2021-12-12 23:19:18 +00:00
|
|
|
module RL
|
2021-12-12 14:29:08 +00:00
|
|
|
|
2022-01-11 00:31:30 +00:00
|
|
|
export run_rl, LocalCOMEnv
|
2021-12-14 03:03:14 +00:00
|
|
|
|
2022-01-06 00:48:37 +00:00
|
|
|
using Base: OneTo
|
|
|
|
|
2021-12-10 02:16:45 +00:00
|
|
|
using ReinforcementLearning
|
2021-12-12 17:27:56 +00:00
|
|
|
using Flux: InvDecay
|
2021-12-12 14:29:08 +00:00
|
|
|
using Intervals
|
2021-12-12 17:27:56 +00:00
|
|
|
using StaticArrays: SVector
|
2021-12-14 03:03:14 +00:00
|
|
|
using LoopVectorization: @turbo
|
2021-12-12 23:19:18 +00:00
|
|
|
using Random: Random
|
2021-12-13 01:24:34 +00:00
|
|
|
using ProgressMeter: @showprogress
|
2021-12-12 23:19:18 +00:00
|
|
|
|
2022-01-11 17:39:38 +00:00
|
|
|
using ..ReCo:
|
|
|
|
ReCo, Particle, angle2, norm2d, sq_norm2d, Shape, DEFAULT_SKIN_TO_INTERACTION_R_RATIO
|
2021-12-12 14:29:08 +00:00
|
|
|
|
2022-01-06 00:48:37 +00:00
|
|
|
const INITIAL_STATE_IND = 1
|
2022-01-11 00:31:30 +00:00
|
|
|
const INITIAL_REWARD = 0.0
|
|
|
|
|
|
|
|
method_not_implemented() = error("Method not implemented!")
|
2022-01-06 00:48:37 +00:00
|
|
|
|
2022-01-11 00:31:30 +00:00
|
|
|
function gen_angle_state_space(n_angle_states::Int64)
|
2022-01-06 00:48:37 +00:00
|
|
|
angle_range = range(; start=-π, stop=π, length=n_angle_states + 1)
|
|
|
|
|
|
|
|
angle_state_space = Vector{Interval}(undef, n_angle_states)
|
|
|
|
|
|
|
|
@simd for i in 1:n_angle_states
|
|
|
|
if i == 1
|
|
|
|
bound = Closed
|
|
|
|
else
|
|
|
|
bound = Open
|
|
|
|
end
|
|
|
|
|
|
|
|
angle_state_space[i] = Interval{Float64,bound,Closed}(
|
|
|
|
angle_range[i], angle_range[i + 1]
|
|
|
|
)
|
|
|
|
end
|
|
|
|
|
|
|
|
return angle_state_space
|
|
|
|
end
|
2021-12-14 03:03:14 +00:00
|
|
|
|
2022-01-11 00:31:30 +00:00
|
|
|
function gen_distance_state_space(
|
|
|
|
min_distance::Float64, max_distance::Float64, n_distance_states::Int64
|
|
|
|
)
|
|
|
|
@assert min_distance >= 0.0
|
|
|
|
@assert max_distance > min_distance
|
|
|
|
@assert n_distance_states > 1
|
|
|
|
|
|
|
|
distance_range = range(;
|
|
|
|
start=min_distance, stop=max_distance, length=n_distance_states + 1
|
|
|
|
)
|
|
|
|
|
|
|
|
distance_state_space = Vector{Interval}(undef, n_distance_states)
|
|
|
|
|
|
|
|
@simd for i in 1:n_distance_states
|
|
|
|
if i == 1
|
|
|
|
bound = Closed
|
|
|
|
else
|
|
|
|
bound = Open
|
|
|
|
end
|
|
|
|
|
|
|
|
distance_state_space[i] = Interval{Float64,bound,Closed}(
