Note
This section refers to the technical application. If you are looking for information regarding the status of this project and the original repo, please look here.
Wraps the open-source Stockfish chess engine for easy integration into python.
$ pip install stockfishPlease note that as this is a third-party library, you'll also need to install the Stockfish engine in some way. See various options on their site.
See API Documentation for more information.
- The codebase is compatible with Python 3.10+
- Run
pip install -r requirements.txtto get all necessary dependencies. - Pytest is used to run the unit tests in
/tests - For linting use
pre-commitby runningpre-commit installonce and the pre-commit hooks will be executed automatically
You should install the stockfish engine in your operating system globally or specify path to binary file in class constructor
from stockfish import Stockfish
stockfish = Stockfish(path="/Users/zhelyabuzhsky/Work/stockfish/stockfish-9-64")There are some default engine settings used by this wrapper. For increasing Stockfish's strength and speed, the "Threads" and "Hash" parameters can be modified (note that the latter shouldn't be set before the former).
{
"Debug Log File": "",
"Contempt": 0,
"Min Split Depth": 0,
"Threads": 1, # More threads will make the engine stronger, but should be kept at less than the number of logical processors on your computer.
"Ponder": False,
"Hash": 16, # Default size is 16 MB. It's recommended that you increase this value, to however many MBs of RAM you're willing to allocate (e.g., 2048 for 2GB of RAM).
"MultiPV": 1,
"Skill Level": 20,
"Move Overhead": 10,
"Minimum Thinking Time": 20,
"Slow Mover": 100,
"UCI_Chess960": False,
"UCI_LimitStrength": False,
"UCI_Elo": 1350
}You can change them, as well as the default search depth, during your Stockfish class initialization:
stockfish = Stockfish(path="/Users/zhelyabuzhsky/Work/stockfish/stockfish-9-64", depth=18, parameters={"Threads": 2, "Minimum Thinking Time": 30})These parameters can also be updated at any time by calling the "update_engine_parameters" function:
stockfish.update_engine_parameters({"Hash": 2048, "UCI_Chess960": True}) # Gets stockfish to use a 2GB hash table, and also to play Chess960.As for the depth, it can also be updated, by using the following function. Note that if you don't set depth to a value yourself, the python module will initialize it to 15 by default.
stockfish.set_depth(12)When you're done using the Stockfish engine process, you can send the "quit" uci command to it with:
stockfish.send_quit_command()The __del__() method of the Stockfish class will call send_quit_command(), but it's technically not guaranteed python will call __del__() when the Stockfish object goes out of scope. So even though it'll probably not be needed, it doesn't hurt to call send_quit_command() yourself.
stockfish.make_moves_from_start(["e2e4", "e7e6"])If you'd just like to set up the starting position without making any moves from it, just call this function without sending an argument:
stockfish.make_moves_from_start()This function takes a Sequence[str] as its argument. Each string represents a move, and must have the format of the starting coordinate followed by the ending coordinate. If a move leads to a pawn promoting, then an additional character must be appended at the end (to indicate what piece the pawn promotes into).
Other types of special moves (e.g., checks, captures, checkmates, en passants) do not need any special notation; the starting coordinate followed by the ending coordinate is all the information that's needed. Note that castling is represented by the starting coordinate of the king followed by the ending coordinate of the king. So "e1g1" would be used for white castling kingside, assuming the white king is still on e1 and castling is legal.
Example call (assume in the current position, it is White's turn):
stockfish.make_moves_from_current_position(["g4d7", "a8b8", "f1d1", "b2b1q"]) # Moves the white piece on g4 to d7, then the black piece on a8 to b8, then the white piece on f1 to d1, and finally pushes the black b2-pawn to b1, promoting it into a queen.Note that if you want to play Chess960, it's recommended you first update the "UCI_Chess960" engine parameter to be True, before calling set_fen_position.
stockfish.set_fen_position("rnbqkbnr/pppp1ppp/4p3/8/4P3/8/PPPP1PPP/RNBQKBNR w KQkq - 0 2")This function returns a bool saying whether the passed in FEN is valid (both syntax wise and whether the position represented is legal). The function isn't perfect and won't catch all cases, but generally it should return the correct answer. For example, one exception is positions which are legal, but have no legal moves. I.e., for checkmates and stalemates, this function will incorrectly say the fen is invalid.
