MangoCarrier¶

MangoCarrier integrates with the mango-agents framework for networked multi-agent deployments. The algorithm code is identical to the SimpleCarrier version — only the wiring changes.

Note

mango-agents is an optional dependency. Install it with:

pip install "distributed-resource-optimization[mango]"

Roles¶

Two mango Role classes are provided:

DistributedOptimizationRole: Wraps a DistributedAlgorithm (e.g. a COHDA participant or ADMM flex actor) as a mango role. Messages are routed to the algorithm’s on_exchange_message().
CoordinatorRole: Wraps a Coordinator (e.g. an ADMM coordinator). Listens for a StartCoordinatedDistributedOptimization message, runs the optimisation, then broadcasts OptimizationFinishedMessage to all peers.

Example — COHDA with MangoCarrier¶

import asyncio
import mango
from distributed_resource_optimization import (
    DistributedOptimizationRole,
    create_cohda_participant,
    create_cohda_start_message,
)

async def main():
    c = mango.create_tcp_container(addr=("127.0.0.1", 5555))

    a1 = c.register(mango.agent_composed_of(
        DistributedOptimizationRole(
            create_cohda_participant(1, [[0.0, 1, 2], [1, 2, 3]]))))
    a2 = c.register(mango.agent_composed_of(
        DistributedOptimizationRole(
            create_cohda_participant(2, [[0.0, 1, 2], [1, 2, 3]]))))

    async with mango.activate(c):
        start = create_cohda_start_message([1.2, 2.0, 3.0])
        await a1.send_message(start, a2.addr)
        await asyncio.sleep(0.5)

asyncio.run(main())

Example — ADMM Consensus with MangoCarrier¶

import asyncio
import mango
from distributed_resource_optimization import (
    DistributedOptimizationRole,
    CoordinatorRole,
    StartCoordinatedDistributedOptimization,
    OptimizationFinishedMessage,
    create_admm_flex_actor_one_to_many,
    create_consensus_target_reach_admm_coordinator,
    create_admm_start_consensus,
)

async def main():
    c = mango.create_tcp_container(addr=("127.0.0.1", 5555))

    flex1 = create_admm_flex_actor_one_to_many(10, [0.6, 0.4])
    flex2 = create_admm_flex_actor_one_to_many(10, [0.6, 0.4])
    coordinator = create_consensus_target_reach_admm_coordinator()

    a1 = c.register(mango.agent_composed_of(DistributedOptimizationRole(flex1)))
    a2 = c.register(mango.agent_composed_of(DistributedOptimizationRole(flex2)))
    ca = c.register(mango.agent_composed_of(CoordinatorRole(coordinator)))

    async with mango.activate(c):
        start = StartCoordinatedDistributedOptimization(
            create_admm_start_consensus([1.0, 2.0])
        )
        await ca.send_message(start, ca.addr)
        await asyncio.sleep(5.0)

    print("Actor 1:", flex1.x.round(3))
    print("Actor 2:", flex2.x.round(3))

asyncio.run(main())

Topology¶

The MangoCarrier uses mango’s topology system to discover neighbours. Set up a topology before activating the container so that each role knows which agents to communicate with. Refer to the mango topology documentation for details.

See Also¶

MangoCarrier, DistributedOptimizationRole, CoordinatorRole
StartCoordinatedDistributedOptimization, OptimizationFinishedMessage
SimpleCarrier — lightweight in-process carrier