mirror of
https://gitlab.rlp.net/mobitar/ReCo.jl.git
synced 2024-12-21 00:51:21 +00:00
Save env_helper and added states and actions labels
This commit is contained in:
parent
3e713fd44c
commit
839e766206
4 changed files with 64 additions and 8 deletions
|
@ -10,12 +10,16 @@ mutable struct EnvSharedProps{n_state_dims}
|
|||
state_id_space::OneTo{Int64}
|
||||
state_id::Int64
|
||||
|
||||
action_spaces_labels::Vector{Vector{LaTeXStrings.LaTeXString}}
|
||||
state_spaces_labels::Vector{Vector{LaTeXStrings.LaTeXString}}
|
||||
|
||||
reward::Float64
|
||||
terminated::Bool
|
||||
|
||||
function EnvSharedProps(
|
||||
n_states::Int64, # Can be different from the sum of state_id_tensor_dims
|
||||
state_id_tensor_dims::NTuple{n_state_dims,Int64};
|
||||
state_id_tensor_dims::NTuple{n_state_dims,Int64},
|
||||
state_spaces_labels::Vector{Vector{LaTeXStrings.LaTeXString}};
|
||||
n_v_actions::Int64=2,
|
||||
n_ω_actions::Int64=3,
|
||||
max_v::Float64=40.0,
|
||||
|
@ -29,6 +33,10 @@ mutable struct EnvSharedProps{n_state_dims}
|
|||
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)
|
||||
|
||||
action_spaces_labels = gen_action_spaces_labels(
|
||||
("v", "\\omega"), (v_action_space, ω_action_space)
|
||||
)
|
||||
|
||||
n_actions = n_v_actions * n_ω_actions
|
||||
|
||||
action_space = Vector{SVector{2,Float64}}(undef, n_actions)
|
||||
|
@ -61,6 +69,8 @@ mutable struct EnvSharedProps{n_state_dims}
|
|||
state_id_tensor,
|
||||
state_id_space,
|
||||
INITIAL_STATE_IND,
|
||||
action_spaces_labels,
|
||||
state_spaces_labels,
|
||||
INITIAL_REWARD,
|
||||
false,
|
||||
)
|
||||
|
@ -92,3 +102,20 @@ end
|
|||
function RLBase.is_terminated(env::Env)
|
||||
return env.shared.terminated
|
||||
end
|
||||
|
||||
function gen_action_space_labels(action_label::String, action_space::AbstractRange)
|
||||
labels = Vector{LaTeXStrings.LaTeXString}(undef, length(action_space))
|
||||
for (action_ind, action) in enumerate(action_space)
|
||||
labels[action_ind] = LaTeXStrings.latexstring(
|
||||
action_label * " = $(round(action; digits=2))"
|
||||
)
|
||||
end
|
||||
|
||||
return labels
|
||||
end
|
||||
|
||||
function gen_action_spaces_labels(
|
||||
actions_labels::NTuple{N,String}, action_spaces::NTuple{N,AbstractRange}
|
||||
) where {N}
|
||||
return [gen_action_space_labels(actions_labels[i], action_spaces[i]) for i in 1:N]
|
||||
end
|
|
@ -9,7 +9,7 @@ struct OriginEnv <: Env
|
|||
direction_angle_state_space::Vector{Interval}
|
||||
|
||||
function OriginEnv(;
|
||||
n_distance_states::Int64=4, n_direction_angle_states::Int64=3, args
|
||||
n_distance_states::Int64=3, n_direction_angle_states::Int64=3, args
|
||||
)
|
||||
@assert n_distance_states > 1
|
||||
@assert n_direction_angle_states > 1
|
||||
|
@ -25,7 +25,13 @@ struct OriginEnv <: Env
|
|||
|
||||
n_states = n_distance_states * n_direction_angle_states
|
||||
|
||||
shared = EnvSharedProps(n_states, (n_distance_states, n_direction_angle_states))
|
||||
state_spaces_labels = gen_state_spaces_labels(
|
||||
("d", "\\theta"), (distance_state_space, direction_angle_state_space)
|
||||
)
|
||||
|
||||
shared = EnvSharedProps(
|
||||
n_states, (n_distance_states, n_direction_angle_states), state_spaces_labels
|
||||
)
|
||||
|
||||
return new(shared, distance_state_space, direction_angle_state_space)
|
||||
end
|
||||
|
|
29
src/RL/RL.jl
29
src/RL/RL.jl
|
@ -11,6 +11,8 @@ using StaticArrays: SVector
|
|||
using LoopVectorization: @turbo
|
||||
using Random: Random
|
||||
using ProgressMeter: ProgressMeter
|
||||
using JLD2: JLD2
|
||||
using LaTeXStrings: LaTeXStrings, @L_str
|
||||
|
||||
using ..ReCo: ReCo
|
||||
|
||||
|
@ -54,7 +56,7 @@ end
|
|||
|
||||
function run_rl(;
|
||||
EnvType::Type{E},
|
||||
parent_dir_appendix::String,
|
||||
process_dir::String,
|
||||
elliptical_a_b_ratio::Float64=1.0,
|
||||
n_episodes::Int64=200,
|
||||
episode_duration::Float64=50.0,
|
||||
|
@ -85,7 +87,6 @@ function run_rl(;
|
|||
n_particles = sim_consts.n_particles # Not always equal to the input!