Note that the function checks whether a position is legal by temporarily creating a new Stockfish process, and then seeing if it can return a best move (and also not crash). Whatever the outcome may be though, this temporary SF process should terminate after the function call.
stockfish.is_fen_valid("rnbqkbnr/pppppppp/8/8/8/8/PPPPPPPP/RNBQKBNR w KQkq - 0 1")True
stockfish.is_fen_valid("rnbqkbnr/pppppppp/8/8/8/8/PPPPPPPP/RNBQKBNR w KQkq -") # will return False, in this case because the FEN is missing two of the six required fields.False
stockfish.get_best_move()d2d4
It's possible to specify remaining time on black and/or white clock. Time is in milliseconds.
stockfish.get_best_move(wtime=1000, btime=1000)stockfish.get_best_move_time(1000)Time constraint is in milliseconds
e2e4
Returns True if the passed in move is legal in the current position.
stockfish.is_move_correct('a2a3')True
Returns a list of dictionaries, where each dictionary represents a move's info. Each dictionary will contain a value for the 'Move' key, and either the 'Centipawn' or 'Mate' value will be a number (the other will be None).
Positive values mean advantage White, negative values mean advantage Black (unless you're using the turn perspective option, in which case positive is for the side to move).
Note that if you have stockfish on a weaker elo or skill level setting, the top moves returned by this function will still be for full strength.
Let's consider an example where Black is to move, and the top 3 moves are a mate, winning material, or being slightly worse. We'll assume the turn perspective setting is off.
stockfish.get_top_moves(3)
# [
# {'Move': 'f5h3', 'Centipawn': None, 'Mate': -1}, # The move f5h3 leads to a mate in 1 for Black.
# {'Move': 'f5d7', 'Centipawn': -713, 'Mate': None}, # f5d7 leads to an evaluation of 7.13 in Black's favour.
# {'Move': 'f5h5', 'Centipawn': 31, 'Mate': None} # f5h5 leads to an evaluation of 0.31 in White's favour.
# ]Optional parameter verbose (default False) specifies whether to include the full info from the engine in the returned dictionary, including SelectiveDepth, Nodes, NodesPerSecond, Time, MultiPVLine, and WDL if available.
stockfish.get_top_moves(1, verbose=True)
# [{
# "Move": "d6e7",
# "Centipawn": -408,
# "Mate": None,
# "Nodes": 2767506,
# "NodesPerSecond": 526442,
# "Time": 5257,
# "SelectiveDepth": 31,
# "MultiPVLine": 1,
# "WDL": "0 0 1000"
# }]Optional parameter num_nodes specifies the number of nodes to search. If num_nodes is 0, then the engine will search until the configured depth is reached.
The perft command is used to test the move generation. It counts the total number of leaf nodes to a certain depth, and shows how this node count is divided amongst all legal moves of the current position.
The depth parameter should be an integer greater than zero and specifies the search depth.
stockfish.get_perft(3)
# (8902, {'a2a3': 380, 'b2b3': 420, 'c2c3': 420, 'd2d3': 539,
# 'e2e3': 599, 'f2f3': 380, 'g2g3': 420, 'h2h3': 380,
# 'a2a4': 420, 'b2b4': 421, 'c2c4': 441, 'd2d4': 560,
# 'e2e4': 600, 'f2f4': 401, 'g2g4': 421, 'h2h4': 420,
# 'b1a3': 400, 'b1c3': 440, 'g1f3': 440, 'g1h3': 400})Flip the side to move.
stockfish.get_fen_position()
# rnbqkbnr/pppppppp/8/8/8/8/PPPPPPPP/RNBQKBNR w KQkq - 0 1
stockfish.flip()
stockfish.get_fen_position()
# rnbqkbnr/pppppppp/8/8/8/8/PPPPPPPP/RNBQKBNR b KQkq - 0 1You can set the perspective of the evaluation to be from the perspective of the side to move, or from the perspective of White. Currently this setting only applies to get_top_moves().
# Set the perspective of the evaluation to be from the point of view of the side to move
stockfish.set_turn_perspective(True)
# Set the perspective of the evaluation to be from White's perspective
stockfish.set_turn_perspective(False)
# Get the current perspective of the evaluation
is_turn_perspective = stockfish.get_turn_perspective()Before calling this function, it is recommended that you first check if your version of Stockfish is recent enough to display WDL stats. To do this, use the "does_current_engine_version_have_wdl_option()" function below.
stockfish.get_wdl_stats()[87, 894, 19]
Optional arguments:
get_as_tuple: if you'd like to have a tuple returned instead of a list.time: if you'd like to constrain the search by a duration rather than the current depth.
