GardenOptim/src/GardenOptim.jl
2020-02-20 22:10:46 +01:00

135 lines
3.7 KiB
Julia

module GardenOptim
using DocStringExtensions
using Logging
using CSV
using Tables
export loadplants, loadgarden, loadcosts, update!, randomgardenevolution!, outputgarden
@template (FUNCTIONS, METHODS, MACROS) =
"""
$(TYPEDSIGNATURES)
$(DOCSTRING)
"""
function loadplants()::Vector{String}
plants = readlines("data/plants.txt")
@info "loaded $(length(plants)) plants"
plants
end
function loadgarden(plants::Vector{String})::Tuple{Matrix{Int}, Matrix{Bool}}
garden = CSV.read("data/garden.csv")
garden = coalesce.(garden, "")
mask = convert(Matrix, garden .== "empty")
garden = indexin(convert(Matrix, garden), plants)
garden = replace(garden, nothing=>0)
@assert size(garden) == size(mask)
@info "loaded garden of size $(size(garden))"
garden, mask
end
function loadcosts()::Matrix{Float64}
df = CSV.read("data/costs.csv")
df = coalesce.(df, 0) # replace missing values by 0
costs = convert(Matrix, df[:, 2:end])
@info "loaded cost matrix of size $(size(costs))"
# ensure the matrix is symmetric: keep the max of itself and its transpose
costs = Float64.(max.(costs, permutedims(costs)))
end
"Return a random index to be filled from the garden mask."
function randomindex(mask::Matrix{Bool})::Int
while true
i = rand(1:length(mask))
if mask[i]
return i
end
end
end
"Swap to the elements corresponding to the two provided indices."
function swap!(garden::Matrix{Int}, i::Int, j::Int)
t = garden[i]
garden[i] = garden[j]
garden[j] = t
garden
end
"Return the neighbours to be filled of the cell at the given index."
function neighbours(garden::Matrix{Int}, idx::Int)::Vector{Int}
m, n = size(garden)
j, i = divrem(idx - 1, m)
i += 1
j += 1
neighbourindices = [(i, j-1), (i, j+1), (i-1, j), (i+1, j)]
# cells filled with 0 are not part of the garden
[
garden[k, l] for (k, l) in neighbourindices
if 0 < k <= m && 0 < l <= n && garden[k, l] != 0
]
end
"Compute the cost difference when swapping the two provided indices."
function deltacost(garden::Matrix{Int}, costs::Matrix{Float64}, i::Int, j::Int)::Float64
cost = 0
for k in neighbours(garden, i)
cost += costs[k, garden[j]] - costs[k, garden[i]]
end
for k in neighbours(garden, j)
cost += costs[k, garden[i]] - costs[k, garden[j]]
end
cost
end
"Update the garden using Metropolis-Hastings, using the inverse temperature beta."
function update!(
garden::Matrix{Int},
mask::Matrix{Bool},
costs::Matrix{Float64},
beta::Float64 = 10.0
)
N = length(garden)
i = randomindex(mask)
j = randomindex(mask)
while i == j
j = randomindex(mask)
end
d = deltacost(garden, costs, i, j)
@debug "cost difference $d"
if rand() < exp(- beta * d)
@debug "swapping indices $i and $j"
return swap!(garden, i, j)
end
garden
end
"Fill the garden randomly with a predefined number of plants."
function randomfillgarden!(garden::Matrix{Int}, mask::Matrix{Bool}, plantcount::Int)
garden[mask] = rand(1:plantcount, sum(mask))
garden
end
"Update the garden for a given number of steps, starting from a random initialisation."
function randomgardenevolution!(
garden::Matrix{Int},
mask::Matrix{Bool},
costs::Matrix{Float64};
steps::Int = 10000
)
m = size(costs, 1)
garden = randomfillgarden!(garden, mask, m)
for i = 1:steps
update!(garden, mask, costs, 10.0)
end
garden
end
"Save the garden to a CSV file."
function outputgarden(garden::Matrix{Int}, plants::Vector{String})
output = vcat([""], plants)[garden .+ 1]
CSV.write("output.csv", Tables.table(output), writeheader=false)
end
end # module