise.npz", "md5": "4185d0e47f620e07a888f9a74ae2ba5e"},
"Walker2d-v3-medium-10-val-noise.npz": {"url": "http://datasets.polixir.site/v2/Walker2d-v3/Walker2d-v3-medium-10-val-noise.npz", "md5": "953c3336eae683d562881891442b618c"},
"Walker2d-v3-medium-100-val-noise.npz": {"url": "http://datasets.polixir.site/v2/Walker2d-v3/Walker2d-v3-medium-100-val-noise.npz", "md5": "d0d902ec6c97550fd3c696d65277c867"},
"Walker2d-v3-medium-1000-val-noise.npz": {"url": "http://datasets.polixir.site/v2/Walker2d-v3/Walker2d-v3-medium-1000-val-noise.npz", "md5": "4a71647e36989c9a64724f5e46987f20"},
"Walker2d-v3-medium-100-train-noise.npz": {"url": "http://datasets.polixir.site/v2/Walker2d-v3/Walker2d-v3-medium-100-train-noise.npz", "md5": "6a9f76691df4a4f386c9cb5da564063b"},
"Walker2d-v3-medium-1000-train-noise.npz": {"url": "http://datasets.polixir.site/v2/Walker2d-v3/Walker2d-v3-medium-1000-train-noise.npz", "md5": "a93baf6c53b578cd8c2286ccc1f837d9"},
"Walker2d-v3-medium-10000-train-noise.npz": {"url": "http://datasets.polixir.site/v2/Walker2d-v3/Walker2d-v3-medium-10000-train-noise.npz", "md5": "2a0efcc3b3e4f4afabdf2f3ec4869980"},
"Walker2d-v3.yaml": {"url": "http://datasets.polixir.site/v2/Walker2d-v3/Walker2d-v3.yaml", "md5": "3a8a7a19a96db3d94a56644f8dfbe6f8"},
"Walker2d-v3_reward.py": {"url": "http://datasets.polixir.site/v2/Walker2d-v3/Walker2d-v3_reward.py", "md5": "66a63b9a685374464219ba0d92e2e4aa"},
"citylearn-expert-10-val-noise.npz": {"url": "http://datasets.polixir.site/v2/citylearn/citylearn-expert-10-val-noise.npz", "md5": "76c4a9babad4695011fa5b2f1013b178"},
"citylearn-expert-100-val-noise.npz": {"url": "http://datasets.polixir.site/v2/citylearn/citylearn-expert-100-val-noise.npz", "md5": "3ba1b46f058aae6fb497f7a1bcdfc440"},
"citylearn-expert-1000-val-noise.npz": {"url": "http://datasets.polixir.site/v2/citylearn/citylearn-expert-1000-val-noise.npz", "md5": "e84cab7d8a97cd63a224313096ff2303"},
"citylearn-expert-100-train-noise.npz": {"url": "http://datasets.polixir.site/v2/citylearn/citylearn-expert-100-train-noise.npz", "md5": "b2e0099b41885ef2f59c33a0c9934554"},
"citylearn-expert-1000-train-noise.npz": {"url": "http://datasets.polixir.site/v2/citylearn/citylearn-expert-1000-train-noise.npz", "md5": "696fd67b1203c9e07c108286d29ee21b"},
"citylearn-expert-10000-train-noise.npz": {"url": "http://datasets.polixir.site/v2/citylearn/citylearn-expert-10000-train-noise.npz", "md5": "d9448197c5d05986a9195a73c4a22232"},
"citylearn-high-10-val-noise.npz": {"url": "http://datasets.polixir.site/v2/citylearn/citylearn-high-10-val-noise.npz", "md5": "4ed4dbb1021695758388fd8861b3c5d9"},
"citylearn-high-100-val-noise.npz": {"url": "http://datasets.polixir.site/v2/citylearn/citylearn-high-100-val-noise.npz", "md5": "979bcc8eebb17ebbf0c75e524065cc3a"},
"citylearn-high-1000-val-noise.npz": {"url": "http://datasets.polixir.site/v2/citylearn/citylearn-high-1000-val-noise.npz", "md5": "36d61c8b8bc43f3e8ca92f1fe6f468cd"},
"citylearn-high-100-train-noise.npz": {"url": "http://datasets.polixir.site/v2/citylearn/citylearn-high-100-train-noise.npz", "md5": "2cdf2dedf870784cc99c63fa1fc75630"},
"citylearn-high-1000-train-noise.npz": {"url": "http://datasets.polixir.site/v2/citylearn/citylearn-high-1000-train-noise.npz", "md5": "ff1b6a521a937b56331b253d60963100"},
"citylearn-high-10000-train-noise.npz": {"url": "http://datasets.polixir.site/v2/citylearn/citylearn-high-10000-train-noise.npz", "md5": "9c87cd5d18c411c33f9b4f4813a97367"},
"citylearn-low-10-val-noise.npz": {"url": "http://datasets.polixir.site/v2/citylearn/citylearn-low-10-val-noise.npz", "md5": "d52ad7284835dfa3fd467616b35c23c1"},
"citylearn-low-100-val-noise.npz": {"url": "http://datasets.polixir.site/v2/citylearn/citylearn-low-100-val-noise.npz", "md5": "e15e34479c8eaa2fb9f14fafe592e0bb"},