stockfish.does_current_engine_version_have_wdl_option()True
stockfish.info()info depth 15 seldepth 18 multipv 1 score cp 39 nodes 64450 nps 555603 hashfull 21 tbhits 0 time 116 pv e2e4 e7e5 g1f3 g8f6 d2d4 f6e4 d4e5 d7d5 f1d3 b8c6 e1g1 c8e6 b1c3 e4c3 b2c3
stockfish.set_skill_level(15)stockfish.set_elo_rating(1350)stockfish.resume_full_strength()stockfish.set_depth(15)Creates and returns a dictionary representing the engine's current parameters.
stockfish.get_engine_parameters(){
"Debug Log File": "",
"Contempt": 0,
"Min Split Depth": 0,
"Threads": 1,
"Ponder": False,
"Hash": 16,
"MultiPV": 1,
"Skill Level": 20,
"Move Overhead": 10,
"Minimum Thinking Time": 20,
"Slow Mover": 100,
"UCI_Chess960": False,
"UCI_LimitStrength": False,
"UCI_Elo": 1350
}
stockfish.reset_engine_parameters()stockfish.get_fen_position()rnbqkbnr/pppppppp/8/8/8/8/PPPPPPPP/RNBQKBNR w KQkq - 0 1
stockfish.get_board_visual()+---+---+---+---+---+---+---+---+
| r | n | b | q | k | b | n | r | 8
+---+---+---+---+---+---+---+---+
| p | p | p | p | p | p | p | p | 7
+---+---+---+---+---+---+---+---+
| | | | | | | | | 6
+---+---+---+---+---+---+---+---+
| | | | | | | | | 5
+---+---+---+---+---+---+---+---+
| | | | | | | | | 4
+---+---+---+---+---+---+---+---+
| | | | | | | | | 3
+---+---+---+---+---+---+---+---+
| P | P | P | P | P | P | P | P | 2
+---+---+---+---+---+---+---+---+
| R | N | B | Q | K | B | N | R | 1
+---+---+---+---+---+---+---+---+
a b c d e f g h
This function has an optional boolean (True by default) as a parameter that indicates whether the board should be seen from the view of white. So it is possible to get the board from black's point of view like this:
stockfish.get_board_visual(False)+---+---+---+---+---+---+---+---+
| R | N | B | K | Q | B | N | R | 1
+---+---+---+---+---+---+---+---+
| P | P | P | P | P | P | P | P | 2
+---+---+---+---+---+---+---+---+
| | | | | | | | | 3
+---+---+---+---+---+---+---+---+
| | | | | | | | | 4
+---+---+---+---+---+---+---+---+
| | | | | | | | | 5
+---+---+---+---+---+---+---+---+
| | | | | | | | | 6
+---+---+---+---+---+---+---+---+
| p | p | p | p | p | p | p | p | 7
+---+---+---+---+---+---+---+---+
| r | n | b | k | q | b | n | r | 8
+---+---+---+---+---+---+---+---+
h g f e d c b a
Stockfish searches to the specified depth and evaluates the current position:
stockfish.get_evaluation()Instead of using the depth, you can also specify the time the engine should take to evaluate:
stockfish.get_evaluation(searchtime=2000) # searchtime in millisecondsA dictionary is returned representing the evaluation. Two example return values:
{"type":"cp", "value":12}
{"type":"mate", "value":-3}
If stockfish.get_turn_perspective() is True, then the eval value is relative to the side to move. Otherwise, positive is advantage white, negative is advantage black.
stockfish.get_static_eval()Sends the 'eval' command to Stockfish. This will get it to 'directly' evaluate the current position (i.e., no search is involved), and output a float value (not a whole number centipawn).
If one side is in check or mated, recent versions of Stockfish will output 'none' for the static eval. In this case, the function will return None.
Some example return values:
-5.27
0.28
None
params = BenchmarkParameters(**kwargs)parameters required to run the benchmark function. kwargs can be used to set custom values.
ttSize: range(1,128001)
threads: range(1,513)
limit: range(1,10001)
fenFile: "path/to/file.fen"
limitType: "depth", "perft", "nodes", "movetime"
evalType: "mixed", "classical", "NNUE"
stockfish.benchmark(params)This will run the bench command with BenchmarkParameters. It is an additional custom non-UCI command, mainly for debugging. Do not use this command during a search!
