Merge pull request #605 from BDonnot/bd_dev

Add 90% of the "init state" feature

.circleci/config.yml

@@ -164,7 +164,7 @@ jobs:
                    grid2op.testinstall
 
     legacy_lightsim:               
-        executor: python38
+        executor: python38  # needs to be 38: whl of lightsim were not released for 3.10 at the time
         resource_class: small
         steps:
             - checkout

.github/workflows/main.yml

@@ -64,15 +64,15 @@ jobs:
 
       - name: Install wheel
         run: |
-          pip3 install dist/*.whl --user
+          pip3 install dist/*.whl
           pip freeze
 
       - name: Check package can be imported
         run: |
           python3 -c "import grid2op"
           python3 -c "from grid2op import *"
           python3 -c "from grid2op.Action._backendAction import _BackendAction"
-
+          
       - name: Upload wheel
         uses: actions/upload-artifact@v2
         with:

.gitignore

@@ -403,6 +403,11 @@ grid2op/tests/requirements.txt
 grid2op/tests/venv_test_311/
 issue_577/
 junk.py
+grid2op/tests/20240429_failed_tests.txt
+grid2op/tests/20240429_failed_tests_small.txt
+grid2op/tests/20240429_teq_test.txt
+grid2op/tests/req_38_np121
+test_make_2_envs.py
 
 # profiling files
 **.prof

CHANGELOG.rst

@@ -33,17 +33,28 @@ Change Log
 
 [1.10.2] - 2024-xx-yy
 -------------------------
+- [BREAKING] the `runner.run_one_episode` now returns an extra first argument: 
+  `chron_id, chron_name, cum_reward, timestep, max_ts = runner.run_one_episode()` which 
+  is consistant with `runner.run(...)` (previously it returned only 
+  `chron_name, cum_reward, timestep, max_ts = runner.run_one_episode()`)
+- [BREAKING] the runner now has no `chronics_handler` attribute (`runner.chronics_handler` 
+  is not defined)
 - [ADDED] it is now possible to call `change_reward` directly from 
   an observation (no need to do it from the Observation Space)
 - [ADDED] method to change the reward from the observation (observation_space
   is not needed anymore): you can use `obs.change_reward`
 - [ADDED] a way to automatically set the `experimental_read_from_local_dir` flags 
   (with automatic class creation). For now it is disable by default, but you can 
   activate it transparently (see doc)
-- [ADDED] TODO the possibility to set the grid in an initial state (using an action) TODO
+- [ADDED] possibility to set the grid to an initial state (using an action) when using the
+  "time series" classes. The supported classes are `GridStateFromFile` - and all its derivative, 
+  `FromOneEpisodeData`, `FromMultiEpisodeData`, `FromNPY` and `FromHandlers`. The classes `ChangeNothing`
+  and `FromChronix2grid` are not supported at the moment.
+- [ADDED] an "Handler" (`JSONInitStateHandler`) that can set the grid to an initial state (so as to make
+  compatible the `FromHandlers` time series class with this new feature)
 - [FIXED] a small issue that could lead to having 
   "redispatching_unit_commitment_availble" flag set even if the redispatching
-  data was not loded correctly
+  data was not loaded correctly
 - [FIXED] EducPandaPowerBackend now properly sends numpy array in the class attributes
   (instead of pandas series)
 - [FIXED] an issue when loading back data (with `EpisodeData`): when there were no storage units
@@ -52,6 +63,11 @@ Change Log
   grid layout was set
 - [FIXED] notebook 5 on loading back data with `EpisodeData`.
 - [FIXED] converter between backends (could not handle more than 2 busbars)
+- [FIXED] a bug in `BaseMultiProcessEnvironment`: set_filter had no impact
+- [FIXED] an issue in the `Runner` (`self.chronics_handler` was sometimes used, sometimes not
+  and most of the time incorrectly)
+- [FIXED] on `RemoteEnv` class (impact all multi process environment): the kwargs used to build then backend
+  where not used which could lead to"wrong" backends being used in the sub processes.
 - [IMPROVED] documentation about `obs.simulate` to make it clearer the 
   difference between env.step and obs.simulate on some cases
 - [IMPROVED] type hints on some methods of `GridObjects`

docs/conf.py

@@ -22,7 +22,7 @@
 author = 'Benjamin Donnot'
 
 # The full version, including alpha/beta/rc tags
-release = '1.10.2.dev1'
+release = '1.10.2.dev2'
 version = '1.10'
 
 

grid2op/Action/actionSpace.py

@@ -10,6 +10,8 @@
 import copy
 from typing import Dict, List, Any, Literal
 
+import grid2op
+from grid2op.typing_variables import DICT_ACT_TYPING
 from grid2op.Action.baseAction import BaseAction
 from grid2op.Action.serializableActionSpace import SerializableActionSpace
 