"citylearn-low-1000-val-noise.npz": {"url": "http://datasets.polixir.site/v2/citylearn/citylearn-low-1000-val-noise.npz", "md5": "c47352dac7f5243c1b2cbdaabc69f3b9"},
"citylearn-low-100-train-noise.npz": {"url": "http://datasets.polixir.site/v2/citylearn/citylearn-low-100-train-noise.npz", "md5": "ff01f936cc4e5ccc8b8b28a5840c47d3"},
"citylearn-low-1000-train-noise.npz": {"url": "http://datasets.polixir.site/v2/citylearn/citylearn-low-1000-train-noise.npz", "md5": "5e33f0edce169850583512170285bcd7"},
"citylearn-low-10000-train-noise.npz": {"url": "http://datasets.polixir.site/v2/citylearn/citylearn-low-10000-train-noise.npz", "md5": "772e32bc96603d68d9f66bd65249528f"},
"citylearn-medium-10-val-noise.npz": {"url": "http://datasets.polixir.site/v2/citylearn/citylearn-medium-10-val-noise.npz", "md5": "346c9cbc8205102ec6d226680c6636b2"},
"citylearn-medium-100-val-noise.npz": {"url": "http://datasets.polixir.site/v2/citylearn/citylearn-medium-100-val-noise.npz", "md5": "46c821efc87f6a1d273862d626f9bcbc"},
"citylearn-medium-1000-val-noise.npz": {"url": "http://datasets.polixir.site/v2/citylearn/citylearn-medium-1000-val-noise.npz", "md5": "7555ec64d1f7f2587f7bc81de1114a46"},
"citylearn-medium-100-train-noise.npz": {"url": "http://datasets.polixir.site/v2/citylearn/citylearn-medium-100-train-noise.npz", "md5": "8663bf343e1ee569f62950705a09bc4d"},
"citylearn-medium-1000-train-noise.npz": {"url": "http://datasets.polixir.site/v2/citylearn/citylearn-medium-1000-train-noise.npz", "md5": "8d25c9cd6b55376ad189d119cb86380b"},
"citylearn-medium-10000-train-noise.npz": {"url": "http://datasets.polixir.site/v2/citylearn/citylearn-medium-10000-train-noise.npz", "md5": "d31644dbbe891023fc54fc048c2fac5a"},
"citylearn.yaml": {"url": "http://datasets.polixir.site/v2/citylearn/citylearn.yaml", "md5": "da63379d53ad05221f089afc5b44f673"},
"citylearn_reward.py": {"url": "http://datasets.polixir.site/v2/citylearn/citylearn_reward.py", "md5": "d147bd875461e83132352291afcb4431"},
"finance-expert-10-val-noise.npz": {"url": "http://datasets.polixir.site/v2/finance/finance-expert-10-val-noise.npz", "md5": "2f84951ba2082f64c76c18e409f3d798"},
"finance-expert-100-val-noise.npz": {"url": "http://datasets.polixir.site/v2/finance/finance-expert-100-val-noise.npz", "md5": "4a25d252852997d19b632ff02c87cba2"},
"finance-expert-100-train-noise.npz": {"url": "http://datasets.polixir.site/v2/finance/finance-expert-100-train-noise.npz", "md5": "e063a523452967696f8f38e77bbe8a88"},
"finance-expert-1000-train-noise.npz": {"url": "http://datasets.polixir.site/v2/finance/finance-expert-1000-train-noise.npz", "md5": "47406f520713f34479ef39674a1fbb5f"},
"finance-high-10-val-noise.npz": {"url": "http://datasets.polixir.site/v2/finance/finance-high-10-val-noise.npz", "md5": "75cef37a50575fb695d9d0ff67261d0d"},
"finance-high-100-val-noise.npz": {"url": "http://datasets.polixir.site/v2/finance/finance-high-100-val-noise.npz", "md5": "c311c59401d2bd6b86292f196ae78070"},
"finance-high-100-train-noise.npz": {"url": "http://datasets.polixir.site/v2/finance/finance-high-100-train-noise.npz", "md5": "74455f5a13db80b2999bbf352613bcec"},
"finance-high-1000-train-noise.npz": {"url": "http://datasets.polixir.site/v2/finance/finance-high-1000-train-noise.npz", "md5": "b4445c67bf842c6b614f384aaec5d581"},
"finance-low-10-val-noise.npz": {"url": "http://datasets.polixir.site/v2/finance/finance-low-10-val-noise.npz", "md5": "4833b468490219f4715b71d369446e4b"},
"finance-low-100-val-noise.npz": {"url": "http://datasets.polixir.site/v2/finance/finance-low-100-val-noise.npz", "md5": "65bd1cf7b60a832488f84f0c57c40acd"},
"finance-low-100-train-noise.npz": {"url": "http://datasets.polixir.site/v2/finance/finance-low-100-train-noise.npz", "md5": "01539166db945a5007fbc791c1175fe4"},
