[WIP] Add initial support for swtiches

.circleci/config.yml

@@ -31,7 +31,7 @@ executors:
 jobs:
     test:
         executor: grid2op-executor
-        resource_class: medium+
+        resource_class: large
         parallelism: 4
         steps:
             - checkout

.github/workflows/main.yml

@@ -311,11 +311,68 @@ jobs:
       - name: Test the automatic generation of classes in the env folder
         run: |
           python -m unittest grid2op/tests/automatic_classes.py -v -f
+
+  detailed_topology:
+    name: Test detailed topology on ${{ matrix.config.name }}
+    runs-on: ${{ matrix.config.os }}
+    strategy:
+      matrix:
+        config:
+          # - {
+          #   name: darwin,
+          #   os: macos-latest,
+          # }
+          # - {
+          #   name: windows,
+          #   os: windows-2019,
+          # }
+          - { 
+            name: ubuntu,
+            os: ubuntu-latest,
+          }
+        python:
+          - {
+            name: cp39,
+            version: '3.9',
+          }  
+          - {
+            name: cp313,
+            version: '3.13',
+          }  
+
+    steps:
+
+      - name: Checkout sources
+        uses: actions/checkout@v3
+        with:
+          submodules: true
+
+      - name: Setup Python
+        uses: actions/setup-python@v2
+        with:
+          python-version: ${{ matrix.python.version }}
+
+      - name: Install Python dependencies
+        run: |
+          python -m pip install --upgrade pip wheel setuptools gymnasium build
+
+      - name: Build wheel
+        run: python -m build --wheel .
+
+      - name: Install wheel
+        shell: bash
+        run: |
+          python -m pip install dist/*.whl --user
+          pip freeze
+
+      - name: Test the working of the detailed topo representation
+        run: |
+          python -m unittest grid2op/tests/indep_test_detailed_topo_extended.py -f
 
   package:
     name: Test install
     runs-on: ubuntu-latest
-    needs: [manylinux_build, macos_windows_build, auto_class_in_file]
+    needs: [manylinux_build, macos_windows_build, auto_class_in_file, detailed_topology]
 
     steps:
       - name: Download wheels

.gitignore

@@ -416,7 +416,17 @@ getting_started/venv_310_ray/
 grid2op/tests/venv_test_autoclass/
 test_eduardo.py
 grid2op/tests/failed_test*
+grid2op/tests/ampl_from_iidm.py
+grid2op/tests/do_I_repeat.txt
+grid2op/tests/parse_res35_working.txt
+grid2op/tests/parse_res_35 (copie).txt
+grid2op/tests/parse_res_35.txt
+grid2op/tests/res_algo_topo_35.txt
+grid2op/tests/test_topo_ampl/
+grid2op/tests/test_topo_ampl2/
 venv_312
+venv_*
+grid2op/tests/internal_*.pickle
 grid2op_test_issue_discord.ipynb
 grid2op_test_issue_discord_orig.ipynb
 grid_bug_pp3.json

CHANGELOG.rst

@@ -100,6 +100,19 @@ Native multi agents support:
 - add detachment
 - add change_bus / set_bus
 
+[1.13.0] - DETAILED TOPO
+---------------------------
+- [FIXED] some issues with gym classes
+- [IMPROVED] the serialization of class definition 
+  when using "automatic_classes" or "experimental_read_from_local_dir"
+- [IMPROVED] when using gym MultiDiscrete as an action space,
+  if the dimension is 0 (for one of the MultiDiscrete) then it is now
+  removed from the action space (for example, if your environment does not
+  contain storage unit, the corresponding MultiDiscrete will have one less 
+  dimension, because you cannot act on storage unit.)
+- [IMPROVED] classes for action and observation classes are now easier to 
+  initialize (added the `finalize_class_definition` classmethod)
+
 [1.12.2] - 2025-10-xx
 ----------------------
 - [FIXED] an issue preventing to change the way
@@ -296,9 +309,13 @@ Native multi agents support:
 
 [1.10.4] - 2024-10-14
 -------------------------
-- [FIXED] an issue in the backend: if the backend failed to be
+- [FIXED] an issue in the backend: if the backend failed to be copied
   created the `_grid` attribute was set to `None` and not set back to
+  its original value in the copied backend.
 - [FIXED] the `self.skip_if_needed()` was missing for one of the test suite.
+- [FIXED] the correct `AmbiguousAction` is now raised when grid2op does not understand
+  what an action should be doing (an incorrect `IllegalAction` used to be sent)
+- [FIXED] a test in `test_ActionProperties` did not test the correct property
 - [FIXED] an error in the descirption of the `educ_case14_storage` environment
   (wrong sign for the slack generator)
 - [FIXED] the environment would not load in case of an incorrect "layout.json"
@@ -311,6 +328,7 @@ Native multi agents support:
   in the readme)
 - [ADDED] numpy 2 support (now that pandapower allows it)
 - [IMPROVED] error message when forecasts are not correctly set-up
+- [IMPROVED] an error message when loading a grid with forecasts
 
