fixing some bugs in the forecast env

Signed-off-by: DONNOT Benjamin <[email protected]>

Author: BDonnot

grid2op/Backend/backend.py

@@ -15,7 +15,7 @@
 from abc import ABC, abstractmethod
 import numpy as np
 import pandas as pd
-from typing import Tuple, Optional, Any, Dict, Union
+from typing import Tuple, Optional, Any, Dict, Type, Union
 
 try:
     from typing import Self
@@ -38,6 +38,7 @@
 from grid2op.Space import GridObjects, ElTypeInfo, DEFAULT_N_BUSBAR_PER_SUB, DEFAULT_ALLOW_DETACHMENT
 import grid2op.Observation  # for type hints
 import grid2op.Action  # for type hints
+import grid2op.Action._BackendAction  # for type hints
 
 
 # TODO method to get V and theta at each bus, could be in the same shape as check_kirchoff
@@ -118,8 +119,8 @@ class Backend(GridObjects, ABC):
     IS_BK_CONVERTER : bool = False
 
     # action to set me
-    my_bk_act_class : "Optional[grid2op.Action._backendAction._BackendAction]" = None
-    _complete_action_class : "Optional[grid2op.Action.CompleteAction]" = None
+    my_bk_act_class : "Optional[Type[grid2op.Action._BackendAction._BackendAction]]" = None
+    _complete_action_class : "Optional[Type[grid2op.Action.CompleteAction]]" = None
 
     ERR_INIT_POWERFLOW : str = "Power cannot be computed on the first time step, please check your data."
     ERR_DETACHMENT : str = ("One or more {} were isolated from the grid "
@@ -1972,7 +1973,7 @@ def update_from_obs(self,
         }
 
         if cls.shunts_data_available and type(obs).shunts_data_available:
-            if cls.n_shunt > 0 and "_shunt_bus" not in type(obs).attr_list_set:
+            if cls.n_shunt > 0 and "_shunt_bus" not in (type(obs).attr_list_set | set(type(obs).attr_list_json)):
                 raise BackendError(
                     "Impossible to set the backend to the state given by the observation: shunts data "
                     "are not present in the observation."
@@ -1993,6 +1994,7 @@ def update_from_obs(self,
         act.update(dict_)
         backend_action += act
         self.apply_action(backend_action)
+        backend_action.reset()  # already processed by the backend
         return backend_action
 
     def assert_grid_correct(self, _local_dir_cls=None) -> None:

grid2op/Backend/pandaPowerBackend.py

@@ -858,7 +858,6 @@ def apply_action(self, backendAction: Union["grid2op.Action._backendAction._Back
         """
         if backendAction is None:
             return
-    
         cls = type(self)
 
         (

grid2op/Environment/baseEnv.py

@@ -3550,7 +3550,11 @@ def step(self, action: BaseAction) -> Tuple[BaseObservation,
             # and this regardless of the 
             _ = action.get_topological_impact(powerline_status, _store_in_cache=True, _read_from_cache=False)
 
-            is_legal, reason = self._game_rules(action=action, env=self)
+            if self._last_obs is not None:
+                is_legal, reason = self._game_rules(action=action, env=self)
+            else:
+                is_legal = True
+                reason = None
             if not is_legal:
                 # action is replace by do nothing
                 action.reset_cache_topological_impact()
@@ -4640,11 +4644,6 @@ def _reset_to_orig_state(self, obs):
 
         # soft overflow
         self._timestep_overflow[:] = obs.timestep_overflow
-        
-        # previous "connected" state of all elements
-        self._previous_conn_state.update_from_other(obs._prev_conn)  # already done with _cst_prev_XXX
-        self._backend_action.last_topo_registered.values[:] = obs._prev_conn._topo_vect
-        # self._backend_action.current_topo.values[:] = obs._prev_conn._topo_vect
 
     def forecasts(self):
         # ensure that the "env.chronics_handler.forecasts" is called at most once per step

grid2op/Environment/environment.py

@@ -955,6 +955,9 @@ def reset_grid(self,
         If the thermal has been modified, it also modify them into the new backend.
 
