Grid2op

--------------------
- [FIXED] `Issue 54 <https://github.com/Grid2Op/grid2op/issues/54>`_: Setting the bus for disconnected lines no
longer counts as a substation operation.
- [FIXED] if no redispatch actions are taken, then the game can no more invalid a provided action due to error in the
redispatching. This behavior was caused by increase / decrease of the system losses that was higher (in absolute
value) than the ramp of the generators connected to the slack bus. This has been fixed by removing the losses
of the powergrid in the computation of the redispatching algorithm. **side effect** for the generator connected
to the slack bus, the ramp min / up as well as pmin / pmax might not be respected in the results data provided
in the observation for example.
- [FIXED] a bug in the computation of cascading failure that lead (sometimes) to diverging powerflow when in the fact
the powerflow did not diverge.
- [FIXED] a bug in the `OneChangeThenNothing` agent.
- [FIXED] a bug that lead to impossibility to load a powerline after a cascading failure in some cases. Now fixed by
resetting the appropriate vectors when calling "env.reset".
- [FIXED] function `env.attach_render` that uses old names for the grid layout
- [ADDED] Remember last line buses: Reconnecting a line without providing buses will reconnect it to the buses it
was previously connected to (origin and extremity).
- [ADDED] Change lines status (aka. switch_line_status) unitary actions for subclasses of AgentWithConverter.
- [ADDED] Dispatching unitary actions for subclasses of AgentWithConverter.
- [ADDED] CombinedReward. A reward combiner to compute a weighted sum of other rewards.
- [ADDED] CloseToOverflowReward. A reward that penalize agents when lines have almost reached max capacity.
- [ADDED] DistanceReward. A reward based on how far way from the original topology the current grid is.
- [ADDED] BridgeReward. A reward based on graph connectivity, see implementation in grid2op.Reward.BridgeReward for
details

---------------------
- [BREAKING] `grid2op.GameRules` module renamed to `grid2op.RulesChecker`
- [BREAKING] `grid2op.Converters` module renamed `grid2op.Converter`
- [BREAKING] `grid2op.ChronicsHandler` renamed to `grid2op.Chronics`
- [BREAKING] `grid2op.PandaPowerBackend` is moved to `grid2op.Backend.PandaPowerBackend`
- [BREAKING] `RulesChecker.Allwayslegal` is now `Rules.Alwayslegal`
- [BREAKING] Plotting utils are now located in their own module `grid2op.Plot`
- [DEPRECATION] `HelperAction` is now called `ActionSpace` to better suit open ai gym name. Use of `HelperAction`
will be deprecated in future versions.
- [DEPRECATION] `ObservationHelper` is now called `ObservationSpace` to better suit open ai gym name.
Use of `ObservationHelper` will be deprecated in future versions.
- [DEPRECATION] `Action` class has been split into `BaseAction` that serve as an abstract base class for all
action class, and `CompleteAction` (that inherit from BaseAction) for the class allowing to perform every
modification implemented in grid2op.
- [DEPRECATION] `Observation` class has renamed `BaseObservation` that serve as an abstract base class for all
observation classes. Name Observation will be deprecated in future versions.
- [DEPRECATION] `Agent` class has renamed `BaseAgent` that serve as an abstract base class for all
agent classes. Name Agent will be deprecated in future versions.
- [DEPRECATION] `Reward` class has renamed `BaseReward` that serve as an abstract base class for all
reward classes. Name Reward will be deprecated in future versions.
- [DEPRECATION] `LegalAction` class has renamed `BaseRules` that serve as an abstract base class for all
type of rules classes. Name `LegalAction` will be deprecated in future versions.
- [DEPRECATION] typo fixed in `PreventReconection` class (now properly named `PreventReconnection`)
- [ADDED] different kind of "Opponent" can now be implemented if needed (missing deep testing, different type of
class, and good documentation)
- [ADDED] implement other "rewards" to look at. It is now possible to have an environment that will compute more rewards
that are given to the agent through the "information" return argument of `env.step`. See the documentation of
Environment.other_rewards.
- [ADDED] Alternative method to load datasets based on new dataset format: `MakeEnv.make2`
- [ADDED] Layout of the powergrid is part of the `GridObject` and is serialized along with the
action_space and observation_space. Plotting utilities no longer require specific layout (custom layout
can still be provided)
- [ADDED] A new kind of actions that can change the value (and buses) to which shunt are connected. This support will
be helpfull for the `VoltageControler` class.
- [FIXED] Loading L2RPN_2019 dataset
- [FIXED] a bug that prevents the voltage controler to be changed when using `grid2op.make`.
- [FIXED] `time_before_cooldown_line` vector were output twice in observation space
(see `issue 47 <https://github.com/Grid2Op/grid2op/issues/47>`_ part 1)
- [FIXED] the number of active bus on a substation was not computed properly, which lead to some unexpected
behavior regarding the powerlines switches (depending on current stats of powerline, changing the buses of some
powerline has different effect)
(see `issue 47 <https://github.com/Grid2Op/grid2op/issues/47>`_ part 2)
- [FIXED] wrong voltages were reported for PandapowerBackend that causes some isolated load to be not detected
(see `issue 51 <https://github.com/Grid2Op/grid2op/issues/51>`_ )
- [FIXED] improve the install script to not crash when numba can be installed, but cannot be loaded.
(see `issue 50 <https://github.com/Grid2Op/grid2op/issues/50>`_ )
- [UPDATED] import documentation of `Space` especially in case someone wants to build other type of Backend

