1
0
Fork 0
mirror of https://gitlab.rlp.net/mobitar/ReCo.jl.git synced 2024-09-19 19:01:17 +00:00

Restructured code to include all environments

This commit is contained in:
Mo8it 2022-01-11 01:31:30 +01:00
parent 9c00da84ea
commit bb3246a1e7
5 changed files with 265 additions and 191 deletions

140
src/RL/LocalCOMEnv.jl Normal file
View file

@ -0,0 +1,140 @@
export LocalCOMEnv
struct LocalCOMEnv <: Env
params::EnvParams
distance_state_space::Vector{Interval}
direction_angle_state_space::Vector{Interval}
max_distance::Float64
function LocalCOMEnv(
sim_consts; n_distance_states::Int64=3, n_direction_angle_states::Int64=3
)
@assert n_direction_angle_states > 1
direction_angle_state_space = gen_angle_state_space(n_direction_angle_states)
min_distance = 0.0
max_distance = sim_consts.skin_r
distance_state_space = gen_distance_state_space(
min_distance, max_distance, n_distance_states
)
n_states = n_distance_states * n_direction_angle_states + 1
state_space = Vector{SVector{2,Interval}}(undef, n_states - 1)
ind = 1
for distance_state in distance_state_space
for direction_angle_state in direction_angle_state_space
state_space[ind] = SVector(distance_state, direction_angle_state)
ind += 1
end
end
# Last state is when no particle is in the skin radius
params = EnvParams(n_states, state_space)
return new(params, distance_state_space, direction_angle_state_space, max_distance)
end
end
struct LocalCOMEnvHelper <: EnvHelper
params::EnvHelperParams
vec_to_neighbour_sums::Vector{SVector{2,Float64}}
n_neighbours::Vector{Int64}
function LocalCOMEnvHelper(params::EnvHelperParams)
return new(
params, fill(SVector(0.0, 0.0), params.n_particles), fill(0, params.n_particles)
)
end
end
function gen_env_helper(::LocalCOMEnv, env_helper_params::EnvHelperParams)
return LocalCOMEnvHelper(env_helper_params)
end
function pre_integration_hook(env_helper::LocalCOMEnvHelper)
@simd for id in 1:(env_helper.params.n_particles)
env_helper.vec_to_neighbour_sums[id] = SVector(0.0, 0.0)
env_helper.n_neighbours[id] = 0
end
return nothing
end
function state_update_helper_hook(
env_helper::LocalCOMEnvHelper, id1::Int64, id2::Int64, r⃗₁₂::SVector{2,Float64}
)
env_helper.vec_to_neighbour_sums[id1] += r⃗₁₂
env_helper.vec_to_neighbour_sums[id2] -= r⃗₁₂
env_helper.n_neighbours[id1] += 1
env_helper.n_neighbours[id2] += 1
return nothing
end
function state_update_hook(env_helper::LocalCOMEnvHelper, particles::Vector{Particle})
n_particles = env_helper.params.n_particles
@turbo for id in 1:(n_particles)
env_helper.params.old_states_ind[id] = env_helper.params.states_ind[id]
end
env = env_helper.params.env
for id in 1:n_particles
n_neighbours = env_helper.n_neighbours[id]
if n_neighbours == 0
state_ind = env.params.n_states
else
vec_to_local_center_of_mass =
env_helper.vec_to_neighbour_sums[id] / n_neighbours
distance = sqrt(
vec_to_local_center_of_mass[1]^2 + vec_to_local_center_of_mass[2]^2
)
distance_state = find_state_interval(distance, env.distance_state_space)
si, co = sincos(particles[id].φ)
direction_angle = angle2(SVector(co, si), vec_to_local_center_of_mass)
direction_angle_state = find_state_interval(
direction_angle, env.direction_angle_state_space
)
state = SVector{2,Interval}(distance_state, direction_angle_state)
state_ind = find_state_ind(state, env.params.state_space)
end
env_helper.params.states_ind[id] = state_ind
end
return nothing
end
function update_reward!(env::LocalCOMEnv, env_helper::LocalCOMEnvHelper, particle::Particle)
id = particle.id
normalization = (env.max_distance * env_helper.params.n_particles)
n_neighbours = env_helper.n_neighbours[id]
if n_neighbours == 0
env.params.reward = -(env.max_distance^2) / normalization
else
vec_to_local_center_of_mass = env_helper.vec_to_neighbour_sums[id] / n_neighbours # TODO: Reuse vec_to_local_center_of_mass from state_update_hook
env.params.reward =
-(vec_to_local_center_of_mass[1]^2 + vec_to_local_center_of_mass[2]^2) /
normalization
end
return nothing
end

