Module server.matchmaker.search

Functions

def get_average_rating(searches)
Expand source code
def get_average_rating(searches):
    return statistics.mean(itertools.chain(*[s.displayed_ratings for s in searches]))

Classes

class CombinedSearch (*searches: Search)
Expand source code
class CombinedSearch(Search):
    def __init__(self, *searches: Search):
        assert searches
        rating_type = searches[0].rating_type
        assert all(map(lambda s: s.rating_type == rating_type, searches))

        self.rating_type = rating_type
        self.searches = searches

    @property
    def players(self) -> list[Player]:
        return list(itertools.chain(*[s.players for s in self.searches]))

    @property
    def ratings(self) -> list[Rating]:
        return list(itertools.chain(*[s.ratings for s in self.searches]))

    @property
    def cumulative_rating(self) -> float:
        return sum(s.cumulative_rating for s in self.searches)

    @property
    def average_rating(self) -> float:
        return get_average_rating(self.searches)

    @property
    def raw_ratings(self) -> list[Rating]:
        return list(itertools.chain(*[s.raw_ratings for s in self.searches]))

    @property
    def displayed_ratings(self) -> list[float]:
        return list(itertools.chain(*[s.displayed_ratings for s in self.searches]))

    @property
    def failed_matching_attempts(self) -> int:
        return max(search.failed_matching_attempts for search in self.searches)

    def register_failed_matching_attempt(self):
        for search in self.searches:
            search.register_failed_matching_attempt()

    @property
    def match_threshold(self) -> float:
        """
        Defines the threshold for game quality
        """
        return min(s.match_threshold for s in self.searches)

    @property
    def is_matched(self) -> bool:
        return all(s.is_matched for s in self.searches)

    def done(self) -> bool:
        return all(s.done() for s in self.searches)

    @property
    def is_cancelled(self) -> bool:
        return any(s.is_cancelled for s in self.searches)

    def match(self, other: "Search"):
        """
        Mark as matched with given opponent
        """
        self._logger.info("Combined search matched %s with %s", self.players, other.players)

        for s in self.searches:
            s.match(other)

    async def await_match(self):
        """
        Wait for this search to complete
        """
        await asyncio.wait({s.await_match() for s in self.searches})

    def cancel(self):
        """
        Cancel searching for a match
        """
        for s in self.searches:
            s.cancel()

    def __str__(self):
        return f"CombinedSearch({', '.join(str(s) for s in self.searches)})"

    def __repr__(self):
        return f"CombinedSearch({', '.join(str(s) for s in self.searches)})"

    def get_original_searches(self) -> list[Search]:
        """
        Returns the searches of which this CombinedSearch is comprised
        """
        return list(self.searches)

Represents the state of a users search for a match.

Ancestors

Instance variables

prop average_rating : float
Expand source code
@property
def average_rating(self) -> float:
    return get_average_rating(self.searches)
prop cumulative_rating : float
Expand source code
@property
def cumulative_rating(self) -> float:
    return sum(s.cumulative_rating for s in self.searches)
prop failed_matching_attempts : int
Expand source code
@property
def failed_matching_attempts(self) -> int:
    return max(search.failed_matching_attempts for search in self.searches)
prop is_cancelled : bool
Expand source code
@property
def is_cancelled(self) -> bool:
    return any(s.is_cancelled for s in self.searches)
prop is_matched : bool
Expand source code
@property
def is_matched(self) -> bool:
    return all(s.is_matched for s in self.searches)
prop match_threshold : float
Expand source code
@property
def match_threshold(self) -> float:
    """
    Defines the threshold for game quality
    """
    return min(s.match_threshold for s in self.searches)

Defines the threshold for game quality

prop players : list[Player]
Expand source code
@property
def players(self) -> list[Player]:
    return list(itertools.chain(*[s.players for s in self.searches]))
prop ratings : list[Rating]
Expand source code
@property
def ratings(self) -> list[Rating]:
    return list(itertools.chain(*[s.ratings for s in self.searches]))
prop raw_ratings : list[Rating]
Expand source code
@property
def raw_ratings(self) -> list[Rating]:
    return list(itertools.chain(*[s.raw_ratings for s in self.searches]))