"finance-low-1000-train-noise.npz": {"url": "http://datasets.polixir.site/v2/finance/finance-low-1000-train-noise.npz", "md5": "48afa7cdb9dc8593906cfcc8390d6b67"},
"finance-medium-10-val-noise.npz": {"url": "http://datasets.polixir.site/v2/finance/finance-medium-10-val-noise.npz", "md5": "98f199b8725692dd06b030b89fbe7bd0"},
"finance-medium-100-val-noise.npz": {"url": "http://datasets.polixir.site/v2/finance/finance-medium-100-val-noise.npz", "md5": "57e1e80de5b115873decf443269e3e40"},
"finance-medium-100-train-noise.npz": {"url": "http://datasets.polixir.site/v2/finance/finance-medium-100-train-noise.npz", "md5": "75a902e0bdc06996a679554b90a8df8a"},
"finance-medium-1000-train-noise.npz": {"url": "http://datasets.polixir.site/v2/finance/finance-medium-1000-train-noise.npz", "md5": "90f80385b21763a6fe3e26f168926c49"},
"finance.yaml": {"url": "http://datasets.polixir.site/v2/finance/finance.yaml", "md5": "a7ea340f912a2ead1b7b48c1d8f9d140"},
"finance_reward.py": {"url": "http://datasets.polixir.site/v2/finance/finance_reward.py", "md5": "9414a483d6bbf9236e9ff281db285f3e"},
"ib-high-10-val.npz": {"url": "http://datasets.polixir.site/v2/ib/ib-high-10-val.npz", "md5": "dbd918c5c68ad16d51c279516467de8f"},
"ib-high-100-val.npz": {"url": "http://datasets.polixir.site/v2/ib/ib-high-100-val.npz", "md5": "40ec5ab4def7879ddca1eb79bbf55f00"},
"ib-high-1000-val.npz": {"url": "http://datasets.polixir.site/v2/ib/ib-high-1000-val.npz", "md5": "7cd0589aadc7202c9e1d38171b6ddd8d"},
"ib-high-100-train.npz": {"url": "http://datasets.polixir.site/v2/ib/ib-high-100-train.npz", "md5": "1e7750ef0e8bac223ed56f47f8429d87"},
"ib-high-1000-train.npz": {"url": "http://datasets.polixir.site/v2/ib/ib-high-1000-train.npz", "md5": "bb1e5e75e15c805cda8c3270c10b9294"},
"ib-high-10000-train.npz": {"url": "http://datasets.polixir.site/v2/ib/ib-high-10000-train.npz", "md5": "38f34227286d47e8a4aa723b82d583c2"},
"ib-low-10-val.npz": {"url": "http://datasets.polixir.site/v2/ib/ib-low-10-val.npz", "md5": "22920bc28f9c81c60321ec3f2fc7b71a"},
"ib-low-100-val.npz": {"url": "http://datasets.polixir.site/v2/ib/ib-low-100-val.npz", "md5": "37b4f544ed0deb76043d67ed644ad2e7"},
"ib-low-1000-val.npz": {"url": "http://datasets.polixir.site/v2/ib/ib-low-1000-val.npz", "md5": "621805e19e95467ef4bf82b946ab35b5"},
"ib-low-100-train.npz": {"url": "http://datasets.polixir.site/v2/ib/ib-low-100-train.npz", "md5": "d0b4fa78b27dbf86411b3b41de20d71a"},
"ib-low-1000-train.npz": {"url": "http://datasets.polixir.site/v2/ib/ib-low-1000-train.npz", "md5": "b1f1bc865c7910ff87e9277d1b23e745"},
"ib-low-10000-train.npz": {"url": "http://datasets.polixir.site/v2/ib/ib-low-10000-train.npz", "md5": "bd5a44b98ea574517cff50fa38bcae7a"},
"ib-medium-10-val.npz": {"url": "http://datasets.polixir.site/v2/ib/ib-medium-10-val.npz", "md5": "eb146a26eef350ad116901bc36a1d833"},
"ib-medium-100-val.npz": {"url": "http://datasets.polixir.site/v2/ib/ib-medium-100-val.npz", "md5": "27839d327ac8a933a0d0a91ed567935b"},
"ib-medium-1000-val.npz": {"url": "http://datasets.polixir.site/v2/ib/ib-medium-1000-val.npz", "md5": "deed6007666f35af468cc22d9026bb1d"},
"ib-medium-100-train.npz": {"url": "http://datasets.polixir.site/v2/ib/ib-medium-100-train.npz", "md5": "afd8973406bc7401ce77d9b0b4695947"},
"ib-medium-1000-train.npz": {"url": "http://datasets.polixir.site/v2/ib/ib-medium-1000-train.npz", "md5": "8aeededf5c3baf3eed30f6ca8080fe02"},
"ib-medium-10000-train.npz": {"url": "http://datasets.polixir.site/v2/ib/ib-medium-10000-train.npz", "md5": "7f601ab6adae6790635801812aa66b7f"},
"ib.yaml": {"url": "http://datasets.polixir.site/v2/ib/ib.yaml", "md5": "1ddfbf446e77fbe51ab2704c1632f32a"},
"ib_reward.py": {"url": "http://datasets.polixir.site/v2/ib/ib_reward.py", "md5": "5d1a7476b6aa5c1084a5194e58425d5b"}
}