|
|
|
|
distance_range[i], distance_range[i + 1]
|
|
|
|
)
|
|
|
|
end
|
|
|
|
|
|
|
|
return distance_state_space
|
|
|
|
end
|
|
|
|
|
|
|
|
abstract type Env <: AbstractEnv end
|
|
|
|
|
2022-01-11 17:39:38 +00:00
|
|
|
mutable struct EnvSharedProps{state_dims}
|
2021-12-20 23:31:44 +00:00
|
|
|
n_actions::Int64
|
|
|
|
action_space::Vector{SVector{2,Float64}}
|
2022-01-06 00:48:37 +00:00
|
|
|
action_ind_space::OneTo{Int64}
|
2021-12-14 03:03:14 +00:00
|
|
|
|
|
|
|
n_states::Int64
|
2022-01-11 00:31:30 +00:00
|
|
|
state_space::Vector{SVector{state_dims,Interval}}
|
2022-01-06 00:48:37 +00:00
|
|
|
state_ind_space::OneTo{Int64}
|
2021-12-20 23:31:44 +00:00
|
|
|
state_ind::Int64
|
2021-12-14 03:03:14 +00:00
|
|
|
|
2021-12-10 02:16:45 +00:00
|
|
|
reward::Float64
|
2021-12-20 23:31:44 +00:00
|
|
|
terminated::Bool
|
2021-12-10 02:16:45 +00:00
|
|
|
|
2022-01-11 17:39:38 +00:00
|
|
|
function EnvSharedProps(
|
2022-01-11 00:31:30 +00:00
|
|
|
n_states::Int64,
|
|
|
|
state_space::Vector{SVector{state_dims,Interval}};
|
2022-01-06 00:48:37 +00:00
|
|
|
n_v_actions::Int64=2,
|
2021-12-28 22:39:24 +00:00
|
|
|
n_ω_actions::Int64=3,
|
|
|
|
max_v::Float64=40.0,
|
2021-12-16 12:50:38 +00:00
|
|
|
max_ω::Float64=π / 2,
|
2022-01-11 00:31:30 +00:00
|
|
|
) where {state_dims}
|
2021-12-10 02:16:45 +00:00
|
|
|
@assert n_v_actions > 1
|
2021-12-12 23:19:18 +00:00
|
|
|
@assert n_ω_actions > 1
|
2021-12-10 02:16:45 +00:00
|
|
|
@assert max_v > 0
|
|
|
|
@assert max_ω > 0
|
|
|
|
|
2022-01-06 00:48:37 +00:00
|
|
|
v_action_space = range(; start=0.0, stop=max_v, length=n_v_actions)
|
|
|
|
ω_action_space = range(; start=-max_ω, stop=max_ω, length=n_ω_actions)
|
2021-12-10 02:16:45 +00:00
|
|
|
|
|
|
|
n_actions = n_v_actions * n_ω_actions
|
|
|
|
|
2021-12-20 23:31:44 +00:00
|
|
|
action_space = Vector{SVector{2,Float64}}(undef, n_actions)
|
2021-12-10 02:16:45 +00:00
|
|
|
|
|
|
|
ind = 1
|
|
|
|
for v in v_action_space
|
|
|
|
for ω in ω_action_space
|
2021-12-20 23:31:44 +00:00
|
|
|
action_space[ind] = SVector(v, ω)
|
2021-12-10 02:16:45 +00:00
|
|
|
ind += 1
|
|
|
|
end
|
|
|
|
end
|
|
|
|
|
2022-01-06 00:48:37 +00:00
|
|
|
action_ind_space = OneTo(n_actions)
|
2021-12-13 01:24:34 +00:00
|
|
|
|
2022-01-06 00:48:37 +00:00
|
|
|
state_ind_space = OneTo(n_states)
|
2021-12-20 23:31:44 +00:00
|
|
|
|
2022-01-11 00:31:30 +00:00
|
|
|
return new{state_dims}(
|
2021-12-20 23:31:44 +00:00