         """
+        # force the first observation to be generated properly
+        self._last_obs = None
+        
         # the real powergrid of the environment
         self.backend.reset(self._init_grid_path)  
 
@@ -972,6 +975,9 @@ def reset_grid(self,
             self._backend_action.last_topo_registered.values[:] = self._init_obs._prev_conn._topo_vect
         else:
             self._backend_action = self._backend_action_class()
+        
+        if self._init_obs is not None:
+            self._reset_to_orig_state(self._init_obs)
 
         init_action = None
         if not self._parameters.IGNORE_INITIAL_STATE_TIME_SERIE:
@@ -983,9 +989,15 @@ def reset_grid(self,
         else:
             # do as if everything was connected to busbar 1
             # TODO logger: log that
-            init_action = self._helper_action_env({"set_bus": np.ones(type(self).dim_topo, dtype=dt_int)})
-            if type(self).shunts_data_available:
-                init_action += self._helper_action_env({"shunt": {"set_bus": np.ones(type(self).n_shunt, dtype=dt_int)}})
+            if self._init_obs is None:
+                init_action = self._helper_action_env({"set_bus": np.ones(type(self).dim_topo, dtype=dt_int)})
+                if type(self).shunts_data_available:
+                    init_action += self._helper_action_env({"shunt": {"set_bus": np.ones(type(self).n_shunt, dtype=dt_int)}})
+            else:
+                init_action = self._helper_action_env({"set_bus": self._init_obs.topo_vect})
+                if type(self).shunts_data_available:
+                    init_action += self._helper_action_env({"shunt": {"set_bus": self._init_obs._shunt_bus}})
+                    
         if init_action is None:
             # default behaviour for grid2op < 1.10.2
             init_action = self._helper_action_env({})
@@ -1003,7 +1015,7 @@ def reset_grid(self,
                 raise Grid2OpException(f"kwargs `method` used to set the initial state of the grid "
                                        f"is not understood (use one of `combine` or `ignore` and "
                                        f"not `{method}`)")
-        init_action._set_topo_vect.nonzero()
+                
         *_, fail_to_start, info = self.step(init_action)
         if fail_to_start:
             raise Grid2OpException(
@@ -1383,6 +1395,7 @@ def reset(self,
 
         if options is not None and "init datetime" in options:
             self.chronics_handler.set_current_datetime(options["init datetime"])
+            
         self.reset_grid(init_state, method)
         if self.viewer_fig is not None:
             del self.viewer_fig
@@ -1423,11 +1436,6 @@ def reset(self,
         # and reset also the "simulated env" in the observation space
         self._observation_space.reset(self)
         self._observation_space.set_real_env_kwargs(self)
-
-        self._last_obs = None  # force the first observation to be generated properly
-        
-        if self._init_obs is not None:
-            self._reset_to_orig_state(self._init_obs)
         return self.get_obs()
 
     def render(self, mode="rgb_array"):

grid2op/tests/test_forecast_env.py

@@ -32,9 +32,8 @@ def _aux_normal_obs(self, obs, line_id=0):
         for_env = obs.get_forecast_env()
         for_obs = for_env.reset()
         assert (for_obs.topo_vect == obs.topo_vect).all(), f"{(for_obs.topo_vect != obs.topo_vect).nonzero()}"
-        
         for_obs = for_env.reset(options={"init state": {"set_line_status": [(line_id, -1)]}})
-        assert (for_obs.topo_vect != obs.topo_vect).sum() == 2
+        assert (for_obs.topo_vect != obs.topo_vect).sum() == 2, f"{(for_obs.topo_vect != obs.topo_vect).nonzero()}"
         assert for_obs.topo_vect[type(self.env).line_or_pos_topo_vect[line_id]] == -1
         assert for_obs.topo_vect[type(self.env).line_ex_pos_topo_vect[line_id]] == -1