Skip to content

refactored multi agent proposal #926

New issue

Have a question about this project? Sign up for a free GitHub account to open an issue and contact its maintainers and the community.

Sign up for GitHub

By clicking “Sign up for GitHub”, you agree to our terms of service and privacy statement. We’ll occasionally send you account related emails.

Already on GitHub? Sign in to your account

Open
wants to merge 2 commits into
base: main
Choose a base branch
from
Open

refactored multi agent proposal #926

Show file tree
Hide file tree
Changes from all commits
Commits
Show all changes
2 commits
Select commit Hold shift + click to select a range
798a5dd
use NamedTuples for plan!, Base.push! for sim. env
Mytolo Jul 11, 2023
ec34889
Merge branch 'main' into refactor-multiAgent-proposal
Mytolo Jul 19, 2023
File filter

Filter by extension

Filter by extension
Conversations
Failed to load comments.
Loading
Jump to
Jump to file
Failed to load files.
Loading
Diff view
Diff view
34 changes: 21 additions & 13 deletions src/ReinforcementLearningCore/src/policies/agent/base.jl
View file
Open in desktop
Original file line number Diff line number Diff line change
Expand Up @@ -51,10 +51,6 @@ function RLBase.optimise!(policy::AbstractPolicy, stage::AbstractStage, trajecto

@functor Agent (policy,)

function Base.push!(agent::Agent, ::PreActStage, env::AbstractEnv)
push!(agent, state(env))
end

# !!! TODO: In async scenarios, parameters of the policy may still be updating
# (partially), which will result to incorrect action. This should be addressed
# in Oolong.jl with a wrapper
Expand All @@ -64,11 +60,8 @@ function RLBase.plan!(agent::Agent{P,T,C}, env::AbstractEnv) where {P,T,C}
action
end

# Multiagent Version
function RLBase.plan!(agent::Agent{P,T,C}, env::E, p::Symbol) where {P,T,C,E<:AbstractEnv}
action = RLBase.plan!(agent.policy, env, p)
push!(agent.trajectory, agent.cache, action)
action
function Base.push!(agent::Agent, ::PreActStage, env::AbstractEnv)
push!(agent, state(env))
end

function Base.push!(agent::Agent{P,T,C}, ::PostActStage, env::E) where {P,T,C,E<:AbstractEnv}
Expand All @@ -79,11 +72,26 @@ function Base.push!(agent::Agent, ::PostExperimentStage, env::E) where {E<:Abstr
RLBase.reset!(agent.cache)
end

function Base.push!(agent::Agent, ::PostExperimentStage, env::E, player::Symbol) where {E<:AbstractEnv}
RLBase.reset!(agent.cache)
end

function Base.push!(agent::Agent{P,T,C}, state::S) where {P,T,C,S}
push!(agent.cache, state)
end

# Multiagent Version
function RLBase.plan!(agent::Agent{P,T,C}, env::E, p::Symbol) where {P,T,C,E<:AbstractEnv}
action = RLBase.plan!(agent.policy, env, p)
push!(agent.trajectory, agent.cache, action)
action
end

# for simultaneous DynamicStyle environments, we have to define push! operations
function Base.push!(agent::Agent, ::PreActStage, env::AbstractEnv, player::Symbol)
push!(agent, state(env, player))
end

function Base.push!(agent::Agent{P,T,C}, ::PostActStage, env::E, player::Symbol) where {P,T,C,E<:AbstractEnv}
push!(agent.cache, reward(env, player), is_terminated(env))
end

function Base.push!(agent::Agent, ::PostExperimentStage, env::E, player::Symbol) where {E<:AbstractEnv}
RLBase.reset!(agent.cache)
end
12 changes: 7 additions & 5 deletions src/ReinforcementLearningCore/src/policies/agent/multi_agent.jl
View file
Open in desktop
Original file line number Diff line number Diff line change
Expand Up @@ -133,10 +133,12 @@ function Base.run(

if check_stop(stop_condition, policy, env)
is_stop = true
@timeit_debug timer "push!(policy) PreActStage" push!(multiagent_policy, PreActStage(), env)
@timeit_debug timer "optimise! PreActStage" optimise!(multiagent_policy, PreActStage())
@timeit_debug timer "push!(hook) PreActStage" push!(multiagent_hook, PreActStage(), policy, env)
@timeit_debug timer "plan!" RLBase.plan!(multiagent_policy, env) # let the policy see the last observation
if !is_terminated(env)
@timeit_debug timer "push!(policy) PreActStage" push!(multiagent_policy, PreActStage(), env)
@timeit_debug timer "optimise! PreActStage" optimise!(multiagent_policy, PreActStage())
@timeit_debug timer "push!(hook) PreActStage" push!(multiagent_hook, PreActStage(), policy, env)
@timeit_debug timer "plan!" RLBase.plan!(multiagent_policy, env) # let the policy see the last observation
end
break
end

Expand Down Expand Up @@ -228,7 +230,7 @@ function Base.push!(composed_hook::ComposedHook{T},
end

function RLBase.plan!(multiagent::MultiAgentPolicy, env::E) where {E<:AbstractEnv}
return (RLBase.plan!(multiagent[player], env, player) for player in players(env))
return NamedTuple(player => RLBase.plan!(multiagent[player], env, player) for player in players(env))
end

function RLBase.optimise!(multiagent::MultiAgentPolicy, stage::S) where {S<:AbstractStage}
Expand Down