|
||||
|
||||
env_args = (skin_radius=sim_consts.skin_radius, half_box_len=sim_consts.half_box_len)
|
||||
|
||||
env = EnvType(; args=env_args)
|
||||
|
||||
agent = gen_agent(env.shared.n_states, env.shared.n_actions, ϵ_stable)
|
||||
|
@ -104,7 +105,7 @@ function run_rl(;
|
|||
|
||||
env_helper = gen_env_helper(env, env_helper_shared; args=env_helper_args)
|
||||
|
||||
parent_dir = "RL_" * parent_dir_appendix
|
||||
parent_dir = "RL/" * process_dir
|
||||
|
||||
# Pre experiment
|
||||
hook(PRE_EXPERIMENT_STAGE, agent, env)
|
||||
|
@ -137,9 +138,31 @@ function run_rl(;
|
|||
# Post experiment
|
||||
hook(POST_EXPERIMENT_STAGE, agent, env)
|
||||
|
||||
process_dir = ReCo.DEFAULT_EXPORTS_DIR * "/$parent_dir"
|
||||
|
||||
JLD2.save_object("$process_dir/env_helper.jld2", env_helper)
|
||||
|
||||
return env_helper
|
||||
end
|
||||
|
||||
function gen_state_space_labels(state_label::String, state_space::Vector{Interval})
|
||||
labels = Vector{LaTeXStrings.LaTeXString}(undef, length(state_space))
|
||||
for (state_ind, state) in enumerate(state_space)
|
||||
labels[state_ind] = LaTeXStrings.latexstring(
|
||||
state_label *
|
||||
" = $(round(state.first; digits=2)):$(round(state.last, digits=2))",
|
||||
)
|
||||
end
|
||||
|
||||
return labels
|
||||
end
|
||||
|
||||
function gen_state_spaces_labels(
|
||||
states_labels::NTuple{N,String}, state_spaces::NTuple{N,Vector{Interval}}
|
||||
) where {N}
|
||||
return [gen_state_space_labels(states_labels[i], state_spaces[i]) for i in 1:N]
|
||||
end
|
||||
|
||||
include("Envs/LocalCOMWithAdditionalShapeRewardEnv.jl")
|
||||
include("Envs/OriginEnv.jl")
|
||||
|
||||
|
|
|
@ -1,6 +1,6 @@
|
|||
const DEFAULT_PACKING_RATIO = 0.5
|
||||
const DEFAULT_δt = 1e-5
|
||||
const DEFAULT_SKIN_TO_INTERACTION_R_RATIO = 2.0
|
||||
const DEFAULT_SKIN_TO_INTERACTION_R_RATIO = 2.5
|
||||
const DEFAULT_EXPORTS_DIR = "exports"
|
||||
const DEFAULT_PARENT_DIR = ""
|
||||
const DEFAULT_COMMENT = ""
|
||||
|
@ -69,7 +69,7 @@ function gen_sim_consts(
|
|||
|
||||
skin_radius = skin_to_interaction_radius_ratio * interaction_radius
|
||||
|
||||
buffer = 2.5
|
||||
buffer = 3
|
||||
max_approach_after_one_integration_step = buffer * (2 * v₀ * δt)
|
||||
@assert skin_radius >= interaction_radius + max_approach_after_one_integration_step
|
||||
|
||||
|
|
Loading…
Reference in a new issue