youainti
/
Orbits
1
0
Fork 0
Code Issues 10 Pull Requests Packages Projects Releases Wiki Activity

current work on making operator level maximization as opposed to just the planner level.

Browse Source
temporaryWork^2
youainti 5 years ago
parent 2414e15735
commit 5d60a23270
2 changed files with 1221 additions and 64 deletions
Show Stats Download Patch File Download Diff File

167
julia_code/BellmanResidualMinimization.jl
Unescape Escape View File

@ -69,8 +69,8 @@ end
# ╔═╡ 5b45b29e-f0f4-41e9-91e7-d444687feb4e
#implement survival function
function survival(
stocks::Array
,debris::Array
stocks::Array{Float32}
,debris::Array{Float32}
,physical_model::BasicModel
)
return exp.(
@ -82,9 +82,9 @@ end
# ╔═╡ 152f3a3c-a565-41bb-8e59-6ab0d2315ffb
#debris evolution
function H(
stocks::Array
,debris::Array
,launches::Array
stocks::Array{Float32}
,debris::Array{Float32}
,launches::Array{Float32}
, physical_model::BasicModel
)
#get changes in debris from natural dynamics
@ -103,9 +103,9 @@ end
# ╔═╡ 25ac9438-2b1d-4f6b-9ff1-1695e1d52b51
#stock update rules
function G(
stocks::Array
,debris::Array
,launches::Array
stocks::Array{Float32}
,debris::Array{Float32}
,launches::Array{Float32}
, physical_model::BasicModel
)
return LinearAlgebra.diagm(survival(stocks,debris,physical_model) .- physical_model.decay_rate)*stocks + launches
@ -117,42 +117,44 @@ md"""
"""
# ╔═╡ f7aabe43-9a2c-4fe0-8099-c29cdf66566c
function value_function_generator(number_params=10)
function value_function_generator(number_params=32)
return Flux.Chain(
Flux.Parallel(vcat
#parallel joins together stocks and debris, after a little bit of preprocessing
,Flux.Chain(
Flux.Dense(N_constellations, N_states*2,Flux.relu)
,Flux.Dense(N_states*2, N_states*2,Flux.σ)
Flux.Dense(N_constellations, number_params*2,Flux.relu)
,Flux.Dense(number_params*2, number_params*2,Flux.σ)
)
,Flux.Chain(
Flux.Dense(N_debris, N_states,Flux.relu)
,Flux.Dense(N_states, N_states,Flux.σ)
Flux.Dense(N_debris, number_params,Flux.relu)
,Flux.Dense(number_params, number_params,Flux.σ)
)
)
#Apply some transformations to the preprocessed data.
,Flux.Dense(N_states*3,number_params,Flux.σ)
,Flux.Dense(number_params*3,number_params,Flux.σ)
,Flux.Dense(number_params,number_params,Flux.σ)
,Flux.Dense(number_params,1)
)
end
# ╔═╡ d816b252-bdca-44ba-ac5c-cb21163a1e9a
function policy_function_generator(number_params=10)
function planner_policy_function_generator(number_params=32)
return Flux.Chain(
Flux.Parallel(vcat
#parallel joins together stocks and debris
,Flux.Chain(
Flux.Dense(N_constellations, N_states*2,Flux.relu)
,Flux.Dense(N_states*2, N_states*2,Flux.σ)
Flux.Dense(N_constellations, number_params*2,Flux.relu)
,Flux.Dense(number_params*2, number_params*2,Flux.σ)
)
,Flux.Chain(
Flux.Dense(N_debris, N_states,Flux.relu)
,Flux.Dense(N_states, N_states)
Flux.Dense(N_debris, number_params,Flux.relu)
,Flux.Dense(number_params, number_params)
)
)
#Apply some transformations
,Flux.Dense(N_states*3,number_params,Flux.σ)
,Flux.Dense(number_params*3,number_params,Flux.σ)
,Flux.Dense(number_params,number_params,Flux.σ)
,Flux.Dense(number_params,N_constellations,Flux.relu)
)
@ -176,6 +178,12 @@ begin
Flux.@functor Split
(m::Split)(x::AbstractArray) = tuple(map(f -> f(x), m.paths))
struct Duplicate
