Merge pull request #58 from msainsburydale/auto-juliaformatter-pr

Automatic JuliaFormatter.jl run

Author: msainsburydale

src/Estimators.jl

@@ -492,20 +492,17 @@ function PosteriorEstimator(network, d::Integer, dstar::Integer = d; q = Normali
     end
 
     # Distribution used to approximate the posterior 
-    q = q(d, dstar; kwargs...) 
+    q = q(d, dstar; kwargs...)
 
     # Initialise the estimator
     return PosteriorEstimator(q, network)
 end
 
 # Constructor for consistent argument ordering
-function PosteriorEstimator(network, q::A) where A <: ApproximateDistribution
+function PosteriorEstimator(network, q::A) where {A <: ApproximateDistribution}
     return PosteriorEstimator(q, network)
 end
 
-
-
-
 #TODO maybe its better to not have a tuple, and just allow the arguments to be passed as normal... Just have to change DeepSet definition to allow two arguments in some places (this is more natural). Can easily allow backwards compat in this case too. 
 @doc raw"""
 	RatioEstimator <: NeuralEstimator
@@ -615,8 +612,6 @@ end
 # Z = simulate(θ, m)
 # r̂(Z, θ)                                   # log of the likelihood-to-evidence ratios
 
-
-
 @doc raw"""
 	PiecewiseEstimator <: NeuralEstimator
 	PiecewiseEstimator(estimators::Vector{BayesEstimator}, changepoints::Vector{Integer})

src/assess.jl

@@ -269,7 +269,7 @@ function assess(
     # Add estimator name if it was provided
     if !isnothing(estimator_names) # deprecation coercion
         estimator_name = estimator_names
-    end 
+    end
     if !isnothing(estimator_name)
         θ̂[!, "estimator"] .= estimator_name
         runtime[!, "estimator"] .= estimator_name
@@ -296,7 +296,7 @@ function assess(
         df_s = DataFrame(
             parameter = repeat(parameter_names, inner = N),
             truth = repeat(θ[:, idx], inner = N),
-            draw = repeat(1:N, outer = d), 
+            draw = repeat(1:N, outer = d),
             value = vec(S')
         )
         df_s[!, "k"] .= ((idx - 1) % K) + 1

src/inference.jl

@@ -139,11 +139,10 @@ function sampleposterior(
     est::RatioEstimator,
     Z,
     N::Integer = 1000;
-    logprior::Function = θ -> 0f0,
+    logprior::Function = θ -> 0.0f0,
     θ_grid = nothing, theta_grid = nothing,
     # θ₀ = nothing, theta0 = nothing, # NB a basic MCMC sampler could be initialised with θ₀
     kwargs...)
-
     Z = f32(Z)
 
     # Check duplicated arguments that are needed so that the R interface uses ASCII characters only
@@ -168,7 +167,6 @@ function sampleposterior(
     end
 end
 
-
 function sampleposterior(estimator::PosteriorEstimator, Z, N::Integer = 1000; kwargs...)
     # Determine if we are using the gpu 
     args = (; kwargs...)
@@ -207,7 +205,7 @@ See also [`posteriormedian()`](@ref), [`posteriormean()`](@ref).
 """
 function posteriormode(
     est::RatioEstimator, Z;
-    logprior::Function = θ -> 0f0, penalty::Union{Function, Nothing} = nothing,
+    logprior::Function = θ -> 0.0f0, penalty::Union{Function, Nothing} = nothing,
     θ_grid = nothing, theta_grid = nothing,
     θ₀ = nothing, theta0 = nothing,
     kwargs...

src/train.jl

@@ -107,8 +107,8 @@ function findlr(opt)
 end
 
 function _train(
-    estimator, 
-    sampler, 
+    estimator,
+    sampler,
     simulator;
     m = nothing,
     ξ = nothing, xi = nothing,
@@ -190,7 +190,6 @@ function _train(
     local min_val_risk = val_risk # minimum validation loss, monitored for early stopping
     local early_stopping_counter = 0
     train_time = @elapsed for epoch ∈ 1:epochs
-
         GC.gc(false)
 
         if store_entire_train_set
@@ -322,7 +321,6 @@ function _train(estimator, θ_train::P, θ_val::P, simulator;
     local min_val_risk = val_risk
     local early_stopping_counter = 0
     train_time = @elapsed for epoch = 1:epochs
-
         GC.gc(false)
 
         sim_time = 0.0
@@ -441,7 +439,6 @@ function _train(estimator, θ_train::P, θ_val::P, Z_train::T, Z_val::T;
     local min_val_risk = val_risk
     local early_stopping_counter = 0
     train_time = @elapsed for epoch = 1:epochs
-
         GC.gc(false)
 
         # For each batch update estimator and compute the training loss

src/utility.jl

@@ -289,7 +289,6 @@ function _checkgpu(use_gpu; verbose::Bool = true)
     return (device)
 end
 
-
 # Here, we define _manualgc() for the case that CUDA has not been loaded (so, we will be using the CPU)
 # For the case that CUDA is loaded, the function is overloaded in ext/NeuralEstimatorsCUDAExt.jl
 # NB Julia complains if we overload functions in package extensions... to get around this, here we