View file

@ -1,6 +1,6 @@
module RL
export run_rl
export run_rl, LocalCOMEnv
using Base: OneTo
@ -12,12 +12,14 @@ using LoopVectorization: @turbo
using Random: Random
using ProgressMeter: @showprogress
using ..ReCo: ReCo, Particle, angle2, Shape
using ..ReCo: ReCo, Particle, angle2, Shape, DEFAULT_SKIN_TO_INTERACTION_R_RATIO
const INITIAL_REWARD = 0.0
const INITIAL_STATE_IND = 1
const INITIAL_REWARD = 0.0
function angle_state_space(n_angle_states::Int64)
method_not_implemented() = error("Method not implemented!")
function gen_angle_state_space(n_angle_states::Int64)
angle_range = range(; start=-π, stop=π, length=n_angle_states + 1)
angle_state_space = Vector{Interval}(undef, n_angle_states)
@ -37,40 +39,61 @@ function angle_state_space(n_angle_states::Int64)
return angle_state_space
end
mutable struct Env <: AbstractEnv
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
mutable struct EnvParams{state_dims}
n_actions::Int64
action_space::Vector{SVector{2,Float64}}
action_ind_space::OneTo{Int64}
distance_state_space::Vector{Interval}
direction_angle_state_space::Vector{Interval}
n_states::Int64
state_space::Vector{SVector{2,Interval}}
state_space::Vector{SVector{state_dims,Interval}}
state_ind_space::OneTo{Int64}
state_ind::Int64
reward::Float64
terminated::Bool
function Env(;
max_distance::Float64,
min_distance::Float64=0.0,
function EnvParams(
n_states::Int64,
state_space::Vector{SVector{state_dims,Interval}};
n_v_actions::Int64=2,
n_ω_actions::Int64=3,
max_v::Float64=40.0,
max_ω::Float64=π / 2,
n_distance_states::Int64=3,
n_direction_angle_states::Int64=3,
)
@assert min_distance >= 0.0
@assert max_distance > min_distance
) where {state_dims}
@assert n_v_actions > 1
@assert n_ω_actions > 1
@assert max_v > 0
@assert max_ω > 0
@assert n_distance_states > 1
@assert n_direction_angle_states > 1
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)
@ -89,47 +112,12 @@ mutable struct Env <: AbstractEnv
action_ind_space = OneTo(n_actions)
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
direction_angle_state_space = angle_state_space(n_direction_angle_states)
n_states = n_distance_states * n_direction_angle_states + 1
state_space = Vector{SVector{2,Interval}}(undef, n_states - 1)
ind = 1
for distance_state in distance_state_space
for direction_angle_state in direction_angle_state_space
state_space[ind] = SVector(distance_state, direction_angle_state)
ind += 1
end
end
# Last state is when no particle is in the skin radius
state_ind_space = OneTo(n_states)
return new(
return new{state_dims}(
n_actions,
action_space,
action_ind_space,
distance_state_space,
direction_angle_state_space,
n_states,
state_space,
state_ind_space,
@ -141,94 +129,78 @@ mutable struct Env <: AbstractEnv
end
function reset!(env::Env)
env.state_ind = env.n_states
env.terminated = false
env.params.terminated = false
return nothing
end
RLBase.state_space(env::Env) = env.state_ind_space
RLBase.state_space(env::Env) = env.params.state_ind_space
RLBase.state(env::Env) = env.state_ind
RLBase.state(env::Env) = env.params.state_ind
RLBase.action_space(env::Env) = env.action_ind_space
RLBase.action_space(env::Env) = env.params.action_ind_space
RLBase.reward(env::Env) = env.reward
RLBase.reward(env::Env) = env.params.reward
RLBase.is_terminated(env::Env) = env.terminated
RLBase.is_terminated(env::Env) = env.params.terminated
struct Params{H<:AbstractHook}
struct EnvHelperParams{H<:AbstractHook}
env::Env
agent::Agent
hook::H
n_steps_before_actions_update::Int64
goal_gyration_tensor_eigvals_ratio::Float64
n_particles::Int64
old_states_ind::Vector{Int64}
states_ind::Vector{Int64}
actions::Vector{SVector{2,Float64}}
actions_ind::Vector{Int64}
n_steps_before_actions_update::Int64
goal_gyration_tensor_eigvals_ratio::Float64
n_particles::Int64
max_distance::Float64
