renamed most files to simpler things; test and tested notebooks; remo…

…ved old dx bins methods and types; cut commented sections in notebooks;

examples/fit_choice_synthetic_data.ipynb

@@ -264,26 +264,6 @@
     "Create an instance of the `choiceoptions` composite type, using `lb`, `ub`, and `fit`. `choiceoptions` specifies which parameters to fit, and the lower and upper bounds of the optimization."
    ]
   },
-  {
-   "cell_type": "code",
-   "execution_count": 14,
-   "metadata": {},
-   "outputs": [],
-   "source": [
-    "#options = choiceoptions(fit=fit, lb=lb, ub=ub)"
-   ]
-  },
-  {
-   "cell_type": "code",
-   "execution_count": 15,
-   "metadata": {},
-   "outputs": [],
-   "source": [
-    "#options = choiceoptions(lb=vcat([0., 2.,  -5., 0.,   0.,  0., 0.005, -5.0, 0.0]),\n",
-    "#    ub = [30., 100., 5., 200., 10., 1.2,  1., 5.0, 1.0], \n",
-    "#    fit = trues(9))"
-   ]
-  },
   {
    "cell_type": "code",
    "execution_count": 16,

examples/fit_neural_real_data.ipynb

@@ -326,7 +326,7 @@
       "(numbers below include barrier contribution)\n",
       "Iter     Function value   Gradient norm \n",
       "     0     1.214329e+05     9.804548e+02\n",
-      " * time: 0.029542207717895508\n",
+      " * time: 0.015299797058105469\n",
       "\n",
       "Exiting inner optimizer with x = [0.16252272717900118, 20.03605663315885, 0.14271078200888496, 20.009599383205845, 1.0937330142867439, 0.7755103122118522, 0.9654754019793073, 1.7487100729426448, -0.5224949966918031, 1.3023422724530407, 0.6142154691698731, 1.272268011036716, 0.3360566806700428]\n",
       "Current distance to box: 0.0345246\n",
@@ -360,7 +360,7 @@
        "    |g(x)|                 = 2.98e+01 ≰ 1.0e-03\n",
        "\n",
        " * Work counters\n",
-       "    Seconds run:   28  (vs limit 170000)\n",
+       "    Seconds run:   45  (vs limit 170000)\n",
        "    Iterations:    1\n",
        "    f(x) calls:    2\n",
        "    ∇f(x) calls:   2\n",