@@ -74,22 +76,9 @@ def __init__(
 
     def __call__(
         self,
-        dict_: Dict[Literal["injection",
-                            "hazards",
-                            "maintenance",
-                            "set_line_status",
-                            "change_line_status",
-                            "set_bus", 
-                            "change_bus",
-                            "redispatch",
-                            "set_storage",
-                            "curtail",
-                            "raise_alarm",
-                            "raise_alert"], Any] = None,
+        dict_: DICT_ACT_TYPING = None,
         check_legal: bool = False,
-        env: "grid2op.Environment.BaseEnv" = None,
-        *,
-        injection=None,
+        env: "grid2op.Environment.BaseEnv" = None
     ) -> BaseAction:
         """
         This utility allows you to build a valid action, with the proper sizes if you provide it with a valid
@@ -132,9 +121,15 @@ def __call__(
             An action that is valid and corresponds to what the agent want to do with the formalism defined in
             see :func:`Action.udpate`.
 
+        Notes
+        -----
+        
+        This function is not in the "SerializableActionSpace" because the
+        "legal_action" is not serialized. TODO ?
+        
         """
         # build the action
-        res = self.actionClass()
+        res : BaseAction = self.actionClass()
 
         # update the action
         res.update(dict_)

grid2op/Action/baseAction.py

@@ -10,13 +10,17 @@
 import numpy as np
 import warnings
 from typing import Tuple, Dict, Literal, Any, List
+
+
 try:
     from typing import Self
 except ImportError:
     from typing_extensions import Self
 
 from packaging import version
 
+import grid2op
+from grid2op.typing_variables import DICT_ACT_TYPING
 from grid2op.dtypes import dt_int, dt_bool, dt_float
 from grid2op.Exceptions import *
 from grid2op.Space import GridObjects
@@ -624,6 +628,7 @@ def as_serializable_dict(self) -> dict:
 
         """
         res = {}
+        cls = type(self)
         # bool elements
         if self._modif_alert:
             res["raise_alert"] = [
@@ -645,7 +650,7 @@ def as_serializable_dict(self) -> dict:
             self._aux_serialize_add_key_change("gen_change_bus", "generators_id", res["change_bus"])
             self._aux_serialize_add_key_change("line_or_change_bus", "lines_or_id", res["change_bus"])
             self._aux_serialize_add_key_change("line_ex_change_bus", "lines_ex_id", res["change_bus"])
-            if hasattr(type(self), "n_storage") and type(self).n_storage:
+            if hasattr(cls, "n_storage") and cls.n_storage:
                 self._aux_serialize_add_key_change("storage_change_bus", "storages_id", res["change_bus"])
             if not res["change_bus"]:
                 del res["change_bus"]
@@ -664,7 +669,7 @@ def as_serializable_dict(self) -> dict:
             self._aux_serialize_add_key_set("gen_set_bus", "generators_id", res["set_bus"])
             self._aux_serialize_add_key_set("line_or_set_bus", "lines_or_id", res["set_bus"])
             self._aux_serialize_add_key_set("line_ex_set_bus", "lines_ex_id", res["set_bus"])
-            if hasattr(type(self), "n_storage") and type(self).n_storage:
+            if hasattr(cls, "n_storage") and cls.n_storage:
                 self._aux_serialize_add_key_set("storage_set_bus", "storages_id", res["set_bus"])
             if not res["set_bus"]:
                 del res["set_bus"]
@@ -715,7 +720,7 @@ def as_serializable_dict(self) -> dict:
             if not res["injection"]:
                 del res["injection"]
 
-        if type(self).shunts_data_available:
+        if cls.shunts_data_available:
             res["shunt"] = {}
             if np.isfinite(self.shunt_p).any():
                 res["shunt"]["shunt_p"] = [
@@ -2094,11 +2099,31 @@ def _digest_redispatching(self, dict_):
 
     def _digest_storage(self, dict_):
         if "set_storage" in dict_:
-            self.storage_p = dict_["set_storage"]
-
+            try:
+                self.storage_p = dict_["set_storage"]
+            except IllegalAction as exc_:
+                cls = type(self)
+                # only raise the error if I am not in compat mode
+                if cls.glop_version == grid2op.__version__:
+                    raise exc_
+                else:
+                    # TODO be more specific on the version
+                    warnings.warn(f"Ignored error on storage units, because "
+                                  f"you are in a backward compatibility mode.")
+
     def _digest_curtailment(self, dict_):
         if "curtail" in dict_:
-            self.curtail = dict_["curtail"]
+            try:
+                self.curtail = dict_["curtail"]
+            except IllegalAction as exc_:
+                cls = type(self)
+                # only raise the error if I am not in compat mode
+                if cls.glop_version == grid2op.__version__:
+                    raise exc_
+                else:
+                    # TODO be more specific on the version
+                    warnings.warn(f"Ignored error on curtailment, because "
+                                  f"you are in a backward compatibility mode.")
 
     def _digest_alarm(self, dict_):
         """
@@ -2125,7 +2150,7 @@ def _reset_vect(self):
         self._subs_impacted = None
         self._lines_impacted = None
 
-    def update(self, dict_):
+    def update(self, dict_: DICT_ACT_TYPING):
         """
         Update the action with a comprehensible format specified by a dictionary.
 