 
 [1.10.3] - 2024-07-12
@@ -381,7 +399,6 @@ Native multi agents support:
   processes (using `fork` at least), see `ep_data.make_serializable`
 - [IMPROVED] chronix2grid tests are now done independantly on the CI
 
-
 [1.10.2] - 2024-05-27
 -------------------------
 - [BREAKING] the `runner.run_one_episode` now returns an extra argument (first position): 
@@ -1168,7 +1185,7 @@ Native multi agents support:
 - [ADDED]: function to retrieve the maximum duration of the current episode.
 - [ADDED]: a new kind of opponent that is able to attack at "more random" times with "more random" duration.
   See the `GeometricOpponent`.
-- [IMPROVED]: on windows at least, grid2op does not work with gym < 0.17.2 Checks are performed in order to make sure
+- [IMPROVED]: on windows at least, grid2op does not work with `gym < 0.17.2` Checks are performed in order to make sure
   the installed open ai gym package meets this requirement (see issue
   `Issue#185 <https://github.com/Grid2Op/grid2op/issues/185>`_ )
 - [IMPROVED] the seed of openAI gym for composed action space (see issue `https://github.com/openai/gym/issues/2166`):

_profiling/profiler_switch.py

@@ -0,0 +1,135 @@
+# Copyright (c) 2019-2020, RTE (https://www.rte-france.com)
+# See AUTHORS.txt
+# This Source Code Form is subject to the terms of the Mozilla Public License, version 2.0.
+# If a copy of the Mozilla Public License, version 2.0 was not distributed with this file,
+# you can obtain one at http://mozilla.org/MPL/2.0/.
+# SPDX-License-Identifier: MPL-2.0
+# This file is part of Grid2Op, Grid2Op a testbed platform to model sequential decision making in power systems.
+
+"""
+This file should be used to assess the performance of grid2op in "runner" mode: the loading time are not studied,
+neither are the import times.
+
+Data are loaded only once, when the environment is "done" the programm stops.
+
+This corresponds to the situation: you have a trained agent, you want to assess its performance using the runner
+"""
+
+import os
+
+from grid2op import make
+from grid2op.Action import CompleteAction
+from grid2op.Space import AddDetailedTopoIEEE
+from grid2op.Parameters import Parameters
+from grid2op.Agent import BaseAgent
+from grid2op.Rules import AlwaysLegal
+from grid2op.Backend import PandaPowerBackend
+import cProfile
+
+from utils_benchmark import run_env, str2bool
+
+try:
+    from lightsim2grid import LightSimBackend
+    light_sim_avail = True
+except ImportError:
+    light_sim_avail = False
+
+ENV_NAME = "rte_case5_example"
+ENV_NAME = "rte_case14_realistic"
+MAX_TS = 1000
+RANDOM_SWITCH_AGENT = False  # do not change this
+# for profiling it has no impact.
+
+
+class RandomBusSwitchAgent(BaseAgent):
+    def act(self, observation, reward, done=False):
+        sub_id = self.space_prng.randint(type(self.action_space).n_sub)
+        new_position = self.space_prng.choice([-1, 1], size=1)
+        return self.action_space({"set_switch": [(sub_id, new_position)]})
+
+
+class RandomElSwitchAgent(BaseAgent):
+    def act(self, observation, reward, done = False):
+        if observation.current_step <= 1:
+            # deactivate the connector between busbars
+            res = self.action_space({"set_switch": [(sub_id, -1)
+                                                    for sub_id in range(type(self.action_space).n_sub)]})
+        else:
+            res = self.action_space()
+
+        do_act = self.space_prng.choice([0, 1], size=1, p = [0.95, 0.05])
+        if not do_act:
+            return res
+
+        n_switch = type(self.action_space).detailed_topo_desc.switches.shape[0]
+        switch_id = self.space_prng.randint(type(self.action_space).n_sub, 
+                                            n_switch,
+                                            size=1)
+        new_position = self.space_prng.choice([-1, 1], size=1)
+        res.update({"set_switch": [(switch_id, new_position)]})
+        return res
+
+def main(max_ts, name, use_lightsim=False, test_env=True):
+    param = Parameters()
+    if use_lightsim:
+        if light_sim_avail:
+            class ThisBackendClass(AddDetailedTopoIEEE, LightSimBackend):
+                pass
+        else:
+            raise RuntimeError("LightSimBackend not available")
+    else:
+        class ThisBackendClass(AddDetailedTopoIEEE, PandaPowerBackend):
+            pass
+
+    backend = ThisBackendClass()
+
+    param.init_from_dict(
+        {"NO_OVERFLOW_DISCONNECTION": True,
+         "STOP_EP_IF_GEN_BREAK_CONSTRAINTS": False,
+         "ENV_DOES_REDISPATCHING": False})
+
+    env_klu = make(name,
+                   backend=backend,
+                   param=param,
+                   gamerules_class=AlwaysLegal,
+                   test=test_env,
+                   action_class=CompleteAction)
+    assert type(env_klu).detailed_topo_desc is not None
+    agent = RandomBusSwitchAgent(action_space=env_klu.action_space)
+    if RANDOM_SWITCH_AGENT:
+        agent = RandomElSwitchAgent(action_space=env_klu.action_space)
+    agent.seed(4)
+
+    cp = cProfile.Profile()
+    cp.enable()
+    nb_ts_klu, *_, time_step = run_env(env_klu, max_ts, agent)
+    cp.disable()
+    print(f'Time for {nb_ts_klu} steps: {time_step} => {time_step / nb_ts_klu} s/step or {nb_ts_klu / time_step:.3e} step / s')
+    nm_f, ext = os.path.splitext(__file__)
+    nm_out = "{}_{}_{}.prof".format(nm_f, "lightsim" if use_ls else "pp", name)
+    cp.dump_stats(nm_out)
+    print("You can view profiling results with:\n\tsnakeviz {}".format(nm_out))
+
+
+if __name__ == "__main__":
+    import argparse
+    parser = argparse.ArgumentParser(description='Benchmark pyKLU and Pandapower Backend for an agent that takes every '
+                                                 'topological action possible')
+    parser.add_argument('--name', default=ENV_NAME, type=str,
+                        help='Environment name to be used for the benchmark.')
+    parser.add_argument('--number', type=int, default=MAX_TS,
+                        help='Maximum number of time steps for which the benchamark will be run.')
+    parser.add_argument("--use_ls", type=str2bool, nargs='?',
+                        const=True, default=False,
+                        help="Use the LightSim2Grid Backend.")
+    parser.add_argument("--no_test", type=str2bool, nargs='?',
+                        const=True, default=False,
+                        help="Do not use a test environment for the profiling (default to False: meaning you use a test env)")
+
+    args = parser.parse_args()
+
+    max_ts = int(args.number)
+    name = str(args.name)
+    use_ls = args.use_ls
+    test_env = not args.no_test
+    main(max_ts, name, use_lightsim=use_ls, test_env=test_env)