Methods

def done(self) ‑> bool
Expand source code
def done(self) -> bool:
    return all(s.done() for s in self.searches)
def get_original_searches(self) ‑> list[Search]
Expand source code
def get_original_searches(self) -> list[Search]:
    """
    Returns the searches of which this CombinedSearch is comprised
    """
    return list(self.searches)

Returns the searches of which this CombinedSearch is comprised

Inherited members

class Search (players: list[Player],
start_time: float | None = None,
rating_type: str = 'ladder_1v1',
on_matched: Callable[[ForwardRef('Search'), ForwardRef('Search')], Any] = <function Search.<lambda>>)
Expand source code
@with_logger
class Search:
    """
    Represents the state of a users search for a match.
    """

    def __init__(
        self,
        players: list[Player],
        start_time: Optional[float] = None,
        rating_type: str = RatingType.LADDER_1V1,
        on_matched: OnMatchedCallback = lambda _1, _2: None
    ):
        assert isinstance(players, list)
        for player in players:
            assert player.ratings[rating_type] is not None

        self.players = players
        self.rating_type = rating_type
        self.start_time = start_time or time.time()
        self._match = asyncio.get_event_loop().create_future()
        self._failed_matching_attempts = 0
        self.on_matched = on_matched

        # Precompute this
        self.quality_against_self = self.quality_with(self)

    def adjusted_rating(self, player: Player) -> Rating:
        """
        Returns an adjusted mean with a simple linear interpolation between current mean and a specified base mean
        """
        mean, dev = player.ratings[self.rating_type]
        game_count = player.game_count[self.rating_type]
        adjusted_mean = ((config.NEWBIE_MIN_GAMES - game_count) * config.NEWBIE_BASE_MEAN
                         + game_count * mean) / config.NEWBIE_MIN_GAMES
        return Rating(adjusted_mean, dev)

    def is_newbie(self, player: Player) -> bool:
        return player.game_count[self.rating_type] <= config.NEWBIE_MIN_GAMES

    def is_single_party(self) -> bool:
        return len(self.players) == 1

    def has_newbie(self) -> bool:
        for player in self.players:
            if self.is_newbie(player):
                return True

        return False

    def num_newbies(self) -> int:
        return sum(self.is_newbie(player) for player in self.players)

    def has_high_rated_player(self) -> bool:
        max_rating = max(self.displayed_ratings)
        return max_rating >= config.HIGH_RATED_PLAYER_MIN_RATING

    def has_top_player(self) -> bool:
        max_rating = max(self.displayed_ratings)
        return max_rating >= config.TOP_PLAYER_MIN_RATING

    @property
    def ratings(self) -> list[Rating]:
        ratings = []
        for player, rating in zip(self.players, self.raw_ratings):
            # New players (less than config.NEWBIE_MIN_GAMES games) match against less skilled opponents
            if self.is_newbie(player):
                rating = self.adjusted_rating(player)
            ratings.append(rating)
        return ratings

    @property
    def cumulative_rating(self) -> float:
        return sum(self.displayed_ratings)

    @property
    def average_rating(self) -> float:
        return statistics.mean(self.displayed_ratings)

    @property
    def raw_ratings(self) -> list[Rating]:
        return [player.ratings[self.rating_type] for player in self.players]

    @property
    def displayed_ratings(self) -> list[float]:
        """
        The client always displays mean - 3 * dev as a player's rating.
        So generally this is perceived as a player's true rating.
        """
        return [rating.displayed() for rating in self.raw_ratings]

    def _nearby_rating_range(self, delta: int) -> tuple[int, int]:
        """
        Returns 'boundary' mu values for player matching. Adjust delta for
        different game qualities.
        """
        mu, _ = self.ratings[0]  # Takes the rating of the first player, only works for 1v1
        rounded_mu = int(math.ceil(mu / 10) * 10)  # Round to 10
        return rounded_mu - delta, rounded_mu + delta

    @property
    def boundary_80(self) -> tuple[int, int]:
        """ Achieves roughly 80% quality. """
        return self._nearby_rating_range(200)

    @property
    def boundary_75(self) -> tuple[int, int]:
        """ Achieves roughly 75% quality. FIXME - why is it MORE restrictive??? """
        return self._nearby_rating_range(100)

    @property
    def failed_matching_attempts(self) -> int:
        return self._failed_matching_attempts

    @property
    def search_expansion(self) -> float:
        """
        Defines how much to expand the search range of game quality due to waiting
        time.