|
|
|
n_actions,
|
2021-12-13 01:24:34 +00:00
|
|
|
action_space,
|
2021-12-14 03:03:14 +00:00
|
|
|
action_ind_space,
|
2021-12-20 23:31:44 +00:00
|
|
|
n_states,
|
2021-12-13 01:24:34 +00:00
|
|
|
state_space,
|
2021-12-14 03:03:14 +00:00
|
|
|
state_ind_space,
|
2022-01-06 00:48:37 +00:00
|
|
|
INITIAL_STATE_IND,
|
2021-12-14 03:03:14 +00:00
|
|
|
INITIAL_REWARD,
|
2021-12-20 23:31:44 +00:00
|
|
|
false,
|
2021-12-12 14:29:08 +00:00
|
|
|
)
|
2021-12-10 02:16:45 +00:00
|
|
|
end
|
|
|
|
end
|
|
|
|
|
2021-12-20 23:31:44 +00:00
|
|
|
function reset!(env::Env)
|
2022-01-11 17:39:38 +00:00
|
|
|
env.shared.terminated = false
|
2021-12-14 03:03:14 +00:00
|
|
|
|
|
|
|
return nothing
|
|
|
|
end
|
|
|
|
|
2022-01-11 17:39:38 +00:00
|
|
|
RLBase.state_space(env::Env) = env.shared.state_ind_space
|
2021-12-10 02:16:45 +00:00
|
|
|
|
2022-01-11 17:39:38 +00:00
|
|
|
RLBase.state(env::Env) = env.shared.state_ind
|
2021-12-10 02:16:45 +00:00
|
|
|
|
2022-01-11 17:39:38 +00:00
|
|
|
RLBase.action_space(env::Env) = env.shared.action_ind_space
|
2021-12-14 03:03:14 +00:00
|
|
|
|
2022-01-11 17:39:38 +00:00
|
|
|
RLBase.reward(env::Env) = env.shared.reward
|
2021-12-14 03:03:14 +00:00
|
|
|
|
2022-01-11 17:39:38 +00:00
|
|
|
RLBase.is_terminated(env::Env) = env.shared.terminated
|
2021-12-14 03:03:14 +00:00
|
|
|
|
2022-01-11 17:39:38 +00:00
|
|
|
struct EnvHelperSharedProps{H<:AbstractHook}
|
2021-12-20 23:31:44 +00:00
|
|
|
env::Env
|
|
|
|
agent::Agent
|
|
|
|
hook::H
|
2021-12-14 03:03:14 +00:00
|
|
|
|
2021-12-20 23:31:44 +00:00
|
|
|
n_steps_before_actions_update::Int64
|
2021-12-14 03:03:14 +00:00
|
|
|
|
2022-01-06 23:30:05 +00:00
|
|
|
goal_gyration_tensor_eigvals_ratio::Float64
|
2021-12-12 17:27:56 +00:00
|
|
|
|
2021-12-20 23:31:44 +00:00
|
|
|
n_particles::Int64
|
2022-01-08 21:44:20 +00:00
|
|
|
|
2022-01-11 00:31:30 +00:00
|
|
|
old_states_ind::Vector{Int64}
|
|
|
|
states_ind::Vector{Int64}
|
|
|
|
|
|
|
|
actions::Vector{SVector{2,Float64}}
|
|
|
|
actions_ind::Vector{Int64}
|
2021-12-28 16:15:00 +00:00
|
|
|
|
2022-01-11 17:39:38 +00:00
|
|
|
function EnvHelperSharedProps(
|
2021-12-20 23:31:44 +00:00
|
|
|
env::Env,
|
|
|
|
agent::Agent,
|
|
|
|
hook::H,
|
2021-12-13 01:24:34 +00:00
|
|
|
n_steps_before_actions_update::Int64,
|
2022-01-06 23:30:05 +00:00
|
|
|
goal_gyration_tensor_eigvals_ratio::Float64,
|
2021-12-20 23:31:44 +00:00
|
|
|
n_particles::Int64,
|
