fix: remove 'simzoo' submodule (#246)

This commit removes the `simzoo` submodule since it is now available as a stand-alone package called
[stable-gym](https://github.com/rickstaa/stable-gym). This was done since having the package in this
package required a lot of complicated code and made the package harder to debug.

BREAKING CHANGE: The `simzoo` package is no longer included and has to be installed
separately through the [stable-gym](https://github.com/rickstaa/stable-gym) repository.

README.md

@@ -15,7 +15,6 @@ Imitation learning (IL) control algorithms directly from real-world data. This f
 is made up of four main modules:
 
 *   [Modeling](./bayesian_learning_control/modeling): Module that uses state of the art System Identification and State Estimation techniques to create an Openai gym environment out of real data.
-*   [Simzoo](https://github.com/rickstaa/simzoo): Module that contains several already created [Bayesian Learning Control](https://rickstaa.github.io/bayesian-learning-control/simzoo/simzoo.html) Openai gym environments.
 *   [Control](./bayesian_learning_control/control): Module used to train several [Bayesian Learning Control](https://rickstaa.github.io/bayesian-learning-control/control/control.html) RL/IL agents on the built gym environments.
 *   [Hardware](./bayesian_learning_control/hardware): Module that can be used to deploy the trained RL/IL agents onto the hardware of your choice.
 

bayesian_learning_control/control/algos/pytorch/lac/lac.py

@@ -32,13 +32,14 @@
 import numpy as np
 import torch
 import torch.nn as nn
+from torch.optim import Adam
+
 from bayesian_learning_control.control.algos.pytorch.common import get_lr_scheduler
 from bayesian_learning_control.control.algos.pytorch.common.buffers import ReplayBuffer
 from bayesian_learning_control.control.algos.pytorch.common.helpers import (
     count_vars,
     retrieve_device,
 )
-
 # fmt: off
 from bayesian_learning_control.control.algos.pytorch.policies.lyapunov_actor_critic import \
     LyapunovActorCritic  # noqa: E501
@@ -58,7 +59,6 @@
     setup_logger_kwargs,
 )
 from bayesian_learning_control.utils.serialization_utils import save_to_json
-from torch.optim import Adam
 
 # Import ray tuner if installed
 tune = lazy_importer(module_name="ray.tune")
@@ -635,8 +635,8 @@ def load_state_dict(self, state_dict, restore_lagrance_multipliers=True):
             ) from e
 
     def state_dict(self):
-        """Simple wrapper around the :meth:`torch.nn.Module.state_dict` method that saves
-        the current class name. This is used to enable easy loading of the model.
+        """Simple wrapper around the :meth:`torch.nn.Module.state_dict` method that
+        saves the current class name. This is used to enable easy loading of the model.
         """
         state_dict = super().state_dict()
         state_dict[
@@ -1316,9 +1316,8 @@ def lac(  # noqa: C901
 
 
 if __name__ == "__main__":
-
-    # Import gym environments
-    import bayesian_learning_control.simzoo.simzoo.envs  # noqa: F401
+    # NOTE: You can import your custom gym environment here.
+    # import stable_gym  # noqa: F401
 
     parser = argparse.ArgumentParser(
         description="Trains a LAC agent in a given environment."

bayesian_learning_control/control/algos/pytorch/sac/sac.py

@@ -34,6 +34,8 @@
 import numpy as np
 import torch
 import torch.nn as nn
+from torch.optim import Adam
+
 from bayesian_learning_control.control.algos.pytorch.common import get_lr_scheduler
 from bayesian_learning_control.control.algos.pytorch.common.buffers import ReplayBuffer
 from bayesian_learning_control.control.algos.pytorch.common.helpers import (
@@ -58,7 +60,6 @@
     setup_logger_kwargs,
 )
 from bayesian_learning_control.utils.serialization_utils import save_to_json
-from torch.optim import Adam
 
