(v3.3.6) - Sinergym normalization fix for DRL evaluations (#419)

* Fix evalLoggerCallback for deactivate normalization update and update mean and var calibration from training env automatically (when NormalizeObservation was activated)

* Update Sinergym version from 3.3.5 to 3.3.6

* Fix opyplus update inconsistency

* Dockerfile: Updated Ubuntu and Python version (24.04 and 3.12.3)

* Added version documentation

.devcontainer/Dockerfile

@@ -1,6 +1,6 @@
 # Base on nrel/energyplus from Nicholas Long but using 
 # Ubuntu, Python 3.10 and BCVTB
-ARG UBUNTU_VERSION=22.04
+ARG UBUNTU_VERSION=24.04
 FROM ubuntu:${UBUNTU_VERSION}
 
 # Arguments for EnergyPlus version (default values of version 8.6.0 if is not specified)
@@ -12,7 +12,8 @@ ARG ENERGYPLUS_SHA=87ed9199d4
 ARG SINERGYM_EXTRAS=[extras]
 
 # Argument for choosing Python version
-ARG PYTHON_VERSION=3.10
+ARG PYTHON_VERSION=3.12
+ENV PIP_BREAK_SYSTEM_PACKAGES=1
 
 # WANDB_API_KEY
 ARG WANDB_API_KEY

Dockerfile

@@ -1,6 +1,6 @@
 # Base on nrel/energyplus from Nicholas Long but using 
 # Ubuntu, Python 3.10 and BCVTB
-ARG UBUNTU_VERSION=22.04
+ARG UBUNTU_VERSION=24.04
 FROM ubuntu:${UBUNTU_VERSION}
 
 # Arguments for EnergyPlus version (default values of version 8.6.0 if is not specified)
@@ -12,7 +12,8 @@ ARG ENERGYPLUS_SHA=87ed9199d4
 ARG SINERGYM_EXTRAS=[extras]
 
 # Argument for choosing Python version
-ARG PYTHON_VERSION=3.10
+ARG PYTHON_VERSION=3.12
+ENV PIP_BREAK_SYSTEM_PACKAGES=1
 
 # WANDB_API_KEY
 ARG WANDB_API_KEY

INSTALL.md

@@ -13,6 +13,7 @@ The table below provides a summary of the versions supported by *Sinergym* acros
 | **1.9.5**            | **22.04 LTS**      | **3.10.6**         | 9.5.0                  | IDF                       |
 | **2.4.0**            | 22.04 LTS          | 3.10.6             | 9.5.0                  | **epJSON**                |
 | **2.5.0**            | 22.04 LTS          | 3.10.6             | **23.1.0**             | epJSON                    |
+| **3.3.6**            | **24.04 LTS**      | **3.12.3**         | 23.1.0                 | epJSON                    |
 
 We recommend always using the latest version of *Sinergym* that is supported by the container. 
 This approach helps you avoid the complexities of the installation process. However, 

docs/source/pages/installation.rst

@@ -20,6 +20,8 @@ The table below provides a summary of the versions supported by *Sinergym* acros
 +----------------------+--------------------+--------------------+------------------------+---------------------------+
 | **2.5.0**            | 22.04 LTS          | 3.10.6             | **23.1.0**             | epJSON                    |
 +----------------------+--------------------+--------------------+------------------------+---------------------------+
+| **3.3.6**            | **24.04 LTS**      | **3.12.3**         | 23.1.0                 | epJSON                    |
++----------------------+--------------------+--------------------+------------------------+---------------------------+
 
 We recommend always using the latest version of *Sinergym* that is supported by the container. 
 This approach helps you avoid the complexities of the installation process. However, 

scripts/train/train_agent.py

@@ -284,7 +284,8 @@ def process_algorithm_parameters(alg_params: dict):
     # Set up Evaluation and saving best model
     if conf.get('evaluation'):
         eval_callback = LoggerEvalCallback(
-            eval_env,
+            eval_env=eval_env,
+            train_env=env,
             best_model_save_path=eval_env.get_wrapper_attr('workspace_path') +
             '/best_model/',
             log_path=eval_env.get_wrapper_attr('workspace_path') +

sinergym/utils/callbacks.py

@@ -241,6 +241,7 @@ class LoggerEvalCallback(EventCallback):
     def __init__(
         self,
         eval_env: Union[gym.Env, VecEnv],
+        train_env: Union[gym.Env, VecEnv],
         callback_on_new_best: Optional[BaseCallback] = None,
         callback_after_eval: Optional[BaseCallback] = None,
         n_eval_episodes: int = 5,
@@ -273,6 +274,7 @@ def __init__(
         #     eval_env = DummyVecEnv([lambda: eval_env])
 
         self.eval_env = eval_env
+        self.train_env = train_env
         self.best_model_save_path = best_model_save_path
         # Logs will be written in ``evaluations.npz``
         if log_path is not None:
@@ -351,6 +353,7 @@ def _on_step(self) -> bool:
                         "Training and eval env are not wrapped the same way, "
                         "see https://stable-baselines3.readthedocs.io/en/master/guide/callbacks.html#evalcallback "
                         "and warning above.") from e
+            self._sync_envs()
 
             # Reset success rate buffer
             self._is_success_buffer = []
@@ -489,3 +492,16 @@ def update_child_locals(self, locals_: Dict[str, Any]) -> None:
         """
         if self.callback:
             self.callback.update_locals(locals_)
+
+    def _sync_envs(self):
+        # normalization
+        if is_wrapped(
+                self.train_env,
+                NormalizeObservation) and is_wrapped(
+                self.eval_env,
+                NormalizeObservation):
+            self.eval_env.get_wrapper_attr('deactivate_update')()
+            self.eval_env.get_wrapper_attr('set_mean')(
+                self.train_env.get_wrapper_attr('mean'))
+            self.eval_env.get_wrapper_attr('set_var')(
+                self.train_env.get_wrapper_attr('var'))

sinergym/utils/common.py

@@ -10,7 +10,7 @@
 import pandas as pd
 import xlsxwriter
 from eppy.modeleditor import IDF
-from opyplus.epm.record import Record
+from opyplus.epgm.record import Record
 
 import sinergym
 from sinergym.utils.constants import YEAR

sinergym/version.txt

@@ -1 +1 @@
-3.3.5
+3.3.6