thoth.adviser.predictors package

Submodules

thoth.adviser.predictors.annealing module

Implementation of Adaptive Simulated Annealing (ASA) used to resolve software stacks.

class thoth.adviser.predictors.annealing.AdaptiveSimulatedAnnealing(*, keep_history=None, temperature_history: List[Tuple[float, bool, float, int]] = NOTHING, temperature: float = 0.0)

Bases: thoth.adviser.predictor.Predictor

Implementation of adaptive simulated annealing looking for stacks based on the scoring function.

plot() → matplotlib.figure.Figure

Plot temperature history of adaptive simulated annealing.

pre_run() → None

Initialization before the actual annealing run.

run() → Tuple[thoth.adviser.state.State, Tuple[str, str, str]]

Run adaptive simulated annealing on top of beam.

thoth.adviser.predictors.hill_climbing module

Implementation of hill climbing in the state space.

class thoth.adviser.predictors.hill_climbing.HillClimbing(history: List[Tuple[float, int]] = NOTHING, *, keep_history=None)

Bases: thoth.adviser.predictor.Predictor

Implementation of hill climbing in the state space.

plot() → matplotlib.figure.Figure

Plot score of the highest rated stack during hill climbing.

pre_run() → None

Initialization before the actual hill climbing run.

run() → Tuple[thoth.adviser.state.State, Tuple[str, str, str]]

Get top state from the beam for the next resolution round.

thoth.adviser.predictors.latest module

Implementation of predictor used for resolving latest stacks in the state space.

class thoth.adviser.predictors.latest.ApproximatingLatest(history: List[Tuple[float, int]] = NOTHING, hop: bool = False, hop_logged: bool = False, packages_heated_up: Set[str] = NOTHING, initial_state: Optional[thoth.adviser.state.State] = None, latest_versions_heat_up: set = NOTHING, *, keep_history=None)

Bases: thoth.adviser.predictors.hill_climbing.HillClimbing

Implementation of predictor used for resolving latest stacks in the state space.

This predictor approximates resolution to the latest software stack. The resolution to the latest is approximated using continuous resolution with an optional randomness not to get stuck in a “trap” if resolution to all latest cannot be satisfied.

pre_run() → None

Initialize local variables before each predictor run per resolver.

run() → Tuple[thoth.adviser.state.State, Tuple[str, str, str]]

Get last state expanded and expand first unresolved dependency.

set_reward_signal(state: thoth.adviser.state.State, package_tuple: Tuple[str, str, str], reward: float) → None

Set hop to True if we did not get resolve any stack with latest.

thoth.adviser.predictors.mcts module

Implementation of Monte-Carlo Tree Search (MCTS) based predictor with adaptive simulated annealing schedule.

class thoth.adviser.predictors.mcts.MCTS(policy: Dict[str, List[Union[float, int]]] = NOTHING, next_state: Optional[thoth.adviser.state.State] = None, *, keep_history=None, temperature_history: List[Tuple[float, bool, float, int]] = NOTHING, temperature: float = 0.0)

Bases: thoth.adviser.predictors.td.TemporalDifference

Implementation of Monte-Carlo Tree Search (MCTS) based predictor with adaptive simulated annealing schedule.

pre_run() → None

Initialize pre-running of this predictor.

run() → Tuple[thoth.adviser.state.State, Tuple[str, str, str]]

Run MCTS with adaptive simulated annealing schedule.

set_reward_signal(state: thoth.adviser.state.State, _: Tuple[str, str, str], reward: float) → None

Note down reward signal of the last action performed.

thoth.adviser.predictors.random_walk module

Implementation of a Random Walk based dependency graph sampling predictor.

class thoth.adviser.predictors.random_walk.RandomWalk(history: List[Tuple[float, int]] = NOTHING, *, keep_history=None)

Bases: thoth.adviser.predictor.Predictor

Implementation of a Random Walk based dependency graph sampling predictor.

plot() → matplotlib.figure.Figure

Plot score of the highest rated stack during sampling.

pre_run() → None

Initialization before the random walk run.

run() → Tuple[thoth.adviser.state.State, Tuple[str, str, str]]

Generate stacks using random walking.

thoth.adviser.predictors.sampling module

Implementation of a random sampling of the state space.

class thoth.adviser.predictors.sampling.Sampling(history: List[Tuple[float, int]] = NOTHING, *, keep_history=None)

Bases: thoth.adviser.predictor.Predictor

Implementation of a random sampling of the state space.

plot() → matplotlib.figure.Figure

Plot score of the highest rated stack during sampling.

pre_run() → None

Initialization before the sampling run.

run() → Tuple[thoth.adviser.state.State, Tuple[str, str, str]]

Get random state and random unresolved dependency from the beam for the next resolution round.

thoth.adviser.predictors.td module

Implementation of Temporal Difference (TD) based predictor with adaptive simulated annealing schedule.

class thoth.adviser.predictors.td.TemporalDifference(policy: Dict[str, List[Union[float, int]]] = NOTHING, *, keep_history=None, temperature_history: List[Tuple[float, bool, float, int]] = NOTHING, temperature: float = 0.0)

Bases: thoth.adviser.predictors.annealing.AdaptiveSimulatedAnnealing

Implementation of Temporal Difference (TD) based predictor with adaptive simulated annealing schedule.

pre_run() → None

Initialize pre-running of this predictor.

run() → Tuple[thoth.adviser.state.State, Tuple[str, str, str]]

Run Temporal Difference (TD) with adaptive simulated annealing schedule.

set_reward_signal(state: thoth.adviser.state.State, _: Tuple[str, str, str], reward: float) → None

Note down reward signal of the last action performed.