2021-12-12 17:27:56 +00:00
|
|
|
) where {H<:AbstractHook}
|
2021-12-20 23:31:44 +00:00
|
|
|
return new{H}(
|
|
|
|
env,
|
|
|
|
agent,
|
|
|
|
hook,
|
2021-12-12 14:29:08 +00:00
|
|
|
n_steps_before_actions_update,
|
2022-01-06 23:30:05 +00:00
|
|
|
goal_gyration_tensor_eigvals_ratio,
|
2021-12-20 23:31:44 +00:00
|
|
|
n_particles,
|
2022-01-11 00:31:30 +00:00
|
|
|
fill(0, n_particles),
|
|
|
|
fill(0, n_particles),
|
2022-01-08 21:44:20 +00:00
|
|
|
fill(SVector(0.0, 0.0), n_particles),
|
|
|
|
fill(0, n_particles),
|
2021-12-12 14:29:08 +00:00
|
|
|
)
|
|
|
|
end
|
|
|
|
end
|
|
|
|
|
2022-01-11 00:31:30 +00:00
|
|
|
abstract type EnvHelper end
|
2022-01-08 21:44:20 +00:00
|
|
|
|
2022-01-11 17:39:38 +00:00
|
|
|
function gen_env_helper(::Env, env_helper_params::EnvHelperSharedProps)
|
2022-01-11 00:31:30 +00:00
|
|
|
return method_not_implemented()
|
|
|
|
end
|
|
|
|
|
|
|
|
function pre_integration_hook(::EnvHelper)
|
|
|
|
return method_not_implemented()
|
2021-12-12 17:27:56 +00:00
|
|
|
end
|
|
|
|
|
2021-12-20 23:31:44 +00:00
|
|
|
function state_update_helper_hook(
|
2022-01-11 00:31:30 +00:00
|
|
|
::EnvHelper, id1::Int64, id2::Int64, r⃗₁₂::SVector{2,Float64}
|
2021-12-12 17:27:56 +00:00
|
|
|
)
|
2022-01-11 00:31:30 +00:00
|
|
|
return method_not_implemented()
|
2021-12-12 17:27:56 +00:00
|
|
|
end
|
|
|
|
|
2022-01-11 00:31:30 +00:00
|
|
|
function find_state_ind(state::S, state_space::Vector{S}) where {S<:SVector}
|
2021-12-20 23:31:44 +00:00
|
|
|
return findfirst(x -> x == state, state_space)
|
2021-12-12 17:27:56 +00:00
|
|
|
end
|
|
|
|
|
2022-01-06 00:48:37 +00:00
|
|
|
function find_state_interval(value::Float64, state_space::Vector{Interval})::Interval
|
|
|
|
for state in state_space
|
|
|
|
if value in state
|
|
|
|
return state
|
|
|
|
end
|
|
|
|
end
|
|
|
|
end
|
|
|
|
|
2022-01-11 00:31:30 +00:00
|
|
|
function state_update_hook(::EnvHelper, particles::Vector{Particle})
|
|
|
|
return method_not_implemented()
|
2021-12-20 23:31:44 +00:00
|
|
|
end
|
|
|
|
|
2022-01-11 00:31:30 +00:00
|
|
|
function get_env_agent_hook(env_helper::EnvHelper)
|
2022-01-11 17:39:38 +00:00
|
|
|
return (env_helper.shared.env, env_helper.shared.agent, env_helper.shared.hook)
|
2021-12-20 23:31:44 +00:00
|
|
|
end
|
|
|
|
|
2022-01-11 00:31:30 +00:00
|
|
|
function update_reward!(::Env, ::EnvHelper, particle::Particle)
|
|
|
|
return method_not_implemented()
|
2022-01-06 00:48:37 +00:00