+ 2
- 2
neorl/md5_util.py View File

@ -4,9 +4,9 @@ from neorl.utils import get_json, LOCAL_JSON_FILE_PATH, DATA_PATH, download_data
def print_all_kv_with_md5(json_file):
items = json_file.items()
for key, value in items:
filename = download_dataset_from_url(value, name=key, to_path=DATA_PATH)
filename = download_dataset_from_url(value, name=key, to_path=DATA_PATH, verbose=0)
__md5 = get_file_md5(filename)
__new_kv = '"' + key + '": {"url": "' + value + '", "md5": "' + __md5 + '"}, '
__new_kv = '"' + key + '": {"url": "' + value["url"] + '", "md5": "' + __md5 + '"}, '
print(__new_kv)


+ 43
- 0
neorl/neorl_envs/Hopper-v3/Hopper-v3_done.py View File

@ -0,0 +1,43 @@
import numpy as np
def get_done(data):
obs = data["obs"]
action = data["action"]
obs_next = data["next_obs"]
singel_done = False
if len(obs.shape) == 1:
singel_done = True
obs = obs.reshape(1, -1)
if len(action.shape) == 1:
action = action.reshape(1, -1)
if len(obs_next.shape) == 1:
obs_next = obs_next.reshape(1, -1)
if isinstance(obs, np.ndarray):
array_type = np
else:
import torch
array_type = torch
z = obs_next[:, 1:2]
angle = obs_next[:, 2:3]
state = obs_next[:, 3:]
min_state, max_state = (-100.0, 100.0)
min_z, max_z = (0.7, float('inf'))
min_angle, max_angle = (-0.2, 0.2)
healthy_state = array_type.all(array_type.logical_and(min_state < state, state < max_state), axis=-1, keepdim=True)
healthy_z = array_type.logical_and(min_z < z, z < max_z)
healthy_angle = array_type.logical_and(min_angle < angle, angle < max_angle)
is_healthy = array_type.logical_and(array_type.logical_and(healthy_state, healthy_z), healthy_angle)
done = array_type.logical_not(is_healthy)
if singel_done:
done = done
else:
done = done.reshape(-1, 1)
return done