--------------------
- [ADDED] runner now is able to show a progress bar
- [ADDED] add a "max_iter" in the runner.
- [ADDED] a repository in this github for the baseline (work in progress)
- [ADDED] include grid2Viz in a notebook (the notebook "StudyYourAgent")
- [ADDED] when a file is not present in the chronics, the chronics_handler behaves as if
nothing changes. If no files at all are provided, it raises an error.
- [ADDED] possibility to change the controler for the generator voltage setpoints
(See `VoltageControler` for more information). It can be customized as of now.
- [ADDED] lots of new tests for majority of classes (ChronicsHandler, BaseAction, Observations etc.)
- [FIXED] voltages are now set to 0 when the powerline are disconnected, instead of being set to Nan in
pandapower backend.
- [FIXED] `ReadPypowNetData` does not crash when argument "chunk_size" is provided now.
- [FIXED] some typos in the Readme
- [FIXED] some redispatching declared illegal but are in fact legal (due to
a wrong assessment) (see `issue 44 <https://github.com/Grid2Op/grid2op/issues/44>`_)
- [FIXED] reconnecting a powerline now does not count the mandatory actions on both its ends (previously you could not
reconnect a powerline with the L2RPN 2019 rules because it required acting on 2 substations) as "substation action"
- [UPDATED] add a blank environment for easier use.
- [UPDATED] now raise an error if the substations layout does not match the number of substations on the powergrid.
- [UPDATED] better handling of system without numba `issue 42 <https://github.com/Grid2Op/grid2op/issues/42>`_)
- [UPDATED] better display of the error message if all dispatchable generators are set
`issue 39 <https://github.com/Grid2Op/grid2op/issues/39>`_
- [UPDATED] change the link to the doc in the notebook to point to readthedoc and not to local documentation.
- [UPDATED] Simulate action behavior result is the same as stepping given perfect forecasts at t+1

--------------------
- [ADDED] a new environment with consistant voltages based on the case14 grid of pandapower (`case14_relistic`)
- [ADDED] a function to get the name on the element of the graphical representation.
- [ADDED] a new class to (PlotMatPlotlib) to display the grid layout and the position of the element,
as well as their name and ID
- [ADDED] possibility to read by chunk the data (memory efficiency and huge speed up at the beginning of training)
(`issue 21 <https://github.com/Grid2Op/grid2op/issues/21>`_)
- [ADDED] improved method to limit the episode length in chronics handler.
- [ADDED] a method to project some data on the layout of the grid (`GetLayout.plot_info`)
- [FIXED] a bug in the simulated reward (it was not initialized properly)
- [FIXED] add the "prod_charac.csv" for the test environment `case14_test`, `case14_redisp`, `case14_realistic` and
`case5_example`
- [FIXED] fix the display bug in the notebook of the l2rpn starting kit with the layout of the 2 buses
- [UPDATED] now attaching the layout metadata directly into the environment
- [UPDATED] `obs.simulate` now has the same code as `env.step` this include the same signature and the
possibility to simulate redispatching actions as well.
- [UPDATED] Notebook 6 to train agent more efficiently (example: prediction of actions in batch)
- [UPDATED] PlotGraph to derive from `GridObjects` allowing to be inialized at creation and not when first
observation is loaded (usable without observation)
- [UPDATED] new default environment (`case14_realistic`)
- [UPDATED] data for the new created environment.
- [UPDATED] implement redispatching action in `obs.simulate`
- [UPDATED] refactoring `Environment` and `ObsEnv` to inherit from the same base class.

--------------------
- [ADDED] Notebook 6 to explain multi environment
- [ADDED] more type of agents in the notebook 3
- [FIXED] Environment now properly built in MultiEnvironment
- [FIXED] Notebook 3 to now work with both neural network
- [FIXED] remove the "print" that displayed the path of the data used in MultiEnvironment
- [UPDATED] the action space for "IdToAct" now reduces the number of possible actions to only actions that don't
directly cause a game over.

---------------------
- [ADDED] a easier way to set the thermal limits directly from the environment (`env.set_thermal_limit`)
- [ADDED] a new environment with redispatching capabilities (`case14_redisp`) including data
- [ADDED] a new convenient script to download the dataset, run `python3 -m grid2op.download --name "case14_redisp"`
from the command line.
- [ADDED] new rewards to better take into account redispatching (`EconomicReward` and `RedispReward`)
- [ADDED] a method to check if an action is ambiguous (`act.is_ambiguous()`)
- [ADDED] a method to set more efficiently the id of the chronics used in the environment (`env.set_id`)
- [ADDED] env.step now propagate the error in "info" output (but not yet in `obs.simulate`)
- [ADDED] notebooks for redispatching (see `getting_started/5_RedispacthingAgent.ipynb`)
- [ADDED] now able to initialize a runner from a valid environment (see `env.get_params_for_runner`)
- [FIXED] reconnecting too soon a powerline is now forbidden in l2rpn2019 (added the proper legal action)
- [UPDATED] more information in the error when plotly and seaborn are not installed and trying to load the
graph of the grid.
- [UPDATED] setting an object to a busbar higher (or equal) than 2 now leads to an ambiguous action.
- [UPDATED] gitignore to really download the "prod_charac.csv" file
- [UPDATED] private member in action space and observation space (`_template_act` and `_empty_obs`)
to make it clear it's not part of the public API.
- [UPDATED] change default environment to `case14_redisp`
- [UPDATED] notebook 2 now explicitely says the proposed action is ambiguous in a python cell code (and not just
in the comments) see issue (`issue 27 <https://github.com/Grid2Op/grid2op/issues/27>`_)