|
|
|
end
|
|
|
|
|
2021-12-20 23:31:44 +00:00
|
|
|
function update_table_and_actions_hook(
|
2022-01-11 00:31:30 +00:00
|
|
|
env_helper::EnvHelper, particle::Particle, first_integration_step::Bool
|
2021-12-20 23:31:44 +00:00
|
|
|
)
|
2022-01-11 00:31:30 +00:00
|
|
|
env, agent, hook = get_env_agent_hook(env_helper)
|
2021-12-20 23:31:44 +00:00
|
|
|
|
|
|
|
id = particle.id
|
|
|
|
|
|
|
|
if !first_integration_step
|
|
|
|
# Old state
|
2022-01-11 17:39:38 +00:00
|
|
|
env.shared.state_ind = env_helper.shared.old_states_ind[id]
|
2021-12-20 23:31:44 +00:00
|
|
|
|
2022-01-11 17:39:38 +00:00
|
|
|
action_ind = env_helper.shared.actions_ind[id]
|
2021-12-20 23:31:44 +00:00
|
|
|
|
|
|
|
# Pre act
|
|
|
|
agent(PRE_ACT_STAGE, env, action_ind)
|
|
|
|
hook(PRE_ACT_STAGE, agent, env, action_ind)
|
|
|
|
|
|
|
|
# Update to current state
|
2022-01-11 17:39:38 +00:00
|
|
|
env.shared.state_ind = env_helper.shared.states_ind[id]
|
2021-12-20 23:31:44 +00:00
|
|
|
|
|
|
|
# Update reward
|
2022-01-11 00:31:30 +00:00
|
|
|
update_reward!(env, env_helper, particle)
|
2021-12-20 23:31:44 +00:00
|
|
|
|
2021-12-14 03:03:14 +00:00
|
|
|
# Post act
|
2021-12-12 17:27:56 +00:00
|
|
|
agent(POST_ACT_STAGE, env)
|
2021-12-14 03:03:14 +00:00
|
|
|
hook(POST_ACT_STAGE, agent, env)
|
2021-12-12 17:27:56 +00:00
|
|
|
end
|
|
|
|
|
2021-12-20 23:31:44 +00:00
|
|
|
# Update action
|
|
|
|
action_ind = agent(env)
|
2022-01-11 17:39:38 +00:00
|
|
|
action = env.shared.action_space[action_ind]
|
2021-12-20 23:31:44 +00:00
|
|
|
|
2022-01-11 17:39:38 +00:00
|
|
|
env_helper.shared.actions[id] = action
|
|
|
|
env_helper.shared.actions_ind[id] = action_ind
|
2021-12-28 16:15:00 +00:00
|
|
|
|
2021-12-20 23:31:44 +00:00
|
|
|
return nothing
|
|
|
|
end
|
|
|
|
|
|
|
|
act_hook(::Nothing, args...) = nothing
|
|
|
|
|
|
|
|
function act_hook(
|
2022-01-11 00:31:30 +00:00
|
|
|
env_helper::EnvHelper, particle::Particle, δt::Float64, si::Float64, co::Float64
|
2021-12-20 23:31:44 +00:00
|
|
|
)
|
|
|
|
# Apply action
|
2022-01-11 17:39:38 +00:00
|
|
|
action = env_helper.shared.actions[particle.id]
|
2021-12-20 23:31:44 +00:00
|
|
|
|
|
|
|
vδt = action[1] * δt
|
|
|
|
particle.tmp_c += SVector(vδt * co, vδt * si)
|
|
|
|
particle.φ += action[2] * δt
|
|
|
|
|
2021-12-12 17:27:56 +00:00
|
|
|
return nothing
|
|
|
|
end
|
2021-12-12 14:29:08 +00:00
|
|
|
|
2022-01-06 00:48:37 +00:00
|
|
|