 # Import ray tuner if installed
 tune = lazy_importer(module_name="ray.tune")
@@ -605,8 +606,8 @@ def load_state_dict(self, state_dict, restore_lagrance_multipliers=True):
             ) from e
 
     def state_dict(self):
-        """Simple wrapper around the :meth:`torch.nn.Module.state_dict` method that saves
-        the current class name. This is used to enable easy loading of the model.
+        """Simple wrapper around the :meth:`torch.nn.Module.state_dict` method that
+        saves the current class name. This is used to enable easy loading of the model.
         """
         state_dict = super().state_dict()
         state_dict[
@@ -1083,7 +1084,6 @@ def sac(  # noqa: C901
 
         # Update handling
         if (t + 1) >= update_after and ((t + 1) - update_after) % update_every == 0:
-
             # Step based learning rate decay
             if lr_decay_ref.lower() == "step":
                 for scheduler in opt_schedulers:
@@ -1219,9 +1219,8 @@ def sac(  # noqa: C901
 
 
 if __name__ == "__main__":
-
-    # Import gym environments
-    import bayesian_learning_control.simzoo.simzoo.envs  # noqa: F401
+    # NOTE: You can import your custom gym environment here.
+    # import stable_gym  # noqa: F401
 
     parser = argparse.ArgumentParser(
         description="Trains a SAC agent in a given environment."

bayesian_learning_control/control/algos/tf2/lac/lac.py

@@ -28,6 +28,7 @@
 
 import gym
 import numpy as np
+
 from bayesian_learning_control.common.helpers import combine_shapes
 from bayesian_learning_control.control.algos.tf2.common import get_lr_scheduler
 from bayesian_learning_control.control.algos.tf2.common.helpers import (
@@ -1259,8 +1260,8 @@ def lac(  # noqa: C901
 
 
 if __name__ == "__main__":
-    # Import gym environments
-    import bayesian_learning_control.simzoo.simzoo.envs  # noqa: F401
+    # NOTE: You can import your custom gym environment here.
+    # import stable_gym  # noqa: F401
 
     parser = argparse.ArgumentParser(
         description="Trains a LAC agent in a given environment."

bayesian_learning_control/control/algos/tf2/policies/actors/squashed_gaussian_actor.py

@@ -139,7 +139,6 @@ def call(self, obs, deterministic=False, with_logprob=True):
 
         # Compute logprob from Gaussian, and then apply correction for Tanh squashing.
         if with_logprob:
-
             # Transform base_distribution to the policy distribution
             reparm_trick_bijector = tfp.bijectors.Chain(
                 (self._squash_bijector, affine_bijector)

bayesian_learning_control/control/algos/tf2/sac/sac.py

@@ -28,6 +28,7 @@
 
 import gym
 import numpy as np
+
 from bayesian_learning_control.common.helpers import combine_shapes
 from bayesian_learning_control.control.algos.tf2.common import get_lr_scheduler
 from bayesian_learning_control.control.algos.tf2.common.helpers import (
@@ -1180,17 +1181,17 @@ def sac(  # noqa: C901
 
 
 if __name__ == "__main__":
-    # Import gym environments
-    import bayesian_learning_control.simzoo.simzoo.envs  # noqa: F401
+    # NOTE: You can import your custom gym environment here.
+    # import stable_gym  # noqa: F401
 
     parser = argparse.ArgumentParser(
         description="Trains a SAC agent in a given environment."
     )
     parser.add_argument(
         "--env",
         type=str,
-        default="Oscillator-v1",
-        help="the gym env (default: Oscillator-v1)",
+        default="CartPole-v1",
+        help="the gym env (default: CartPole-v1)",
     )
     parser.add_argument(
         "--hid_a",

bayesian_learning_control/control/common/buffers.py

@@ -302,7 +302,6 @@ def finish_path(self, last_val=0):
 