vec_to_neighbour_sums::Vector{SVector{2,Float64}}
n_neighbours::Vector{Int64}
function Params(
function EnvHelperParams(
env::Env,
agent::Agent,
hook::H,
n_steps_before_actions_update::Int64,
goal_gyration_tensor_eigvals_ratio::Float64,
n_particles::Int64,
max_distance::Float64,
) where {H<:AbstractHook}
n_states = env.n_states
return new{H}(
env,
agent,
hook,
fill(0, n_particles),
fill(n_states, n_particles),
fill(SVector(0.0, 0.0), n_particles),
fill(0, n_particles),
n_steps_before_actions_update,
goal_gyration_tensor_eigvals_ratio,
n_particles,
max_distance,
fill(0, n_particles),
fill(0, n_particles),
fill(SVector(0.0, 0.0), n_particles),
fill(0, n_particles),
)
end
end
function pre_integration_hook(rl_params::Params)
@simd for id in 1:(rl_params.n_particles)
rl_params.vec_to_neighbour_sums[id] = SVector(0.0, 0.0)
rl_params.n_neighbours[id] = 0
end
abstract type EnvHelper end
return nothing
function gen_env_helper(::Env, env_helper_params::EnvHelperParams)
return method_not_implemented()
end
function pre_integration_hook(::EnvHelper)
return method_not_implemented()
end
function state_update_helper_hook(
rl_params::Params, id1::Int64, id2::Int64, r⃗₁₂::SVector{2,Float64}
::EnvHelper, id1::Int64, id2::Int64, r⃗₁₂::SVector{2,Float64}
)
rl_params.vec_to_neighbour_sums[id1] += r⃗₁₂
rl_params.vec_to_neighbour_sums[id2] -= r⃗₁₂
rl_params.n_neighbours[id1] += 1
rl_params.n_neighbours[id2] += 1
return nothing
return method_not_implemented()
end
function find_state_ind(state::S, state_space::Vector{S}) where {S<:SVector{2,Interval}}
function find_state_ind(state::S, state_space::Vector{S}) where {S<:SVector}
return findfirst(x -> x == state, state_space)
end
@ -240,89 +212,40 @@ function find_state_interval(value::Float64, state_space::Vector{Interval})::Int
end
end
function state_update_hook(rl_params::Params, particles::Vector{Particle})
@turbo for id in 1:(rl_params.n_particles)
rl_params.old_states_ind[id] = rl_params.states_ind[id]
end
env = rl_params.env
for id in 1:(rl_params.n_particles)
n_neighbours = rl_params.n_neighbours[id]
if n_neighbours == 0
state_ind = env.n_states
else
vec_to_local_center_of_mass = rl_params.vec_to_neighbour_sums[id] / n_neighbours
distance = sqrt(
vec_to_local_center_of_mass[1]^2 + vec_to_local_center_of_mass[2]^2
)
distance_state = find_state_interval(distance, env.distance_state_space)
si, co = sincos(particles[id].φ)
direction_angle = angle2(SVector(co, si), vec_to_local_center_of_mass)
direction_angle_state = find_state_interval(
direction_angle, env.direction_angle_state_space
)
state = SVector{2,Interval}(distance_state, direction_angle_state)
state_ind = find_state_ind(state, env.state_space)
end
rl_params.states_ind[id] = state_ind
end
return nothing
function state_update_hook(::EnvHelper, particles::Vector{Particle})
return method_not_implemented()
end
function get_env_agent_hook(rl_params::Params)
return (rl_params.env, rl_params.agent, rl_params.hook)
function get_env_agent_hook(env_helper::EnvHelper)
return (env_helper.params.env, env_helper.params.agent, env_helper.params.hook)
end
function update_reward!(env::Env, rl_params::Params, particle::Particle)
id = particle.id
normalization = (rl_params.max_distance * rl_params.n_particles)
n_neighbours = rl_params.n_neighbours[id]
if n_neighbours == 0
env.reward = -(rl_params.max_distance^2) / normalization
else
vec_to_local_center_of_mass = rl_params.vec_to_neighbour_sums[id] / n_neighbours # TODO: Reuse vec_to_local_center_of_mass from state_update_hook
env.reward =
-(vec_to_local_center_of_mass[1]^2 + vec_to_local_center_of_mass[2]^2) /
normalization
end
return nothing
function update_reward!(::Env, ::EnvHelper, particle::Particle)
return method_not_implemented()
end
function update_table_and_actions_hook(