@@ -6410,7 +6435,6 @@ def decompose_as_unary_actions(self,
                     tmp += a
             assert tmp == act
         
-        
         Parameters
         ----------
         group_topo : bool, optional
@@ -6473,3 +6497,56 @@ def decompose_as_unary_actions(self,
         if self._modif_curtailment:
             self._aux_decompose_as_unary_actions_curtail(cls, group_curtail, res)
         return res
+
+    def _add_act_and_remove_line_status_only_set(self, other: "BaseAction") -> "BaseAction":
+        """INTERNAL
+        
+        This is used by the environment when combining action in the "set state" in env.reset.
+        
+        It supposes both self and other are only "set" actions
+        
+        .. versionadded:: 1.10.2
+        
+        """
+        self += other
+        cls = type(self)
+        # switch off in self the element disconnected in other
+        switched_off = other._set_line_status == -1
+        switched_off |= other._set_topo_vect[cls.line_or_pos_topo_vect] == -1
+        switched_off |= other._set_topo_vect[cls.line_ex_pos_topo_vect] == -1
+        self._set_topo_vect[cls.line_or_pos_topo_vect[switched_off]] = -1
+        self._set_topo_vect[cls.line_ex_pos_topo_vect[switched_off]] = -1
+        self._set_line_status[switched_off] = -1
+
+        # switch on in self the element reconnected in other
+        switched_on = other._set_line_status == 1
+        switched_on |= other._set_topo_vect[cls.line_or_pos_topo_vect] > 0
+        switched_on |= other._set_topo_vect[cls.line_ex_pos_topo_vect] > 0
+        self._set_line_status[switched_on] = 1
+        # "reconnect" object through topo vect
+        topo_vect = other._set_topo_vect > 0
+        self._set_topo_vect[topo_vect] = other._set_topo_vect[topo_vect]
+
+        if (self._set_line_status != 0).any():
+            self._modif_set_status = True
+        if (self._set_topo_vect != 0).any():
+            self._modif_set_bus = True
+        return self
+
+    def remove_change(self) -> "BaseAction":
+        """This function will modify 'self' and remove all "change" action type. 
+        
+        It is mainly used in the environment, when removing the "change" type for setting the original
+        state of the grid.
+        
+        .. versionadded:: 1.10.2
+        
+        """
+        if self._change_bus_vect.any():
+            warnings.warn("This action modified the buses with `change_bus` ")
+            self._change_bus_vect[:] = False
+            self._modif_change_bus = False
+        if self._switch_line_status.any():
+            self._switch_line_status[:] = False
+            self._modif_change_status = False
+        return self

grid2op/Agent/oneChangeThenNothing.py

@@ -6,6 +6,7 @@
 # SPDX-License-Identifier: MPL-2.0
 # This file is part of Grid2Op, Grid2Op a testbed platform to model sequential decision making in power systems.
 
+import warnings
 from grid2op.Agent.baseAgent import BaseAgent
 
 
@@ -29,6 +30,11 @@ class OneChangeThenNothing(BaseAgent):
 
     .. code-block:: python
 
+        # This class has been deprecated, please use the env.reset()
+        # with proper options instead
+        
+        
+        DEPRECATED ! 
         import grid2op
         from grid2op.Agent import OneChangeThenNothing
         acts_dict_ = [{}, {"set_line_status": [(0,-1)]}]  # list of dictionaries. Each dictionary
@@ -44,12 +50,35 @@ class OneChangeThenNothing(BaseAgent):
             # run 2 episode with it
             res_2 = runner.run(nb_episode=2)
 
+    Notes:
+    ------
+    
+    After grid2op 1.10.2, this class has been deprecated. A cleaner alternative
+    to use it is to set the initial state of grid when calling `env.reset` like this:
+    
+    .. code-block:: python
+    
+        import grid2op
+        
+        env = grid2op.make("l2rpn_case14_sandbox")  # create an environment
+        dict_act_json = ...  # dict representing an action
+        obs = env.reset(options={"init state": dict_act_json})
+    
+    This way of doing offers:
+    
+    - more flexibility: rules are not checked
+    - more flexibility: any type of actions acting on anything can be performed
+      (even if the action would be illegal for the agent)
+    - less trouble: cooldown are not affected
+    
     """
 
     my_dict = {}
 
     def __init__(self, action_space):
         BaseAgent.__init__(self, action_space)
+        cls = type(self)
+        warnings.warn(f"Deprecated class, please use `env.reset(options={{'init state': {self.action_space(cls.my_dict).to_json()}, 'method': 'ignore' }})` instead")
         self.has_changed = False
         self.do_nothing_action = self.action_space({})