E.g., if the engine being used is Stockfish 14.1 or Stockfish 14, then the function would return 14. Meanwhile, if a development build of the engine is being used (not an official release), then the function returns an int with 5 or 6 digits, representing the date the engine was compiled on. For example, 20122 is returned for the development build compiled on January 2, 2022.
stockfish.get_stockfish_major_version()15
stockfish.is_development_build_of_engine()False
This command will clear Stockfish's hash table, which is relatively expensive and should generally only be done if the new position will be completely unrelated to the current one (such as a new game).
stockfish.send_ucinewgame_command()If the square is empty, the None object is returned. Otherwise, one of 12 enum members of a custom Stockfish.Piece enum will be returned. Each of the 12 members of this enum is named in the following pattern: colour followed by underscore followed by piece name, where the colour and piece name are in all caps. The value of each enum member is a char representing the piece (uppercase is white, lowercase is black). For white, it will be one of "P", "N", "B", "R", "Q", or "K". For black the same chars, except lowercase. For example, say the current position is the starting position:
stockfish.get_what_is_on_square("e1") # returns Stockfish.Piece.WHITE_KING
stockfish.get_what_is_on_square("e1").value # result is "K"
stockfish.get_what_is_on_square("d8") # returns Stockfish.Piece.BLACK_QUEEN
stockfish.get_what_is_on_square("d8").value # result is "q"
stockfish.get_what_is_on_square("h2") # returns Stockfish.Piece.WHITE_PAWN
stockfish.get_what_is_on_square("h2").value # result is "P"
stockfish.get_what_is_on_square("g8") # returns Stockfish.Piece.BLACK_KNIGHT
stockfish.get_what_is_on_square("g8").value # result is "n"
stockfish.get_what_is_on_square("b5") # returns NoneThe argument must be a string that represents the move, using the notation that Stockfish uses (i.e., the coordinate of the starting square followed by the coordinate of the ending square). The function will return one of the following enum members from a custom Stockfish.Capture enum: DIRECT_CAPTURE, EN_PASSANT, or NO_CAPTURE. For example, say the current position is the one after 1.e4 Nf6 2.Nc3 e6 3.e5 d5.
stockfish.will_move_be_a_capture("c3d5") # returns Stockfish.Capture.DIRECT_CAPTURE
stockfish.will_move_be_a_capture("e5f6") # returns Stockfish.Capture.DIRECT_CAPTURE
stockfish.will_move_be_a_capture("e5d6") # returns Stockfish.Capture.EN_PASSANT
stockfish.will_move_be_a_capture("f1e2") # returns Stockfish.Capture.NO_CAPTUREThe StockfishException is a newly defined Exception type. It is thrown when the underlying Stockfish process created by the wrapper crashes. This can happen when an incorrect input like an invalid FEN (for example 8/8/8/3k4/3K4/8/8/8 w - - 0 1 with both kings next to each other) is given to Stockfish.
Not all invalid inputs will lead to a StockfishException, but only those which cause the Stockfish process to crash.
To handle a StockfishException when using this library, import the StockfishException from the library and use a try/except-block:
from stockfish import StockfishException
try:
# Evaluation routine
except StockfishException:
# Error handlingYou can (de-)activate the debug view option with the set_debug_view function. Like this you can see all communication between the engine and the library.
stockfish.set_debug_view(True)Most contributions will involve making updates to stockfish/models.py. To test your changes, download a version of stockfish and paste the executable in the stockfish folder. Then, create a file in the stockfish folder called main.py. Both the executable and main.py will be ignored by git.
In main.py, start with something like the following:
from models import Stockfish
def main():
sf = Stockfish(path = "name of your stockfish executable")
# Use this object as you wish to test your changes.
if __name__ == "__main__":
main()Then when navigating to the stockfish folder in the terminal, you can run this main.py file simply with python main.py.
Once you're satisfied with your changes to models.py, see the section below for how to run the project's entire test suite.
For your stockfish executable (the same one mentioned in the previous section), paste it also in the project's root directory. Then in models.py,
temporarily modify this line so that the path's default value is changed from "stockfish" to the name of your stockfish executable:
path: str = "stockfish",Then in the project's root directory, you can run:
$ python setup.py testTo skip some of the slower tests, run:
$ python setup.py skip_slow_testsIf you discover any security related issues, please report it via the Private vulnerability reporting instead of using the issue tracker.
Note This is just a brief summary. For more information, please look here.
Due to the unfortunate death of Ilya Zhelyabuzhsky, the original repo is no longer maintained. For this reason, this fork was created, which continues the project and is currently maintained by johndoknjas and kieferro. The official PyPi releases for the Stockfish package will also be created from this repo in the future.
Please submit all bug reports and PRs to this repo instead of the old one.
- We want to sincerely thank Ilya Zhelyabuzhsky, the original founder of this project for writing and maintaining the code and for his contributions to the open source community.
- We also want to thank all the other contributors for working on this project.
MIT License. Please see License File for more information.