n::Int
end
Flux.@functor Duplicate
(m::Duplicate)(x::Tuple) = [x for i=1:m.n]
### TESTING ###
@ -187,21 +195,55 @@ begin
end
# ╔═╡ 6a3b5f7a-a535-450f-8c5f-19bdcc280146
function operators_policy_function_generator(number_params=32)
function f()
return Flux.Chain(
Flux.Parallel(vcat
#parallel joins together stocks and debris
,Flux.Chain(
Flux.Dense(N_constellations, number_params,Flux.relu)
,Flux.Dense(number_params, number_params,Flux.σ)
)
,Flux.Chain(
Flux.Dense(N_debris, number_params,Flux.relu)
,Flux.Dense(number_params, number_params)
)
)
#Apply some transformations
,Flux.Dense(number_params*3,number_params,Flux.σ)
,Flux.Dense(number_params,number_params,Flux.σ)
,Flux.Dense(number_params,1,Flux.relu)
)
end
a = [f() for i=1:N_constellations]
Flux.Chain(
Duplicate(N_constellations)
)
end
# ╔═╡ 0ba48a75-36db-4003-95c0-2329d4fb29c5
Duplicate(2)(([1.0f0 2],[3f0]))
# ╔═╡ 340da189-f443-4376-a82d-7699a21ab7a2
abstract type EconomicParameters end
# ╔═╡ 206ac4cc-5102-4381-ad8a-777b02dc4d5a
begin #basic linear model
struct EconModel1 <: EconomicParameters
β::Real
payoff_array::Array{Real}
policy_costs::Array{Real}
β::Float32
payoff_array::Array{Float32}
policy_costs::Array{Float32}
end
function payoff1(
s::Vector
,d::Vector
,a::Vector
s::Vector{Float32}
,d::Vector{Float32}
,a::Vector{Float32}
,em::EconModel1
)
return em.payoff_array*s - em.policy_costs*a
@ -212,16 +254,16 @@ end
# ╔═╡ eebb8706-a431-4fd1-b7a5-40f07a63d5cb
begin #basic CES model
struct CESParams <: EconomicParameters
β::Real
r::Real #elasticity of subsititution
payoff_array::Array{Real}
policy_costs::Array{Real}
debris_costs::Array{Real}
β::Float32
r::Float32 #elasticity of subsititution
payoff_array::Array{Float32}
policy_costs::Array{Float32}
debris_costs::Array{Float32}
end
function CES_with_debris(
s::Vector
,d::Vector
,a::Vector
s::Vector{Float32}
,d::Vector{Float32}
,a::Vector{Float32}
,em::CESParams
)
return (em.payoff_array*(s.^em.r) - em.debris_costs*(d.^em.r)).^(1/em.r) - em.policy_costs*a
@ -250,21 +292,25 @@ md"""
"""
# ╔═╡ 65e0b1fa-d5e1-4ff6-8736-c9d6b5f40150
em1 = EconModel1(0.95, [1 0 0 0], [5 0 0 0])
begin
em1_a = EconModel1(0.95, [1.0 0 0 0], [5 0 0 0])
em1_b = EconModel1(0.95, [0 1.0 0 0], [0 5 0 0])
em1_c = EconModel1(0.95, [0 0 1.0 0], [0 0 5 0])
em1_d = EconModel1(0.95, [0 0 0 1.0], [0 0 0 5])
#=
This is the most basic profit model
You earn 1 per operating satellite and it costs 5 per launch.
Only interaction is through debris.
=#
end
# ╔═╡ 19ccfc3a-6dbb-4c64-bf03-e2e219ef0efe
begin
em2_a = EconModel1(0.95, [1 -0.02 -0.02 0], [5 0 0 0])
em2_b = EconModel1(0.95, [-0.02 1 -0.02 0], [0 5 0 0])
em2_c = EconModel1(0.95, [0 -0.02 1 -0.02], [0 0 5 0])
em2_d = EconModel1(0.95, [0 -0.02 -0.02 1], [0 0 0 5])
em2_a = EconModel1(0.95, [1 -0.02 -0.02 0], [5.0 0 0 0])
em2_b = EconModel1(0.95, [-0.02 1 -0.02 0], [0.0 5 0 0])
em2_c = EconModel1(0.95, [0 -0.02 1 -0.02], [0.0 0 5 0])
em2_d = EconModel1(0.95, [0 -0.02 -0.02 1], [0.0 0 0 5])
#=
This is a simple addition to the basic model, where you lose some benefit based
the size of your competitor's satellites.
@ -287,13 +333,13 @@ md"""
"""
# ╔═╡ fb6aacff-c42d-4ec1-88cb-5ce1b2e8874f
policy = policy_function_generator();
policy = planner_policy_function_generator();