+ 43
- 0
neorl/neorl_envs/Walker2d-v3/Walker2d-v3_done.py View File

@ -0,0 +1,43 @@
import numpy as np
def get_done(data):
obs = data["obs"]
action = data["action"]
obs_next = data["next_obs"]
singel_done = False
if len(obs.shape) == 1:
singel_done = True
obs = obs.reshape(1, -1)
if len(action.shape) == 1:
action = action.reshape(1, -1)
if len(obs_next.shape) == 1:
obs_next = obs_next.reshape(1, -1)
if isinstance(obs, np.ndarray):
array_type = np
else:
import torch
array_type = torch
min_z, max_z = (0.8, 2.0)
min_angle, max_angle = (-1.0, 1.0)
min_state, max_state = (-100.0, 100.0)
z = obs_next[:, 1:2]
angle = obs_next[:, 2:3]
state = obs_next[:, 3:]
healthy_state = array_type.all(array_type.logical_and(min_state < state, state < max_state), axis=-1, keepdim=True)
healthy_z = array_type.logical_and(min_z < z, z < max_z)
healthy_angle = array_type.logical_and(min_angle < angle, angle < max_angle)
is_healthy = array_type.logical_and(array_type.logical_and(healthy_state, healthy_z), healthy_angle)
done = array_type.logical_not(is_healthy)
if singel_done:
done = done
else:
done = done.reshape(-1, 1)
return done

+ 2
- 2
neorl/neorl_envs/ib/ib_reward.py View File

@ -15,8 +15,8 @@ def get_reward(data):
CRF = 3.0
CRC = 1.0
fatigue = obs_next[:, -2]
consumption = obs_next[:, -1]
fatigue = obs_next[:, 4]
consumption = obs_next[:, 5]
cost = CRF * fatigue + CRC * consumption


+ 5
- 47
neorl/neorl_envs/ib/industrial_benchmark_python/IBGym.py View File

@ -37,7 +37,7 @@ class IBGym(core.EnvData):
OpenAI Gym Wrapper for the industrial benchmark
"""
def __init__(self, setpoint, reward_type, action_type, observation_type="classic", reset_after_timesteps=1000,
obs_reward=True, init_seed=None, n_past_timesteps=30):
init_seed=None, n_past_timesteps=30):
"""
Initializes the underlying environment, seeds numpy and initializes action / observation spaces
as well as other necessary variables
@ -46,7 +46,6 @@ class IBGym(core.EnvData):
:param action_type: discrete / continuous
:param observation_type: classic / include_past - determines wether single or N state frames used as observation
:param reset_after_timesteps: how many timesteps can the environment run without resetting
:param obs_reward: add fatigue and consumption to observation for reward caculation
:param init_seed: seed for numpy to make environment behavior reproducible
:param n_past_timesteps: if observation type is include_past, this determines how many state frames are used
"""
@ -60,7 +59,6 @@ class IBGym(core.EnvData):
# Used to determine whether to return the absolute value or the relative change in the cost function
self.reward_function = reward_type
self.obs_reward = obs_reward
# Used to set an arbitrary limit of how many time steps the environment can take before resetting
self.reset_after_timesteps = reset_after_timesteps
@ -105,14 +103,9 @@ class IBGym(core.EnvData):
if self.observation_type == "classic": # classic only has the current state frame
self.observation_space = gym.spaces.Box(low=single_low, high=single_high)
elif self.observation_type == "include_past": # time embedding: state contains also past N state frames
if self.obs_reward:
low = np.hstack([single_low] * self.n_past_timesteps + [-100, -100])
high = np.hstack([single_high] * self.n_past_timesteps + [100, 100])
else:
low = np.hstack([single_low] * self.n_past_timesteps)
high = np.hstack([single_high] * self.n_past_timesteps)
elif self.observation_type == "include_past": # time embedding: state contains also past N state frames
low = np.hstack([single_low] * self.n_past_timesteps)
high = np.hstack([single_high] * self.n_past_timesteps)
self.observation_space = gym.spaces.Box(low=low, high=high)
else:
@ -168,10 +161,6 @@ class IBGym(core.EnvData):
' or "delta" for the change in the cost fucntion between steps.')
self.info = self._markovian_state() # entire markov state - not all info is visible in observations
if self.obs_reward:
return_reward = self.get_reward_from_data(return_observation,action,return_observation)
return return_observation, return_reward, self.done, self.info
def reset(self):
@ -253,12 +242,6 @@ class IBGym(core.EnvData):
else:
raise ValueError('Invalid observation_type. observation_type can either be "classic" or "include_past"')
if self.obs_reward:
fatigue = np.array([self.IB.state['f']])
consumption = np.array([self.IB.state['c']])
return_observation = np.concatenate([return_observation,fatigue,consumption], axis=-1)
return return_observation
def _markovian_state(self):
@ -294,29 +277,4 @@ class IBGym(core.EnvData):
self.IB.state['hg']]
info = OrderedDict(zip(markovian_states_variables, markovian_states_values))
return info
def get_reward_from_data(self, obs, action, obs_next):
singel_reward = False
if len(obs.shape) == 1:
singel_reward = True
obs = obs.reshape(1,-1)
if len(action.shape) == 1:
action = action.reshape(1,-1)
if len(obs_next.shape) == 1:
obs_next = obs_next.reshape(1,-1)
CRF = self.IB.CRF
CRC = self.IB.CRC
fatigue = obs_next[:,-2]
consumption = obs_next[:,-1]
cost = CRF * fatigue + CRC * consumption
reward = -cost
if singel_reward:
reward = reward[0].item()
return reward
return info