_profiling/utils_benchmark.py

@@ -10,6 +10,7 @@
 import numpy as np
 from tqdm import tqdm
 import argparse
+from grid2op.Environment import Environment
 from grid2op.Agent import AgentWithConverter
 from grid2op.Converter import IdToAct
 
@@ -179,12 +180,12 @@ def print_res(env_klu, env_pp,
     print("Absolute value of the difference for gen_q: {}".format(np.max(np.abs(gen_q_klu - gen_q_pp))))
 
 
-def run_env(env, max_ts, agent):
+def run_env(env: Environment, max_ts, agent):
     nb_rows = min(env.chronics_handler.max_timestep(), max_ts)
     aor = np.zeros((nb_rows, env.n_line))
     gen_p = np.zeros((nb_rows, env.n_gen))
     gen_q = np.zeros((nb_rows, env.n_gen))
-    obs = env.get_obs()
+    obs = env.get_obs(_do_copy=False)
     done = False
     reward = env.reward_range[0]
     nb_ts = 0
@@ -196,6 +197,7 @@ def run_env(env, max_ts, agent):
             act = agent.act(obs, reward, done)
             beg_step = time.perf_counter()
             obs, reward, done, info = env.step(act)
+            assert not info["is_ambiguous"], info["exception"]
             time_step += time.perf_counter() - beg_step
             aor[nb_ts, :] = obs.a_or
             gen_p[nb_ts, :] = obs.prod_p

docs/detailed_topology.rst

@@ -0,0 +1,50 @@
+.. _detailed-topology-modeling-module:
+
+Dive into the detailed topology "modeling" in grid2op
+===================================================================
+
+.. warning::
+  Work in progress
+
+What is a "detailed" topology in grid2op
+-----------------------------------------
+
+.. warning::
+  Work in progress
+
+
+- Concept of connectivity nodes
+- Switches
+- Processing of the switches to "original topoolgy"
+- 
+
+Impact in grid2op
+------------------
+
+.. warning::
+  Work in progress
+
+- new action
+- new observation
+
+.. danger::
+    Be carefull with convertion fo / from switches !!!
+
+Why did we add it ?
+--------------------
+
+.. warning::
+  Work in progress
+
+What features are actually implemented ?
+-----------------------------------------
+
+.. warning::
+  Work in progress
+
+
+Pros and cons of using it ?
+---------------------------------
+.. warning::
+  Work in progress
+

docs/index.rst

@@ -110,6 +110,7 @@ graph.
    modeled_elements
    grid_graph
    topology
+   detailed_topology
 
 Environments
 ---------------