# ╔═╡ 41271ab4-1ec7-431f-9efb-0f7c3da2d8b4
#Constellation level loss function
function Ξ(
s::Vector
,d::Vector
s::Vector{Float32}
,d::Vector{Float32}
, physical_model::PhysicalParameters
,co::ConstellationOperator
)
@ -322,14 +368,17 @@ end
# ╔═╡ 43b99708-0052-4b78-886c-92ac2b532f29
begin #testing
s1 = ones(N_constellations)
d1 = ones(N_debris)
s1 = ones(Float32,N_constellations)
d1 = ones(Float32,N_debris)
Ξ(s1,d1,bm,operators[1])
end
# ╔═╡ dff642d9-ec5a-4fed-a059-6c07760a3a58
#planner's loss function
function planners_loss(s,d)
function planners_loss(
s::Vector{Float32}
,d::Vector{Float32}
)
l = 0.0
for co in operators
l += Ξ(s,d,bm,co)
@ -340,13 +389,13 @@ end
# ╔═╡ 5abebc1a-370c-4f5f-8826-dc0b143d5166
md"""
## Constructing data
## Constructing data and training
"""
# ╔═╡ a20959be-65e4-4b69-9521-503bc59f0854
begin
N=200 #increase later
data = [(rand(1:500, N_constellations),rand(1:500, N_debris)) for n=1:N]
data = [(rand(1:500f0, N_constellations),rand(1:500f0, N_debris)) for n=1:N]
end
# ╔═╡ 6bf8d29a-7990-4e91-86e6-d9894ed3db27
@ -355,7 +404,7 @@ ADAM = Flux.Optimise.ADAM(0.1)
# ╔═╡ e7ee1a0f-ab9b-439e-a7be-4a6d3b8f160d
begin
accum1 = 0.0
local accum1 = 0.0
for d in data
accum1 += planners_loss(d...)
end
@ -364,8 +413,8 @@ end
# ╔═╡ 74f5fde3-0593-46fc-a688-f1db7ab28c64
# Social planners problem
for epoch in 1:200
data1 = [(rand(1:500, N_constellations),rand(1:500, N_debris)) for n=1:N]
for epoch in 1:20
data1 = [(rand(1:500f0, N_constellations),rand(1:500f0, N_debris)) for n=1:N]
#train the social planner's policy funciton
Flux.Optimise.train!(planners_loss, params(policy), data1, ADAM)
@ -378,7 +427,7 @@ end
# ╔═╡ 02f3fe78-e7a7-453f-9ddf-acddf08d8676
begin
accum = 0.0
local accum = 0.0
for d in data
accum += planners_loss(d...)
@ -391,7 +440,7 @@ policy(data[3])
# ╔═╡ 14e61097-f28f-4029-b6b4-5fb119620fc3
begin
n=1
n=15
[operators[1].value(data[n])
,operators[2].value(data[n])
@ -982,12 +1031,14 @@ uuid = "3f19e933-33d8-53b3-aaab-bd5110c3b7a0"
# ╠═9fa41b7c-1923-4c1e-bfc6-20ce4a1a2ede
# ╟─90446134-4e45-471c-857d-4e165e51937a
# ╠═5b45b29e-f0f4-41e9-91e7-d444687feb4e
# ╟─152f3a3c-a565-41bb-8e59-6ab0d2315ffb
# ╟─25ac9438-2b1d-4f6b-9ff1-1695e1d52b51
# ╠═152f3a3c-a565-41bb-8e59-6ab0d2315ffb
# ╠═25ac9438-2b1d-4f6b-9ff1-1695e1d52b51
# ╠═29ff1777-d276-4e8f-8582-4ca191f2e2ff
# ╟─f7aabe43-9a2c-4fe0-8099-c29cdf66566c
# ╟─d816b252-bdca-44ba-ac5c-cb21163a1e9a
# ╠═f7aabe43-9a2c-4fe0-8099-c29cdf66566c
# ╠═d816b252-bdca-44ba-ac5c-cb21163a1e9a
# ╠═6a3b5f7a-a535-450f-8c5f-19bdcc280146
# ╠═95bfc9d8-8427-41d6-9f0f-f155296eef91
# ╠═0ba48a75-36db-4003-95c0-2329d4fb29c5
# ╠═340da189-f443-4376-a82d-7699a21ab7a2
# ╠═206ac4cc-5102-4381-ad8a-777b02dc4d5a
# ╠═eebb8706-a431-4fd1-b7a5-40f07a63d5cb
@ -1003,7 +1054,7 @@ uuid = "3f19e933-33d8-53b3-aaab-bd5110c3b7a0"
# ╠═cd55e232-493d-4849-8bd7-b0ba85e21bab
# ╠═fb6aacff-c42d-4ec1-88cb-5ce1b2e8874f
# ╠═f30904a7-5caa-449a-a5bd-f2aa78777a9a
# ╟─5abebc1a-370c-4f5f-8826-dc0b143d5166
# ╠═5abebc1a-370c-4f5f-8826-dc0b143d5166
# ╠═a20959be-65e4-4b69-9521-503bc59f0854
# ╠═6bf8d29a-7990-4e91-86e6-d9894ed3db27
# ╠═e7ee1a0f-ab9b-439e-a7be-4a6d3b8f160d

1106
julia_code/BellmanResidual_Operators.jl
View File

File diff suppressed because it is too large Load Diff
Write Preview
Loading…
Cancel
Save