rl_params::Params, particle::Particle, first_integration_step::Bool
env_helper::EnvHelper, particle::Particle, first_integration_step::Bool
)
env, agent, hook = get_env_agent_hook(rl_params)
env, agent, hook = get_env_agent_hook(env_helper)
id = particle.id
if !first_integration_step
# Old state
env.state_ind = rl_params.old_states_ind[id]
env.params.state_ind = env_helper.params.old_states_ind[id]
action_ind = rl_params.actions_ind[id]
action_ind = env_helper.params.actions_ind[id]
# Pre act
agent(PRE_ACT_STAGE, env, action_ind)
hook(PRE_ACT_STAGE, agent, env, action_ind)
# Update to current state
env.state_ind = rl_params.states_ind[id]
env.params.state_ind = env_helper.params.states_ind[id]
# Update reward
update_reward!(env, rl_params, particle)
update_reward!(env, env_helper, particle)
# Post act
agent(POST_ACT_STAGE, env)
@ -331,10 +254,10 @@ function update_table_and_actions_hook(
# Update action
action_ind = agent(env)
action = env.action_space[action_ind]
action = env.params.action_space[action_ind]
rl_params.actions[id] = action
rl_params.actions_ind[id] = action_ind
env_helper.params.actions[id] = action
env_helper.params.actions_ind[id] = action_ind
return nothing
end
@ -342,10 +265,10 @@ end
act_hook(::Nothing, args...) = nothing
function act_hook(
rl_params::Params, particle::Particle, δt::Float64, si::Float64, co::Float64
env_helper::EnvHelper, particle::Particle, δt::Float64, si::Float64, co::Float64
)
# Apply action
action = rl_params.actions[particle.id]
action = env_helper.params.actions[particle.id]
vδt = action[1] * δt
particle.tmp_c += SVector(vδt * co, vδt * si)
@ -378,6 +301,8 @@ function gen_agent(n_states::Int64, n_actions::Int64, ϵ_stable::Float64)
end
function run_rl(;
EnvType::Type{E},
parent_dir_appendix::String,
goal_gyration_tensor_eigvals_ratio::Float64,
n_episodes::Int64=200,
episode_duration::Float64=50.0,
@ -385,8 +310,9 @@ function run_rl(;
n_particles::Int64=100,
seed::Int64=42,
ϵ_stable::Float64=0.0001,
parent_dir::String="",
)
skin_to_interaction_r_ratio::Float64=DEFAULT_SKIN_TO_INTERACTION_R_RATIO,
packing_ratio=0.22,
) where {E<:Env}
@assert 0.0 <= goal_gyration_tensor_eigvals_ratio <= 1.0
@assert n_episodes > 0
@assert episode_duration > 0
@ -398,30 +324,33 @@ function run_rl(;
Random.seed!(seed)
sim_consts = ReCo.gen_sim_consts(
n_particles, 0.0; skin_to_interaction_r_ratio=2.0, packing_ratio=0.22
n_particles,
0.0;
skin_to_interaction_r_ratio=skin_to_interaction_r_ratio,
packing_ratio=packing_ratio,
)
n_particles = sim_consts.n_particles
n_particles = sim_consts.n_particles # This not always equal to the input!
max_distance = sim_consts.skin_r
env = Env(; max_distance=max_distance)
env = EnvType(sim_consts)
agent = gen_agent(env.n_states, env.n_actions, ϵ_stable)
agent = gen_agent(env.params.n_states, env.params.n_actions, ϵ_stable)
n_steps_before_actions_update = round(Int64, update_actions_at / sim_consts.δt)
hook = TotalRewardPerEpisode()
rl_params = Params(
env_helper_params = EnvHelperParams(
env,
agent,
hook,
n_steps_before_actions_update,
goal_gyration_tensor_eigvals_ratio,
n_particles,
max_distance,
)
parent_dir = "RL" * parent_dir
env_helper = gen_env_helper(env, env_helper_params)
parent_dir = "RL_" * parent_dir_appendix
# Pre experiment
hook(PRE_EXPERIMENT_STAGE, agent, env)
@ -439,10 +368,13 @@ function run_rl(;
# Episode
ReCo.run_sim(
dir; duration=episode_duration, seed=rand(1:typemax(Int64)), rl_params=rl_params
dir;
duration=episode_duration,
seed=rand(1:typemax(Int64)),
env_helper=env_helper,
)
env.terminated = true
env.params.terminated = true
# Post episode
hook(POST_EPISODE_STAGE, agent, env)
@ -456,7 +388,9 @@ function run_rl(;
# Post experiment
hook(POST_EXPERIMENT_STAGE, agent, env)
return rl_params
return env_helper
end
include("LocalCOMEnv.jl")
end # module