examples/loading_and_saving.ipynb

@@ -4,28 +4,18 @@
    "cell_type": "code",
    "execution_count": 1,
    "metadata": {},
-   "outputs": [],
-   "source": [
-    "using Distributed\n",
-    "addprocs(8);"
-   ]
-  },
-  {
-   "cell_type": "code",
-   "execution_count": 2,
-   "metadata": {},
    "outputs": [
     {
      "name": "stderr",
      "output_type": "stream",
      "text": [
-      "WARNING: method definition for CIs at /Users/briandepasquale/Documents/GitHub/PulseInputDDM.jl/src/base_model.jl:13 declares type variable T but does not use it.\n",
-      "WARNING: method definition for loglikelihood at /Users/briandepasquale/Documents/GitHub/PulseInputDDM.jl/src/neural_model/RBF_model.jl:326 declares type variable T2 but does not use it.\n"
+      "WARNING: method definition for CIs at /Users/briandepasquale/Documents/GitHub/PulseInputDDM.jl/src/utils.jl:9 declares type variable T but does not use it.\n",
+      "WARNING: method definition for loglikelihood at /Users/briandepasquale/Documents/GitHub/PulseInputDDM.jl/src/neural/RBF_model.jl:326 declares type variable T2 but does not use it.\n"
      ]
     }
    ],
    "source": [
-    "@everywhere using PulseInputDDM, Flatten"
+    "using PulseInputDDM"
    ]
   },
   {
@@ -40,34 +30,25 @@
   },
   {
    "cell_type": "code",
-   "execution_count": 3,
+   "execution_count": 2,
    "metadata": {},
    "outputs": [],
    "source": [
     "θ_syn = θchoice(θz=θz(σ2_i = 1., B = 13., λ = -0.5, σ2_a = 10., σ2_s = 1.0,\n",
     "    ϕ = 0.4, τ_ϕ = 0.02), bias=0.1, lapse=0.1);"
    ]
   },
-  {
-   "cell_type": "code",
-   "execution_count": 4,
-   "metadata": {},
-   "outputs": [],
-   "source": [
-    "xgen = collect(Flatten.flatten(θ_syn));"
-   ]
-  },
   {
    "attachments": {},
    "cell_type": "markdown",
    "metadata": {},
    "source": [
-    "Generate 20K trials of the synthetic data using those parameters. change `rng` to get a different set with the same parameters. `dt` specifies the temporal binning of the data. `1e-2` has worked well."
+    "Generate 80 trials of the synthetic data using those parameters. change `rng` to get a different set with the same parameters. `dt` specifies the temporal binning of the data. `1e-2` has worked well."
    ]
   },
   {
    "cell_type": "code",
-   "execution_count": 5,
+   "execution_count": 3,
    "metadata": {},
    "outputs": [],
    "source": [
@@ -76,14 +57,14 @@
   },
   {
    "cell_type": "code",
-   "execution_count": 6,
+   "execution_count": 4,
    "metadata": {},
    "outputs": [
     {
      "data": {
       "text/plain": [
-       "choiceDDM{θchoice{θz{Float64}, Float64}}\n",
-       "  θ: θchoice{θz{Float64}, Float64}\n",
+       "choiceDDM{θchoice{θz{Float64}, Float64, Float64}}\n",
+       "  θ: θchoice{θz{Float64}, Float64, Float64}\n",
        "  n: Int64 53\n",
        "  cross: Bool false\n",
        "  fit: Array{Bool}((9,)) Bool[1, 1, 1, 1, 1, 1, 1, 1, 1]\n",
@@ -99,17 +80,6 @@
     "model = choiceDDM(θ=θ_syn)\n"
    ]
   },
-  {
-   "cell_type": "code",
-   "execution_count": 7,
-   "metadata": {},
-   "outputs": [],
-   "source": [
-    "#options = choiceoptions(lb=vcat([0., 2.,  -5., 0.,   0.,  0., 0.005, -5.0, 0.0]),\n",
-    "#    ub = [30., 100., 5., 200., 10., 1.2,  1., 5.0, 1.0], \n",
-    "#    fit = trues(9))"
-   ]
-  },
   {
    "attachments": {},
    "cell_type": "markdown",
@@ -122,7 +92,7 @@
   },
   {
    "cell_type": "code",
-   "execution_count": 8,
+   "execution_count": 5,
    "metadata": {},
    "outputs": [],
    "source": [
@@ -140,7 +110,7 @@
   },
   {
    "cell_type": "code",
-   "execution_count": 9,
+   "execution_count": 6,
    "metadata": {},
    "outputs": [],
    "source": [
@@ -157,16 +127,7 @@
   },
   {
    "cell_type": "code",
-   "execution_count": 10,
-   "metadata": {},
-   "outputs": [],
-   "source": [
-    "import PulseInputDDM: save_choice_data"
-   ]
-  },
-  {
-   "cell_type": "code",
-   "execution_count": 11,
+   "execution_count": 7,
    "metadata": {},
    "outputs": [],
    "source": [
@@ -175,7 +136,7 @@
   },
   {
    "cell_type": "code",
-   "execution_count": 12,
+   "execution_count": 8,
    "metadata": {},
    "outputs": [],
    "source": [
@@ -203,7 +164,7 @@
   },
   {
    "cell_type": "code",
-   "execution_count": 13,
+   "execution_count": 9,
    "metadata": {},
    "outputs": [],
    "source": [

src/PulseInputDDM.jl

@@ -51,28 +51,28 @@ export train_and_test, all_Softplus, save_choice_data
 export load_neural_data
 
 include("types.jl")
-include("choice_model/types.jl")
-include("neural_model/types.jl")
-include("neural-choice_model/types.jl")
+include("choice/types.jl")
+include("neural/types.jl")
+include("joint/types.jl")
 
-include("base_model.jl")
+include("core.jl")
 include("utils.jl")
-include("optim_funcs.jl")
-include("sample_model.jl")
+include("optimization.jl")
+include("sampling.jl")
 
-include("choice_model/choice_model.jl")
-include("choice_model/sample_model.jl")
-include("choice_model/process_data.jl")
-include("choice_model/IO.jl")
+include("choice/choice.jl")
+include("choice/sampling.jl")
+include("choice/preprocessing.jl")
+include("choice/IO.jl")
 