+ 4
- 0
neorl/neorl_envs/mujoco.py View File

@ -9,4 +9,8 @@ def make_env(task):
env.get_dataset = env_data.get_dataset
env.set_reward_func = env_data.set_reward_func
env.get_reward_func = env_data.get_reward_func
env.set_done_func = env_data.set_done_func
env.get_done_func = env_data.get_done_func
return env

+ 4
- 4
neorl/tests/test_envdata.py View File

@ -80,13 +80,13 @@ def test_citylearn():
def test_finance():
env = neorl.make("finance")
train_data, val_data = env.get_dataset(train_num=99, data_type="H", path=TEST_DATA_PATH)
train_data, val_data = env.get_dataset(train_num=100, data_type="H", path=TEST_DATA_PATH)
assert int(len(train_data["index"]) * 0.1) == len(val_data["index"])
def test_ib():
env = neorl.make("ib")
train_data, val_data = env.get_dataset(train_num=99, data_type="M", path=TEST_DATA_PATH)
train_data, val_data = env.get_dataset(train_num=100, data_type="M", path=TEST_DATA_PATH)
assert int(len(train_data["index"]) * 0.1) == len(val_data["index"])
@ -98,10 +98,10 @@ def test_ib():
def test_mujoco():
env = neorl.make("HalfCheetah-v3")
train_data, val_data = env.get_dataset(train_num=99, data_type="L", path=TEST_DATA_PATH)
train_data, val_data = env.get_dataset(train_num=100, data_type="L", path=TEST_DATA_PATH)
assert int(len(train_data["index"]) * 0.1) == len(val_data["index"])
env = neorl.make("Walker2d-v3")
train_data, val_data = env.get_dataset(train_num=9, data_type="m", path=TEST_DATA_PATH)
train_data, val_data = env.get_dataset(train_num=10, data_type="m", path=TEST_DATA_PATH)
assert int(len(train_data["index"]) * 0.1) == len(val_data["index"])
env = neorl.make("Hopper-v3")
train_data, val_data = env.get_dataset(train_num=0, data_type="e", path=TEST_DATA_PATH)


+ 5
- 4
neorl/utils.py View File

@ -35,7 +35,7 @@ def get_file_md5(filename):
return my_hash.hexdigest()
def download_dataset_from_url(dataset_url_md5, name, to_path):
def download_dataset_from_url(dataset_url_md5, name, to_path, verbose=1):
"""
Download dataset from url to `to_path + name`.
"""
@ -56,9 +56,10 @@ def download_dataset_from_url(dataset_url_md5, name, to_path):
if local_file_md5 == dataset_md5:
return dataset_filepath
else:
print(f"Local dataset {name} is broken, ready to re-download.")
print(f'Downloading dataset: {dataset_url} to {dataset_filepath}')
if verbose != 0:
print(f"Local dataset {name} is broken, ready to re-download.")
if verbose != 0:
print(f'Downloading dataset: {dataset_url} to {dataset_filepath}')
urllib.request.urlretrieve(dataset_url, dataset_filepath)
if not os.path.exists(dataset_filepath):


Loading…
Cancel
Save