View file

@ -1,6 +1,6 @@
module ReCo
export init_sim, run_sim, run_rl, animate
export init_sim, run_sim, run_rl, animate, LocalCOMEnv
using StaticArrays: SVector
using OrderedCollections: OrderedDict
@ -26,7 +26,7 @@ include("setup.jl")
include("Shape.jl")
using .Shape
include("RL.jl")
include("RL/RL.jl")
using .RL
include("simulation.jl")

View file

@ -6,7 +6,7 @@ function run_sim(
snapshot_at::Float64=0.1,
seed::Int64=42,
n_bundle_snapshots::Int64=100,
rl_params::Union{RL.Params,Nothing}=nothing,
env_helper::Union{RL.EnvHelper,Nothing}=nothing,
)
@assert length(dir) > 0
@assert duration > 0
@ -111,7 +111,7 @@ function run_sim(
n_bundles,
dir,
save_data,
rl_params,
env_helper,
)
return nothing

View file

@ -35,7 +35,7 @@ end
function euler!(
args,
first_integration_step::Bool,
rl_params::Union{RL.Params,Nothing},
env_helper::Union{RL.EnvHelper,Nothing},
state_update_helper_hook::Function,
state_update_hook::Function,
update_table_and_actions_hook::Function,
@ -52,7 +52,7 @@ function euler!(
p1_c, p2.c, args.interaction_r², args.half_box_len
)
state_update_helper_hook(rl_params, id1, id2, r⃗₁₂)
state_update_helper_hook(env_helper, id1, id2, r⃗₁₂)
if overlapping
factor = args.c₁ / (distance²^4) * (args.c₂ / (distance²^3) - 1.0)
@ -64,7 +64,7 @@ function euler!(
end
end
state_update_hook(rl_params, args.particles)
state_update_hook(env_helper, args.particles)
@simd for p in args.particles
si, co = sincos(p.φ)
@ -75,9 +75,9 @@ function euler!(
restrict_coordinates!(p, args.half_box_len)
update_table_and_actions_hook(rl_params, p, first_integration_step)
update_table_and_actions_hook(env_helper, p, first_integration_step)
RL.act_hook(rl_params, p, args.δt, si, co)
RL.act_hook(env_helper, p, args.δt, si, co)
p.φ += args.c₄ * rand_normal01()
@ -91,8 +91,8 @@ Base.wait(::Nothing) = nothing
gen_run_additional_hooks(::Nothing, args...) = false
function gen_run_additional_hooks(rl_params::RL.Params, integration_step::Int64)
return (integration_step % rl_params.n_steps_before_actions_update == 0) ||
function gen_run_additional_hooks(env_helper::RL.EnvHelper, integration_step::Int64)
return (integration_step % env_helper.params.n_steps_before_actions_update == 0) ||
(integration_step == 1)
end
@ -105,7 +105,7 @@ function simulate(
n_bundles::Int64,
dir::String,
save_data::Bool,
rl_params::Union{RL.Params,Nothing},
env_helper::Union{RL.EnvHelper,Nothing},
)
bundle_snapshot_counter = 0
@ -143,10 +143,10 @@ function simulate(
cl = update_verlet_lists!(args, cl)
end
run_additional_hooks = gen_run_additional_hooks(rl_params, integration_step)
run_additional_hooks = gen_run_additional_hooks(env_helper, integration_step)
if run_additional_hooks
RL.pre_integration_hook(rl_params)
RL.pre_integration_hook(env_helper)
state_update_helper_hook = RL.state_update_helper_hook
state_update_hook = RL.state_update_hook
@ -156,7 +156,7 @@ function simulate(
euler!(
args,
first_integration_step,
rl_params,
env_helper,
state_update_helper_hook,
state_update_hook,
update_table_and_actions_hook,