function gen_agent(n_states::Int64, n_actions::Int64, ϵ_stable::Float64)
|
|
|
|
# TODO: Optimize warmup and decay
|
|
|
|
warmup_steps = 200_000
|
|
|
|
decay_steps = 1_000_000
|
|
|
|
|
2021-12-20 23:31:44 +00:00
|
|
|
policy = QBasedPolicy(;
|
|
|
|
learner=MonteCarloLearner(;
|
|
|
|
approximator=TabularQApproximator(;
|
|
|
|
n_state=n_states, n_action=n_actions, opt=InvDecay(1.0)
|
|
|
|
),
|
|
|
|
),
|
2022-01-06 00:48:37 +00:00
|
|
|
explorer=EpsilonGreedyExplorer(;
|
|
|
|
kind=:linear,
|
|
|
|
ϵ_init=1.0,
|
|
|
|
ϵ_stable=ϵ_stable,
|
|
|
|
warmup_steps=warmup_steps,
|
|
|
|
decay_steps=decay_steps,
|
|
|
|
),
|
2021-12-20 23:31:44 +00:00
|
|
|
)
|
|
|
|
|
|
|
|
return Agent(; policy=policy, trajectory=VectorSARTTrajectory())
|
|
|
|
end
|
|
|
|
|
2021-12-15 03:45:15 +00:00
|
|
|
function run_rl(;
|
2022-01-11 00:31:30 +00:00
|
|
|
EnvType::Type{E},
|
|
|
|
parent_dir_appendix::String,
|
2022-01-06 23:30:05 +00:00
|
|
|
goal_gyration_tensor_eigvals_ratio::Float64,
|
2021-12-28 16:15:00 +00:00
|
|
|
n_episodes::Int64=200,
|
2021-12-14 03:03:14 +00:00
|
|
|
episode_duration::Float64=50.0,
|
2021-12-28 16:15:00 +00:00
|
|
|
update_actions_at::Float64=0.1,
|
2021-12-13 01:24:34 +00:00
|
|
|
n_particles::Int64=100,
|
2021-12-14 03:03:14 +00:00
|
|
|
seed::Int64=42,
|
2022-01-06 00:48:37 +00:00
|
|
|
ϵ_stable::Float64=0.0001,
|
2022-01-11 00:31:30 +00:00
|
|
|
skin_to_interaction_r_ratio::Float64=DEFAULT_SKIN_TO_INTERACTION_R_RATIO,
|
|
|
|
packing_ratio=0.22,
|
|
|
|
) where {E<:Env}
|
2022-01-06 23:30:05 +00:00
|
|
|
@assert 0.0 <= goal_gyration_tensor_eigvals_ratio <= 1.0
|
2021-12-12 14:29:08 +00:00
|
|
|
@assert n_episodes > 0
|
|
|
|
@assert episode_duration > 0
|
2021-12-28 16:15:00 +00:00
|
|
|
@assert update_actions_at in 0.001:0.001:episode_duration
|
2021-12-12 14:29:08 +00:00
|
|
|
@assert n_particles > 0
|
2022-01-06 00:48:37 +00:00
|
|
|
@assert 0.0 < ϵ_stable < 1.0
|
2021-12-12 14:29:08 +00:00
|
|
|
|
2021-12-14 03:03:14 +00:00
|
|
|
# Setup
|
|
|
|
Random.seed!(seed)
|
2021-12-10 02:16:45 +00:00
|
|
|
|
2021-12-28 16:15:00 +00:00
|
|
|
sim_consts = ReCo.gen_sim_consts(
|
2022-01-11 00:31:30 +00:00
|
|
|
n_particles,
|
|
|
|
0.0;
|
|
|
|
skin_to_interaction_r_ratio=skin_to_interaction_r_ratio,
|
|
|
|
packing_ratio=packing_ratio,
|
2021-12-28 16:15:00 +00:00
|
|
|
)
|
2022-01-11 00:31:30 +00:00
|
|
|
n_particles = sim_consts.n_particles # This not always equal to the input!
