read me

efmanu · Sep 13, 2021 · 7a06663 · 7a06663
1 parent 88208e1
commit 7a06663
Show file tree

Hide file tree

Showing 6 changed files with 91 additions and 85 deletions.
diff --git a/README.md b/README.md
@@ -1,23 +1,17 @@
-# Dash Optim App
-Web app to optimize various functions with the help of Dash.jl and Optim.jl
+# Turing Dash App
+Web app for Bayesian inference using  Dash.jl and Turing.jl 
 
-NB: Only functions with 2 parameters
 
-### Run DashOptim App
+### Run TuringDash App
 From the project folder, enter the following commands in Julia REPL
 ```julia
     julia>using Pkg
     julia>Pkg.activate(".")
     julia>include("run.jl")
 ```
-##### Step 1: Start Julia with project environment
-![1](https://user-images.githubusercontent.com/22251968/132993635-4f1d52ce-07f7-404b-b442-3b98edb06f43.PNG)
-##### Step 2: Run the app
-![2](https://user-images.githubusercontent.com/22251968/132993640-8be6b1f4-1120-48b7-a422-ab123efb0b60.PNG)
-##### Step 2: Open App in browser
-![3](https://user-images.githubusercontent.com/22251968/132993642-e3b82795-1b00-4369-8115-4aad2e44f27b.PNG)
-##### Step 2: Configurations for optimization
-![Animation](https://user-images.githubusercontent.com/22251968/132994748-f696aee1-dc4c-43b4-8275-e78f1c5ce061.gif)
+
+##### Step 2: Configurations for sampling
+![Animation4](https://user-images.githubusercontent.com/22251968/133113047-30aae141-bd81-417d-bde5-832ac4de3797.gif)
 
 ### Deploy in Herokuapp
 From the project folder, enter the following commands in command line/terminal etc.

diff --git a/TuringDash/src/TuringDash.jl b/TuringDash/src/TuringDash.jl
@@ -9,6 +9,7 @@ using PlotlyJS
 using Turing
 using Random
 using CSV, DataFrames
+using Base64
 
 #Initialize channel to get data during optimization
 chn = Base.Channel{Vector{Float64}}(Inf)
@@ -18,6 +19,7 @@ y_graph1 = []
 itr = 0
 eval_f = nothing
 range_value = []
+df = nothing
 
 function make_app()
     global chn, x_graph, y_graph, itr, eval_f
@@ -28,6 +30,7 @@ function make_app()
 	y_graph = []
     y_graph1 = []
     itr = 0
+    df = nothing
 
     external_stylesheets = "https://cdnjs.cloudflare.com/ajax/libs/font-awesome/4.7.0/css/font-awesome.min.css"
     app = dash(external_stylesheets=[dbc_themes.BOOTSTRAP,external_stylesheets], 
@@ -45,34 +48,37 @@ function make_app()
         State("dtu-user-func-init", "value"),
         State("dtu-upload-datan", "filename")        
     ) do n_clicks, contents, drop_val, user_funcs, funcs_init, filename
+        global chn
+        global chn, x_graph, y_graph, y_graph1, itr, df
+        if (contents isa Nothing) && (drop_val == "0")
+            throw(PreventUpdate())
+        else
+            if contents isa Nothing
+                df = load_default(drop_val)
+            else
+                df = parse_contents(contents, filename)
+            end
+        end
         ctx = callback_context()
         if isempty(ctx.triggered)
             throw(PreventUpdate())
         end
         if !(ctx.triggered[1].prop_id == "dtu-btn-turing.n_clicks")
             throw(PreventUpdate())
         end
-        if (contents isa Nothing) && (drop_val == "0")
-            throw(PreventUpdate())
-        end
+
         if (user_funcs == "") || (funcs_init == "")
             throw(PreventUpdate())
         end
-        global chn
-        global chn, x_graph, y_graph, y_graph1, itr
+
         x_graph = []
         y_graph = []
         y_graph1 = []
         itr = 0
-        if contents isa Nothing
-            df = load_default(drop_val)
-        else
-            df = parse_contents(contents, filename)
-        end        
+
         for colname in Symbol.(names(df))
             @eval $colname = df.$colname
-        end         
-
+        end        
         #evealuate model
         eval(Meta.parse(user_funcs))
 
@@ -85,15 +91,15 @@ function make_app()
         nperiteration = 5
         nsamples = 1000
         @async begin
-            r = sample(rng, model, spl, nperiteration; chain_type=MCMCChains.Chains, save_state=true)
+            r = sample(rng, model, alg, nperiteration; chain_type=MCMCChains.Chains, save_state=true, progress=false)
             put!(chn, Array(r)[end,1:2])
             for i in (nperiteration + 1):nperiteration:nsamples
-                r = Turing.Inference.resume(r, nperiteration, save_state=true)
+                r = Turing.Inference.resume(r, nperiteration, save_state=true, progress=false)
                 put!(chn, Array(r)[end,1:2])
             end
         end
-        @show nsamples
-        return "render_liveupdates()"
+        # @show nsamples
+        return render_liveupdates()
     end
 