        The threshold will expand linearly with every failed matching attempt
        until it reaches the specified MAX.

        Top players use bigger values to make matching easier.
        """
        if self.has_top_player():
            return min(
                self._failed_matching_attempts * config.LADDER_TOP_PLAYER_SEARCH_EXPANSION_STEP,
                config.LADDER_TOP_PLAYER_SEARCH_EXPANSION_MAX
            )
        elif self.has_newbie():
            return min(
                self._failed_matching_attempts * config.LADDER_NEWBIE_SEARCH_EXPANSION_STEP,
                config.LADDER_NEWBIE_SEARCH_EXPANSION_MAX
            )
        else:
            return min(
                self._failed_matching_attempts * config.LADDER_SEARCH_EXPANSION_STEP,
                config.LADDER_SEARCH_EXPANSION_MAX
            )

    def register_failed_matching_attempt(self):
        """
        Signal that matchmaker tried to match this search but was unsuccessful
        and increase internal counter by one.
        """

        self._failed_matching_attempts += 1

    @property
    def match_threshold(self) -> float:
        """
        Defines the threshold for game quality

        The base minimum quality is determined as 80% of the quality of a game
        against a copy of yourself. This is decreased by `self.search_expansion`
        if search is to be expanded.
        """

        return max(0.8 * self.quality_against_self - self.search_expansion, 0)

    def quality_with(self, other: "Search") -> float:
        assert all(other.raw_ratings)
        assert other.players

        team1 = [trueskill.Rating(*rating) for rating in self.ratings]
        team2 = [trueskill.Rating(*rating) for rating in other.ratings]

        return trueskill.quality([team1, team2])

    @property
    def is_matched(self) -> bool:
        return self._match.done() and not self._match.cancelled()

    def done(self) -> bool:
        return self._match.done()

    @property
    def is_cancelled(self) -> bool:
        return self._match.cancelled()

    def matches_with(self, other: "Search") -> bool:
        """
        Determine if this search is compatible with other given search according
        to both wishes.
        """
        if not isinstance(other, Search):
            return False

        quality = self.quality_with(other)
        return self._match_quality_acceptable(other, quality)

    def _match_quality_acceptable(self, other: "Search", quality: float) -> bool:
        """
        Determine if the given match quality is acceptable.

        This gets it's own function so we can call it from the Matchmaker using
        a cached `quality` value.
        """
        # NOTE: We are assuming for optimization purposes that quality is
        # symmetric. If this ever changes, update here
        return (quality >= self.match_threshold and
                quality >= other.match_threshold)

    def match(self, other: "Search"):
        """
        Mark as matched with given opponent
        """
        self._logger.info("Matched %s with %s", self, other)

        self.on_matched(self, other)

        for player, raw_rating in zip(self.players, self.raw_ratings):
            if self.is_newbie(player):
                mean, dev = raw_rating
                adjusted_mean, adjusted_dev = self.adjusted_rating(player)
                self._logger.info(
                    "Adjusted mean rating for %s with %d games from %.1f to %.1f",
                    player.login,
                    player.game_count[self.rating_type],
                    mean,
                    adjusted_mean
                )
        self._match.set_result(other)

    async def await_match(self):
        """
        Wait for this search to complete
        """
        await asyncio.wait_for(self._match, None)
        return self._match

    def cancel(self):
        """
        Cancel searching for a match
        """
        self._match.cancel()

    def __str__(self) -> str:
        return (
            f"Search({self.rating_type}, {self._players_repr()}, threshold="
            f"{self.match_threshold:.2f}, expansion={self.search_expansion:.2f})"
        )

    def _players_repr(self) -> str:
        contents = ", ".join(
            f"Player({p.login}, {p.id}, {p.ratings[self.rating_type]})"
            for p in self.players
        )
        return f"[{contents}]"

    def __repr__(self) -> str:
        """For debugging"""
        return (
            f"Search({[p.login for p in self.players]}, {self.average_rating}"
            f"{f', FMA: {self.failed_matching_attempts}' if self.failed_matching_attempts else ''}"
            f"{', has_newbie)' if self.has_newbie() else ')'}"
        )

    def get_original_searches(self) -> list["Search"]:
        """
        Returns the searches of which this Search is comprised,
        as if it were a CombinedSearch of one
        """
        return [self]

Represents the state of a users search for a match.