-include("neural_model/neural_model.jl")
-include("neural_model/sample_model.jl")
-include("neural_model/process_data.jl")
-include("neural_model/initalize.jl")
-include("neural_model/RBF_model.jl")
+include("neural/neural.jl")
+include("neural/sampling.jl")
+include("neural/preprocessing.jl")
+include("neural/initalize.jl")
+include("neural/RBF_model.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("joint/joint.jl")
+include("joint/sampling.jl")
+include("joint/joint-ALT.jl")
 
 end

src/base_model.jl → src/core.jl

@@ -2,31 +2,6 @@
 const dimz = 7
 
 
-"""
-    CIs(H)
-
-Given a Hessian matrix `H`, compute the 2 std confidence intervals based on the Laplace approximation.
-If `H` is not positive definite (which it should be, but might not be due numerical round off, etc.) compute
-a close approximation to it by adding a correction term. The magnitude of this correction is reported.
-
-"""
-function CIs(H::Array{Float64,2}) where T <: DDM
-
-    HPSD = Matrix(cholesky(Positive, H, Val{false}))
-
-    if !isapprox(HPSD,H)
-        norm_ϵ = norm(HPSD - H)/norm(H)
-        @warn "Hessian is not positive definite. Approximated by closest PSD matrix.
-            ||ϵ||/||H|| is $norm_ϵ"
-    end
-
-    CI = 2*sqrt.(diag(inv(HPSD)))
-
-    return CI, HPSD
-
-end
-
-
 """
     P, M, xc, dx = initialize_latent_model(σ2_i, B, λ, σ2_a, n, dt)
 
@@ -74,18 +49,6 @@ function initialize_latent_model(σ2_i::TT, B::TT, λ::TT, σ2_a::TT,
 end
 
 
-function initialize_latent_model(σ2_i::TT, B::TT, λ::TT, σ2_a::TT,
-     dx::Float64, dt::Float64) where {TT <: Any}
-
-    xc,n = bins(B,dx)
-    P = P0(σ2_i,n,dx,xc,dt)
-    M = transition_M(σ2_a*dt,λ,zero(TT),dx,xc,n,dt)
-
-    return P, M, xc, n
-
-end
-
-
 """
     P0(σ2_i, n dx, xc, dt)
 
@@ -215,33 +178,6 @@ function bins(B::TT, n::Int) where {TT}
 end
 
 
-"""
-    bins(B, dx)
-Computes the bin center locations and number of bins, given the boundary and desired (average) bin spacing.
-### Examples
-```jldoctest
-julia> xc,n = pulse_input_DDM.bins(10.,0.25)
-([-10.25, -9.75, -9.5, -9.25, -9.0, -8.75, -8.5, -8.25, -8.0, -7.75  …  7.75, 8.0, 8.25, 8.5, 8.75, 9.0, 9.25, 9.5, 9.75, 10.25], 81)
-```
-"""
-function bins(B::TT, dx::Float64) where {TT}
-
-    xc = collect(0.:dx:floor(value(B)/dx)*dx)
-
-    if xc[end] == B
-        xc = vcat(xc[1:end-1], B + dx)
-    else
-        xc = vcat(xc, 2*B - xc[end])
-    end
-
-    xc = vcat(-xc[end:-1:2], xc)
-    n = length(xc)
-
-    return xc, n
-
-end
-
-
 """
     expm1_div_x(x)
 

src/neural_model/RBF_model.jl → src/neural/RBF_model.jl

@@ -245,22 +245,6 @@ mutable struct θμ_RBF{T1} <: DDMθ
 end
 
 
-
-function train_and_test(data, options::μ_RBF_options; seed::Int=1, nRBFs = 2:10)
-
-    ntrials = length(data)
-    train = sample(Random.seed!(seed), 1:ntrials, ceil(Int, 0.9 * ntrials), replace=false)
-    test = setdiff(1:ntrials, train)
-
-    ncells = options.ncells; 
-    model = pmap(nRBF-> optimize([data[train]], μ_RBF_options(ncells=ncells, nRBFs=nRBF); show_trace=false)[1], nRBFs)   
-    testLL = map(model-> loglikelihood(model.θ, [data[test]]), model)
-
-    return nRBFs, model, testLL
-
-end
-
-
 function optimize(data, options::μ_RBF_options;
         x_tol::Float64=1e-10, f_tol::Float64=1e-6, g_tol::Float64=1e-3,
         iterations::Int=Int(2e3), show_trace::Bool=true,
@@ -301,9 +285,6 @@ function loglikelihood(x::Vector{T1}, data::Vector{Vector{T2}}, θ::θμ_RBF) wh
 end
 
 
-
-
-
 """
 """
 function θμ_RBF(x::Vector{T}, ncells::Vector{Int}, nRBFs::Int) where {T <: Real}

src/types.jl

@@ -32,12 +32,4 @@ end
     nT::Int
     nL::Vector{Int}
     nR::Vector{Int}
-end
-
-
-@with_kw struct bins
-    #clicks::T
-    xc::Vector{Real}
-    dx::Real
-    n::Int
 end