         # Calculate the advantage and rewards-to-go if buffer contains vals
         if self._contains_vals:
-
             # Get the current trajectory
             path_slice = slice(self.traj_ptr, self.ptr)
             rews = np.append(self.rew_buf[path_slice], last_val)

bayesian_learning_control/control/utils/eval_robustness.py

@@ -1,15 +1,14 @@
-"""A set of functions that can be used to evaluate the stability and robustness of an
-algorithm. This is done by evaluating an algorithm's performance under two types of
-disturbances: A disturbance that is applied during the environment step and a
-perturbation added to the environmental parameters. For the functions in this
-module to work work, these disturbances should be implemented as methods on the
-environment. The Simzoo package contains a
-:class:`~bayesian_learning_control.simzoo.simzoo.common.disturber.Disturber` class from which a Gym environment
-can inherit to add these methods. See the
-`Robustness Evaluation Documentation <https://rickstaa.github.io/bayesian-learning-control/control/robustness_eval.html>`_
+"""Contains a set of functions that can be used to evaluate the stability and robustness
+of an algorithm. This is done by evaluating an algorithm's performance under two
+disturbances: A disturbance applied during the environment step and a perturbation
+added to the environmental parameters. For the functions in this module to work work,
+these disturbances should be implemented as methods on the environment. The
+:stable-gym:`Stable Gym <>` package contains a 	:class:`~stable_gym.common.disturber.Disturber`
+class from which a Gym environment can inherit to add these methods. See the `Robustness
+Evaluation Documentation
+<https://rickstaa.github.io/bayesian-learning control/control/robustness_eval.html>`_
 for more information.
 """  # noqa: E501
-
 import math
 import os
 import sys
@@ -20,6 +19,7 @@
 import numpy as np
 import pandas as pd
 import seaborn as sns
+
 from bayesian_learning_control.control.common.helpers import validate_observations
 from bayesian_learning_control.control.utils.test_policy import load_policy_and_env
 from bayesian_learning_control.utils.log_utils import (
@@ -36,7 +36,7 @@
 
 def _disturber_implemented(env):
     """Checks if the environment inherits from the
-    :class:`~bayesian_learning_control.simzoo.simzoo.common.disturber.Disturber`
+    :class:`~stable_gym.common.disturber.Disturber`
     class or that the methods and attributes that are required for the robustness
     evaluation are present.
 

bayesian_learning_control/control/utils/gym_utils.py

@@ -8,9 +8,10 @@
 from textwrap import dedent
 
 import gym
-from bayesian_learning_control.utils.log_utils import friendly_err
 from gym import spaces
 
+from bayesian_learning_control.utils.log_utils import friendly_err
+
 DISCRETE_SPACES = (
     spaces.Discrete,
     spaces.MultiBinary,
@@ -68,13 +69,14 @@ def validate_gym_env(arg_dict):
     # Import gym environments
     import gym
 
-    # import bayesian_learning_control.simzoo.simzoo  # noqa: F401
+    # Import environment configuration file. This file can be used to inject
+    # custom gym environments into the blc package.
     try:
         import bayesian_learning_control.env_config  # noqa: F401
     except Exception as e:
         raise Exception(
             friendly_err(
-                "Something went wrong when trying to import the 'env_config' file."
+                "Something went wrong when trying to import the 'env_config' " " file."
             )
         ) from e
 

bayesian_learning_control/control/utils/run_entrypoint.py

@@ -13,10 +13,9 @@
 """  # noqa
 
 import base64
+import pickle
 import zlib
 
-import pickle5 as pickle
-
 if __name__ == "__main__":
     import argparse
 

bayesian_learning_control/control/utils/run_utils.py

@@ -21,6 +21,8 @@
 import cloudpickle
 import numpy as np
 import psutil
+from tqdm import trange
+
 from bayesian_learning_control.common.helpers import all_bools, valid_str
 from bayesian_learning_control.user_config import DEFAULT_SHORTHAND, WAIT_BEFORE_LAUNCH
 from bayesian_learning_control.utils.log_utils import (
@@ -30,7 +32,6 @@
 )
 from bayesian_learning_control.utils.mpi_utils.mpi_tools import mpi_fork
 from bayesian_learning_control.utils.serialization_utils import convert_json
-from tqdm import trange
 
 DIV_LINE_WIDTH = 80
 
@@ -112,7 +113,8 @@ def thunk_plus():
             # Import gym environments
             import gym
 