Subclasses

Instance variables

prop average_rating : float
Expand source code
@property
def average_rating(self) -> float:
    return statistics.mean(self.displayed_ratings)
prop boundary_75 : tuple[int, int]
Expand source code
@property
def boundary_75(self) -> tuple[int, int]:
    """ Achieves roughly 75% quality. FIXME - why is it MORE restrictive??? """
    return self._nearby_rating_range(100)

Achieves roughly 75% quality. FIXME - why is it MORE restrictive???

prop boundary_80 : tuple[int, int]
Expand source code
@property
def boundary_80(self) -> tuple[int, int]:
    """ Achieves roughly 80% quality. """
    return self._nearby_rating_range(200)

Achieves roughly 80% quality.

prop cumulative_rating : float
Expand source code
@property
def cumulative_rating(self) -> float:
    return sum(self.displayed_ratings)
prop displayed_ratings : list[float]
Expand source code
@property
def displayed_ratings(self) -> list[float]:
    """
    The client always displays mean - 3 * dev as a player's rating.
    So generally this is perceived as a player's true rating.
    """
    return [rating.displayed() for rating in self.raw_ratings]

The client always displays mean - 3 * dev as a player's rating. So generally this is perceived as a player's true rating.

prop failed_matching_attempts : int
Expand source code
@property
def failed_matching_attempts(self) -> int:
    return self._failed_matching_attempts
prop is_cancelled : bool
Expand source code
@property
def is_cancelled(self) -> bool:
    return self._match.cancelled()
prop is_matched : bool
Expand source code
@property
def is_matched(self) -> bool:
    return self._match.done() and not self._match.cancelled()
prop match_threshold : float
Expand source code
@property
def match_threshold(self) -> float:
    """
    Defines the threshold for game quality

    The base minimum quality is determined as 80% of the quality of a game
    against a copy of yourself. This is decreased by `self.search_expansion`
    if search is to be expanded.
    """

    return max(0.8 * self.quality_against_self - self.search_expansion, 0)

Defines the threshold for game quality

The base minimum quality is determined as 80% of the quality of a game against a copy of yourself. This is decreased by self.search_expansion if search is to be expanded.

prop ratings : list[Rating]
Expand source code
@property
def ratings(self) -> list[Rating]:
    ratings = []
    for player, rating in zip(self.players, self.raw_ratings):
        # New players (less than config.NEWBIE_MIN_GAMES games) match against less skilled opponents
        if self.is_newbie(player):
            rating = self.adjusted_rating(player)
        ratings.append(rating)
    return ratings
prop raw_ratings : list[Rating]
Expand source code
@property
def raw_ratings(self) -> list[Rating]:
    return [player.ratings[self.rating_type] for player in self.players]
prop search_expansion : float
Expand source code
@property
def search_expansion(self) -> float:
    """
    Defines how much to expand the search range of game quality due to waiting
    time.

    The threshold will expand linearly with every failed matching attempt
    until it reaches the specified MAX.

    Top players use bigger values to make matching easier.
    """
    if self.has_top_player():
        return min(
            self._failed_matching_attempts * config.LADDER_TOP_PLAYER_SEARCH_EXPANSION_STEP,
            config.LADDER_TOP_PLAYER_SEARCH_EXPANSION_MAX
        )
    elif self.has_newbie():
        return min(
            self._failed_matching_attempts * config.LADDER_NEWBIE_SEARCH_EXPANSION_STEP,
            config.LADDER_NEWBIE_SEARCH_EXPANSION_MAX
        )
    else:
        return min(
            self._failed_matching_attempts * config.LADDER_SEARCH_EXPANSION_STEP,
            config.LADDER_SEARCH_EXPANSION_MAX
        )

Defines how much to expand the search range of game quality due to waiting time.

The threshold will expand linearly with every failed matching attempt until it reaches the specified MAX.

Top players use bigger values to make matching easier.

