decent_bench.agent

class decent_bench.agent.Agent(agent_id: int, cost_function: CostFunction, activation_scheme: AgentActivationScheme)

Bases: object

Agent with unique id, local cost function, and activation scheme.

property id: int

Unique id for the agent.

property cost_function: CostFunction

Local cost function.

property x: ndarray[tuple[Any, ...], dtype[float64]]

Local optimization variable x.

Raises:

RuntimeError – if x is retrieved before being set or initialized

property received_messages: Mapping[Agent, ndarray[tuple[Any, ...], dtype[float64]]]

Messages received by neighbors.

property aux_vars: dict[str, ndarray[tuple[Any, ...], dtype[float64]]]

Auxiliary optimization variables used by algorithms that require more variables than x.

initialize(*, x: ndarray[tuple[Any, ...], dtype[float64]] | None = None, aux_vars: dict[str, ndarray[tuple[Any, ...], dtype[float64]]] | None = None, received_msgs: dict[Agent, ndarray[tuple[Any, ...], dtype[float64]]] | None = None) → None

Initialize local variables and messages before running an algorithm.

Parameters:

• x – initial x

• aux_vars – initial auxiliary variables

• received_msgs – initial messages from neighbors

class decent_bench.agent.AgentMetricsView(cost_function: CostFunction, x_per_iteration: list[ndarray[tuple[Any, ...], dtype[float64]]], n_evaluate_calls: int, n_gradient_calls: int, n_hessian_calls: int, n_proximal_calls: int, n_sent_messages: int, n_received_messages: int, n_sent_messages_dropped: int)

Bases: object

Immutable view of agent that exposes useful properties for calculating metrics.

cost_function: CostFunction

x_per_iteration: list[ndarray[tuple[Any, ...], dtype[float64]]]

n_evaluate_calls: int

n_gradient_calls: int

n_hessian_calls: int

n_proximal_calls: int

n_sent_messages: int

n_received_messages: int

n_sent_messages_dropped: int

static from_agent(agent: Agent) → AgentMetricsView

Create from agent.