-            # import bayesian_learning_control.simzoo.simzoo  # noqa: F401
+            # Import environment configuration file. This file can be used to inject
+            # custom gym environments into the blc package.
             try:
                 import bayesian_learning_control.env_config  # noqa: F401
             except Exception as e:
@@ -362,14 +364,12 @@ def get_val(v, k):
         # Build the rest of the name by looping through all parameters,
         # and deciding which ones need to go in there.
         for k, v, sh, inn in zip(self.keys, self.vals, self.shs, self.in_names):
-
             # Include a parameter in a name if either 1) it can take multiple
             # values, or 2) the user specified that it must appear in the name.
             # Except, however, when the parameter is 'seed'. Seed is handled
             # differently so that runs of the same experiment, with different
             # seeds, will be grouped by experiment name.
             if (len(v) > 1 or inn) and not (k == "seed"):
-
                 # Use the shorthand if available, otherwise the full name.
                 param_name = sh if sh is not None else k
                 param_name = valid_str(param_name)

bayesian_learning_control/control/utils/safer_eval.py

@@ -13,7 +13,6 @@
 
 import bayesian_learning_control as bayesian_learning_control  # noqa: F401
 import bayesian_learning_control as blc  # noqa: F401
-import bayesian_learning_control.simzoo.simzoo  # noqa: F401
 
 # Import modules to which you want users to have access
 import torch  # noqa: F401

bayesian_learning_control/env_config.py

@@ -7,8 +7,7 @@
     `this issue on the Openai github <https://github.com/openai/gym/blob/master/docs/creating-environments.md>`_
     for more information on how to create custom environments. Environments that are
     added to this file can be called directly without using the module prefix (eg.
-    ``Oscillator-v1`` instead of
-    ``bayesian_learning_control.simzoo.simzoo:Oscillator-v1``).
+    ``Oscillator-v1`` instead of ``stable-gym:Oscillator-v1``).
 
     .. code-block:: python
 

bayesian_learning_control/utils/log_utils/logx.py

@@ -12,12 +12,14 @@
 import json
 import os
 import os.path as osp
+import pickle
 import time
 
 import joblib
 import numpy as np
-import pickle5 as pickle
 import torch
+from torch.utils.tensorboard import SummaryWriter
+
 from bayesian_learning_control.common.helpers import is_scalar
 from bayesian_learning_control.user_config import DEFAULT_STD_OUT_TYPE
 from bayesian_learning_control.utils.import_utils import import_tf
@@ -31,7 +33,6 @@
     load_from_json,
     save_to_json,
 )
-from torch.utils.tensorboard import SummaryWriter
 
 
 class Logger:
@@ -88,7 +89,6 @@ def __init__(
             exp_name (str): Experiment name.
         """
         if proc_id() == 0:
-
             # Parse output_fname to see if csv was requested
             extension = osp.splitext(output_fname)[1]
             self._output_csv = True if extension.lower() == ".csv" else False
@@ -524,7 +524,6 @@ def save_state(self, state_dict, itr=None):
                 to None
         """
         if proc_id() == 0:
-
             # Save training state (environment, ...)
             try:
                 joblib.dump(state_dict, osp.join(self.output_dir, "vars.pkl"))
@@ -586,7 +585,6 @@ def _tf_save(self, itr=None):
                 to None
         """
         if proc_id() == 0:
-
             save_fail_warning = (
                 "The object you tried to save doesn't have a 'save_weights' we "
                 "can use to retrieve the model weights. Please make sure you supplied "
@@ -655,7 +653,6 @@ def _pytorch_save(self, itr=None):
                 to None
         """
         if proc_id() == 0:
-
             save_fail_warning = (
                 "The object you tried to save doesn't have a 'state_dict' we can"
                 "use to retrieve the model weights. Please make sure you supplied the "
@@ -1586,7 +1583,6 @@ def store(
                 counter if global step is not supplied.
         """
         for k, v in kwargs.items():
-
             # Store variable values in epoch_dict and increase global step count
             if not (k in self.epoch_dict.keys()):
                 self.epoch_dict[k] = []
@@ -1719,7 +1715,6 @@ def log_tabular(
                 tb_alias=tb_alias,
             )
         else:
-
             v = self.epoch_dict[key]
             vals = (
                 np.concatenate(v)