Methods

def adjusted_rating(self,
player: Player) ‑> Rating
Expand source code
def adjusted_rating(self, player: Player) -> Rating:
    """
    Returns an adjusted mean with a simple linear interpolation between current mean and a specified base mean
    """
    mean, dev = player.ratings[self.rating_type]
    game_count = player.game_count[self.rating_type]
    adjusted_mean = ((config.NEWBIE_MIN_GAMES - game_count) * config.NEWBIE_BASE_MEAN
                     + game_count * mean) / config.NEWBIE_MIN_GAMES
    return Rating(adjusted_mean, dev)

Returns an adjusted mean with a simple linear interpolation between current mean and a specified base mean

async def await_match(self)
Expand source code
async def await_match(self):
    """
    Wait for this search to complete
    """
    await asyncio.wait_for(self._match, None)
    return self._match

Wait for this search to complete

def cancel(self)
Expand source code
def cancel(self):
    """
    Cancel searching for a match
    """
    self._match.cancel()

Cancel searching for a match

def done(self) ‑> bool
Expand source code
def done(self) -> bool:
    return self._match.done()
def get_original_searches(self) ‑> list['Search']
Expand source code
def get_original_searches(self) -> list["Search"]:
    """
    Returns the searches of which this Search is comprised,
    as if it were a CombinedSearch of one
    """
    return [self]

Returns the searches of which this Search is comprised, as if it were a CombinedSearch of one

def has_high_rated_player(self) ‑> bool
Expand source code
def has_high_rated_player(self) -> bool:
    max_rating = max(self.displayed_ratings)
    return max_rating >= config.HIGH_RATED_PLAYER_MIN_RATING
def has_newbie(self) ‑> bool
Expand source code
def has_newbie(self) -> bool:
    for player in self.players:
        if self.is_newbie(player):
            return True

    return False
def has_top_player(self) ‑> bool
Expand source code
def has_top_player(self) -> bool:
    max_rating = max(self.displayed_ratings)
    return max_rating >= config.TOP_PLAYER_MIN_RATING
def is_newbie(self,
player: Player) ‑> bool
Expand source code
def is_newbie(self, player: Player) -> bool:
    return player.game_count[self.rating_type] <= config.NEWBIE_MIN_GAMES
def is_single_party(self) ‑> bool
Expand source code
def is_single_party(self) -> bool:
    return len(self.players) == 1
def match(self,
other: Search)
Expand source code
def match(self, other: "Search"):
    """
    Mark as matched with given opponent
    """
    self._logger.info("Matched %s with %s", self, other)

    self.on_matched(self, other)

    for player, raw_rating in zip(self.players, self.raw_ratings):
        if self.is_newbie(player):
            mean, dev = raw_rating
            adjusted_mean, adjusted_dev = self.adjusted_rating(player)
            self._logger.info(
                "Adjusted mean rating for %s with %d games from %.1f to %.1f",
                player.login,
                player.game_count[self.rating_type],
                mean,
                adjusted_mean
            )
    self._match.set_result(other)

Mark as matched with given opponent

def matches_with(self,
other: Search) ‑> bool
Expand source code
def matches_with(self, other: "Search") -> bool:
    """
    Determine if this search is compatible with other given search according
    to both wishes.
    """
    if not isinstance(other, Search):
        return False

    quality = self.quality_with(other)
    return self._match_quality_acceptable(other, quality)

Determine if this search is compatible with other given search according to both wishes.

def num_newbies(self) ‑> int
Expand source code
def num_newbies(self) -> int:
    return sum(self.is_newbie(player) for player in self.players)
def quality_with(self,
other: Search) ‑> float
Expand source code
def quality_with(self, other: "Search") -> float:
    assert all(other.raw_ratings)
    assert other.players

    team1 = [trueskill.Rating(*rating) for rating in self.ratings]
    team2 = [trueskill.Rating(*rating) for rating in other.ratings]

    return trueskill.quality([team1, team2])
def register_failed_matching_attempt(self)
Expand source code
def register_failed_matching_attempt(self):
    """
    Signal that matchmaker tried to match this search but was unsuccessful
    and increase internal counter by one.
    """

    self._failed_matching_attempts += 1

Signal that matchmaker tried to match this search but was unsuccessful and increase internal counter by one.