|
2021-12-12 14:29:08 +00:00
|
|
|
|
2022-01-11 00:31:30 +00:00
|
|
|
env = EnvType(sim_consts)
|
2021-12-20 23:31:44 +00:00
|
|
|
|
2022-01-11 17:39:38 +00:00
|
|
|
agent = gen_agent(env.shared.n_states, env.shared.n_actions, ϵ_stable)
|
2021-12-12 14:29:08 +00:00
|
|
|
|
2021-12-12 23:19:18 +00:00
|
|
|
n_steps_before_actions_update = round(Int64, update_actions_at / sim_consts.δt)
|
|
|
|
|
2021-12-20 23:31:44 +00:00
|
|
|
hook = TotalRewardPerEpisode()
|
|
|
|
|
2022-01-11 17:39:38 +00:00
|
|
|
env_helper_params = EnvHelperSharedProps(
|
2021-12-28 16:15:00 +00:00
|
|
|
env,
|
|
|
|
agent,
|
|
|
|
hook,
|
|
|
|
n_steps_before_actions_update,
|
2022-01-06 23:30:05 +00:00
|
|
|
goal_gyration_tensor_eigvals_ratio,
|
2021-12-28 16:15:00 +00:00
|
|
|
n_particles,
|
2021-12-12 17:27:56 +00:00
|
|
|
)
|
|
|
|
|
2022-01-11 00:31:30 +00:00
|
|
|
env_helper = gen_env_helper(env, env_helper_params)
|
|
|
|
|
|
|
|
parent_dir = "RL_" * parent_dir_appendix
|
2021-12-28 16:15:00 +00:00
|
|
|
|
2021-12-14 03:03:14 +00:00
|
|
|
# Pre experiment
|
2021-12-20 23:31:44 +00:00
|
|
|
hook(PRE_EXPERIMENT_STAGE, agent, env)
|
|
|
|
agent(PRE_EXPERIMENT_STAGE, env)
|
2021-12-12 14:29:08 +00:00
|
|
|
|
2021-12-13 01:24:34 +00:00
|
|
|
@showprogress 0.6 for episode in 1:n_episodes
|
2021-12-28 16:15:00 +00:00
|
|
|
dir = ReCo.init_sim_with_sim_consts(sim_consts; parent_dir=parent_dir)
|
2021-12-12 17:27:56 +00:00
|
|
|
|
2021-12-14 03:03:14 +00:00
|
|
|
# Reset
|
2021-12-20 23:31:44 +00:00
|
|
|
reset!(env)
|
2021-12-12 14:29:08 +00:00
|
|
|
|
2021-12-14 03:03:14 +00:00
|
|
|
# Pre espisode
|
2021-12-20 23:31:44 +00:00
|
|
|
hook(PRE_EPISODE_STAGE, agent, env)
|
|
|
|
agent(PRE_EPISODE_STAGE, env)
|
2021-12-14 03:03:14 +00:00
|
|
|
|
|
|
|
# Episode
|
2021-12-16 12:50:38 +00:00
|
|
|
ReCo.run_sim(
|
2022-01-11 00:31:30 +00:00
|
|
|
dir;
|
|
|
|
duration=episode_duration,
|
|
|
|
seed=rand(1:typemax(Int64)),
|
|
|
|
env_helper=env_helper,
|
2021-12-12 14:29:08 +00:00
|
|
|
)
|
2021-12-12 17:27:56 +00:00
|
|
|
|
2022-01-11 17:39:38 +00:00
|
|
|
env.shared.terminated = true
|
2021-12-12 17:27:56 +00:00
|
|
|
|
2021-12-20 23:31:44 +00:00
|
|
|
# Post episode
|
|
|
|
hook(POST_EPISODE_STAGE, agent, env)
|
|
|
|
agent(POST_EPISODE_STAGE, env)
|
2021-12-28 16:15:00 +00:00
|
|
|
|
2022-01-06 00:48:37 +00:00
|
|
|
# TODO: Replace with live plot
|
2021-12-28 16:15:00 +00:00
|
|
|
display(hook.rewards)
|
2022-01-06 00:48:37 +00:00
|
|
|
display(agent.policy.explorer.step)
|
2021-12-12 14:29:08 +00:00
|
|
|
end
|
|
|
|
|
2021-12-14 03:03:14 +00:00
|
|
|
# Post experiment
|
2021-12-20 23:31:44 +00:00
|
|
|
hook(POST_EXPERIMENT_STAGE, agent, env)
|
2021-12-12 17:27:56 +00:00
|
|
|
|
2022-01-11 00:31:30 +00:00
|
|
|
return env_helper
|
2021-12-12 14:29:08 +00:00
|
|
|
end
|
2021-12-10 02:16:45 +00:00
|
|
|
|
2022-01-11 00:31:30 +00:00
|
|
|
include("LocalCOMEnv.jl")
|
|
|
|
|
2021-12-12 14:29:08 +00:00
|
|
|
end # module
|