Module contents

Implementation of predictors used with resolver..

class thoth.adviser.predictors.AdaptiveSimulatedAnnealing(*, keep_history=None, temperature_history: List[Tuple[float, bool, float, int]] = NOTHING, temperature: float = 0.0)

Bases: thoth.adviser.predictor.Predictor

Implementation of adaptive simulated annealing looking for stacks based on the scoring function.

plot() → matplotlib.figure.Figure

Plot temperature history of adaptive simulated annealing.

pre_run() → None

Initialization before the actual annealing run.

run() → Tuple[thoth.adviser.state.State, Tuple[str, str, str]]

Run adaptive simulated annealing on top of beam.

class thoth.adviser.predictors.ApproximatingLatest(history: List[Tuple[float, int]] = NOTHING, hop: bool = False, hop_logged: bool = False, packages_heated_up: Set[str] = NOTHING, initial_state: Optional[thoth.adviser.state.State] = None, latest_versions_heat_up: set = NOTHING, *, keep_history=None)

Bases: thoth.adviser.predictors.hill_climbing.HillClimbing

Implementation of predictor used for resolving latest stacks in the state space.

This predictor approximates resolution to the latest software stack. The resolution to the latest is approximated using continuous resolution with an optional randomness not to get stuck in a “trap” if resolution to all latest cannot be satisfied.

pre_run() → None

Initialize local variables before each predictor run per resolver.

run() → Tuple[thoth.adviser.state.State, Tuple[str, str, str]]

Get last state expanded and expand first unresolved dependency.

set_reward_signal(state: thoth.adviser.state.State, package_tuple: Tuple[str, str, str], reward: float) → None

Set hop to True if we did not get resolve any stack with latest.

class thoth.adviser.predictors.HillClimbing(history: List[Tuple[float, int]] = NOTHING, *, keep_history=None)

Bases: thoth.adviser.predictor.Predictor

Implementation of hill climbing in the state space.

plot() → matplotlib.figure.Figure

Plot score of the highest rated stack during hill climbing.

pre_run() → None

Initialization before the actual hill climbing run.

run() → Tuple[thoth.adviser.state.State, Tuple[str, str, str]]

Get top state from the beam for the next resolution round.

class thoth.adviser.predictors.RandomWalk(history: List[Tuple[float, int]] = NOTHING, *, keep_history=None)

Bases: thoth.adviser.predictor.Predictor

Implementation of a Random Walk based dependency graph sampling predictor.

plot() → matplotlib.figure.Figure

Plot score of the highest rated stack during sampling.

pre_run() → None

Initialization before the random walk run.

run() → Tuple[thoth.adviser.state.State, Tuple[str, str, str]]

Generate stacks using random walking.

class thoth.adviser.predictors.Sampling(history: List[Tuple[float, int]] = NOTHING, *, keep_history=None)

Bases: thoth.adviser.predictor.Predictor

Implementation of a random sampling of the state space.

plot() → matplotlib.figure.Figure

Plot score of the highest rated stack during sampling.

pre_run() → None

Initialization before the sampling run.

run() → Tuple[thoth.adviser.state.State, Tuple[str, str, str]]

Get random state and random unresolved dependency from the beam for the next resolution round.

class thoth.adviser.predictors.TemporalDifference(policy: Dict[str, List[Union[float, int]]] = NOTHING, *, keep_history=None, temperature_history: List[Tuple[float, bool, float, int]] = NOTHING, temperature: float = 0.0)

Bases: thoth.adviser.predictors.annealing.AdaptiveSimulatedAnnealing

Implementation of Temporal Difference (TD) based predictor with adaptive simulated annealing schedule.

pre_run() → None

Initialize pre-running of this predictor.

run() → Tuple[thoth.adviser.state.State, Tuple[str, str, str]]

Run Temporal Difference (TD) with adaptive simulated annealing schedule.

set_reward_signal(state: thoth.adviser.state.State, _: Tuple[str, str, str], reward: float) → None

Note down reward signal of the last action performed.

class thoth.adviser.predictors.MCTS(policy: Dict[str, List[Union[float, int]]] = NOTHING, next_state: Optional[thoth.adviser.state.State] = None, *, keep_history=None, temperature_history: List[Tuple[float, bool, float, int]] = NOTHING, temperature: float = 0.0)

Bases: thoth.adviser.predictors.td.TemporalDifference

Implementation of Monte-Carlo Tree Search (MCTS) based predictor with adaptive simulated annealing schedule.

pre_run() → None

Initialize pre-running of this predictor.

run() → Tuple[thoth.adviser.state.State, Tuple[str, str, str]]

Run MCTS with adaptive simulated annealing schedule.

set_reward_signal(state: thoth.adviser.state.State, _: Tuple[str, str, str], reward: float) → None

Note down reward signal of the last action performed.