removed prior function (and saved privately). removed last case of si…

…gmoid prior in noiselss model. neural and joint tests corrected becuase they use sigmoids (didn't use in elife, or in notebooks)

src/PulseInputDDM.jl

@@ -21,7 +21,6 @@ import StatsFuns: logistic, logit, softplus, xlogy
 import Base.rand
 import Base.Iterators: partition
 import Flatten: flattenable
-#import Polynomials: Poly
 
 export choiceDDM, θchoice, θz, choiceoptions
 export neuralDDM, θneural, θy, neural_options, neuraldata
@@ -59,7 +58,6 @@ include("base_model.jl")
 include("analysis_functions.jl")
 include("optim_funcs.jl")
 include("sample_model.jl")
-include("priors.jl")
 
 include("choice_model/choice_model.jl")
 include("choice_model/sample_model.jl")
@@ -72,21 +70,8 @@ include("neural_model/process_data.jl")
 include("neural_model/noiseless_model.jl")
 include("neural_model/RBF_model.jl")
 
-#include("neural_model/neural_model-sep.jl")
-#include("neural_model/filter/filtered.jl")
-#include("neural_model/neural_model-th.jl")
-
 include("neural-choice_model/neural-choice_model.jl")
 include("neural-choice_model/sample_model.jl")
 include("neural-choice_model/neural-choice_model-ALT.jl")
 
-#include("neural-choice_model/neural-choice_model-negBin.jl")
-#include("neural-choice_model/process_data.jl")
-
-##include("neural_model/null.jl")
-##include("neural_model/polynomial/neural_poly_model.jl")
-##include("neural_model/polynomial/noiseless_model_poly.jl")
-##include("neural_model/load_and_optimize.jl")
-##include("neural_model/sample_model_functions_FP.jl")
-
 end

src/neural_model/noiseless_model.jl

@@ -92,7 +92,7 @@ Returns:
 function fit(model::noiseless_neuralDDM, data, options::neural_options_noiseless;
         x_tol::Float64=1e-10, f_tol::Float64=1e-6, g_tol::Float64=1e-3,
         iterations::Int=Int(2e3), show_trace::Bool=false,
-        outer_iterations::Int=Int(1e1), sig_σ::Float64=1.)
+        outer_iterations::Int=Int(1e1))
 
     @unpack fit, lb, ub = options
     @unpack θ = model
@@ -103,7 +103,7 @@ function fit(model::noiseless_neuralDDM, data, options::neural_options_noiseless
     lb, = unstack(lb, fit)
     ub, = unstack(ub, fit)
     x0,c = unstack(x0, fit)
-    ℓℓ(x) = -(loglikelihood(stack(x,c,fit), model, data) + sigmoid_prior(stack(x,c,fit), θ; sig_σ=sig_σ))
+    ℓℓ(x) = -loglikelihood(stack(x,c,fit), model, data)
 
     output = optimize(x0, ℓℓ, lb, ub; g_tol=g_tol, x_tol=x_tol,
         f_tol=f_tol, iterations=iterations, show_trace=show_trace,

test/joint_model_tests.jl

@@ -31,17 +31,17 @@ options = neural_choice_options(f)
 
 choice_neural_model = neural_choiceDDM(θneural_choice(vcat(x0[1:dimz], 0., 0., x0[dimz+1:end]), f), n, cross)
 
-@test round(choice_loglikelihood(choice_neural_model, data), digits=2) ≈ -0.47
+@test round(choice_loglikelihood(choice_neural_model, data), digits=2) ≈ -0.44
 
-@test round(joint_loglikelihood(choice_neural_model, data), digits=2) ≈ -368.42
+@test round(joint_loglikelihood(choice_neural_model, data), digits=2) ≈ -368.03 
 
 nparams, = PulseInputDDM.nθparams(f)
 
 options = neural_choice_options(fit=vcat(falses(dimz), trues(2), falses.(nparams)...), lb=options.lb, ub=options.ub)
 
 choice_neural_model, = choice_optimize(choice_neural_model, data, options; iterations=2, outer_iterations=1)
 
-@test round(norm(PulseInputDDM.flatten(choice_neural_model.θ)), digits=2) ≈ 56.29
+@test round(norm(PulseInputDDM.flatten(choice_neural_model.θ)), digits=2) ≈ 55.87
 
 choice_neural_model = neural_choiceDDM(θneural_choice(vcat(x0[1:dimz], 0., 0., x0[dimz+1:end]), f), n, cross)
 
@@ -50,4 +50,4 @@ options = neural_choice_options(fit=vcat(trues(dimz), trues(2), trues.(nparams).
 
 choice_neural_model, = choice_optimize(choice_neural_model, data, options; iterations=2, outer_iterations=1)
 
-@test round(norm(PulseInputDDM.flatten(choice_neural_model.θ)), digits=2) ≈ 56.29
+@test round(norm(PulseInputDDM.flatten(choice_neural_model.θ)), digits=2) ≈ 55.87

test/neural_model_tests.jl

@@ -37,19 +37,19 @@ x0 = PulseInputDDM.flatten(θ0)
 @test round(loglikelihood(x0, model0, data), digits=2) ≈ -1495.92
 
 model, = fit(model0, data, options0; iterations=2, outer_iterations=1)
-@test round(norm(PulseInputDDM.flatten(model.θ)), digits=2) ≈ 58.68
+@test round(norm(PulseInputDDM.flatten(model.θ)), digits=2) ≈ 58.28
 
-@test round(norm(gradient(model, data)), digits=2) ≈ 260.59
+@test round(norm(gradient(model, data)), digits=2) ≈ 646.89
 
 x0 = vcat([0.1, 15., -0.1, 20., 0.5, 0.8, 0.008], PulseInputDDM.flatten(model.θ)[dimz+1:end])
 options = neural_options(f)  
 
 model = neuralDDM(θneural(x0, f), n, cross)
 model, = fit(model, data, options; iterations=2, outer_iterations=1)
-@test round(norm(PulseInputDDM.flatten(model.θ)), digits=2) ≈ 55.93
+@test round(norm(PulseInputDDM.flatten(model.θ)), digits=2) ≈ 55.53
 
 H = Hessian(model, data; chunk_size=4)
-@test round(norm(H), digits=2) ≈ 15.52
+@test round(norm(H), digits=2) ≈ 8.6
 
 CI, HPSD = CIs(H)
-@test round(norm(CI), digits=2) ≈ 690.51
+@test round(norm(CI), digits=2) ≈ 1149.39