     callback!(app,
@@ -108,31 +114,32 @@ function make_app()
         global x_graph, y_graph, y_graph1, chn, itr
         if isready(chn)
             val = take!(chn)
-            itr += 1
+            itr += 5
             append!(x_graph, itr)
             append!(y_graph, val[1])
 			append!(y_graph1, val[2])
         else
             st = true
         end
-        return Dict(
-            "data" => [
-                    Dict(
-                        "x" => x_graph,
-                        "y" => y_graph,
-                        "mode" => "line",
-                    )
-                ]
-            ),
-            Dict(
-                "data" => [
-                    Dict(
-                        "x" => x_graph,
-                        "y" => y_graph1,
-                        "mode" => "line",
-                    )
-                ]
-            ), st
+        scatter_plt1 = scatter(
+            ;x=x_graph, y=y_graph, 
+            mode="lines", line_color="blue"
+        )
+        scatter_plt2 = scatter(
+            ;x=x_graph, y=y_graph1, 
+            mode="lines", line_color="blue"
+        )
+        fig1 = PlotlyJS.plot([scatter_plt1], Layout(
+            title="Trace Plot: Var 1",
+            xaxis_title="Iteration",
+            yaxis_title="Value",
+        ))
+        fig2 = PlotlyJS.plot([scatter_plt2], Layout(
+            title="Trace Plot: Var 2",
+            xaxis_title="Iteration",
+            yaxis_title="Value",
+        ))
+        return fig1, fig2, st
     end
     return app
 end

diff --git a/TuringDash/src/render_funcs.jl b/TuringDash/src/render_funcs.jl
@@ -88,9 +88,9 @@ function render_uploader()
                             id="dtu-def-data-drop",
                             options =[
                                 Dict("label" => "Select Data", "value" => "0"),
-                                Dict("label" => "Real Estate", "value" => "1"),
-                                Dict("label" => "Solar", "value" => "2"),
-                                Dict("label" => "House Price", "value" => "3")
+                                Dict("label" => "Data 1", "value" => "1"),
+                                Dict("label" => "Data 2", "value" => "2"),
+                                Dict("label" => "Data 3", "value" => "3")
                             ],
                             value="0"
                             )
@@ -128,11 +128,11 @@ function render_liveupdates()
         html_h3("Plots"),
         dcc_interval(
             id="interval-component",
-            interval=1000, # in milliseconds
+            interval=500, # in milliseconds
             n_intervals=0,
             disabled=false
         ),
-        dcc_graph(id="dtu-params-state2",
+        dcc_graph(id="dtu-params-state1",
             figure=Dict(
                 "data" => [
                     Dict("x" => x_graph, "y" => y_graph),

diff --git a/TuringDash/src/utils.jl b/TuringDash/src/utils.jl
@@ -31,11 +31,11 @@ function parse_contents(contents, filename)
 
   function load_default(val)
     if val == "1"
-      df = CSV.read(download("https://raw.githubusercontent.com/efmanu/fluxdash/master/FluxDash/datasets/real_estate.csv"), DataFrame)
+      df = CSV.read(download("https://raw.githubusercontent.com/efmanu/TuringDashApp.jl/master/TuringDash/datasets/data1.csv"), DataFrame)
     elseif val =="2"
-      df = CSV.read(download("https://raw.githubusercontent.com/efmanu/fluxdash/master/FluxDash/datasets/real_estate.csv"), DataFrame)
+      df = CSV.read(download("https://raw.githubusercontent.com/efmanu/TuringDashApp.jl/master/TuringDash/datasets/data2.csv"), DataFrame)
     else
-      df = CSV.read(download("https://raw.githubusercontent.com/efmanu/fluxdash/master/FluxDash/datasets/real_estate.csv"), DataFrame)
+      df = CSV.read(download("https://raw.githubusercontent.com/efmanu/TuringDashApp.jl/master/TuringDash/datasets/data3.csv"), DataFrame)
     end
     return df
   end
diff --git a/precompile_funcs.jl b/precompile_funcs.jl
@@ -1,38 +1,43 @@
 module PrecompileFuncs
-    using OptimDash
-    using OptimDash.Optim
-    using OptimDash.PlotlyJS
+    using TuringDash
+    using TuringDash.Turing
+    using TuringDash.Random
+    using TuringDash.CSV, TuringDash.DataFrames
 
-function precompile_optim()
+function precompile_turing()
     chn = Base.Channel{Vector{Float64}}(Inf)
-    user_funcs = "f(x) = (1.0 - x[1])^2 + 100.0 * (x[2] - x[1]^2)^2"
-    funcs_init = "[0.0, 0.0]"
-    lb_str = "[-2.0, -2.0]"
-    ub_str = "[2.0, 2.0]"
-    eval_f, init_f, lb_f, ub_f = OptimDash.eval_definition(
-        user_funcs, funcs_init, lb_str, ub_str
-    )
-    function g(f)
-        function _g(x)
-            put!(chn, x)
-            return Base.invokelatest(f, x)
+    df1 = nothing
+    global df1
+    df1 = CSV.read(download("https://raw.githubusercontent.com/efmanu/TuringDashApp.jl/master/TuringDash/datasets/data1.csv"), DataFrame)
+    model_str1 = """
+    @model turingmodel(x, y) = begin
+        a ~ Normal()
+        b ~ Normal()
+        for i in 1:length(x)
+            y[i] ~ Normal(a + b * x[i], 1.0)
         end
-        return _g
     end
-    results = optimize(g(eval_f), init_f)
+    """
+    model_str2 = "turingmodel(x, y)"
+    for colname in Symbol.(names(df1))
+        @eval $colname = df1.$colname
+    end
 
-    bound_range = (ub_f - lb_f)./1000
-    range_value = [collect(1:10) for _ in 1:2]
-    z =  rand(1:10,10,10)
-    heat_plt = heatmap(
-        x = range_value[1], y = range_value[2], 
-        z = z
-    )
-    # scatter_plt = scatter(
-    #     ;x=x_graph, y=y_graph, 
-    #     mode="lines", line_color="black"
-    # )
-    fig = OptimDash.PlotlyJS.plot(heat_plt)
+    eval(Meta.parse(model_str1))
+    model = eval(Meta.parse(model_str2))
+
+    alg = MH()
+    rng = Random.GLOBAL_RNG
+
+    nperiteration = 50
+    nsamples = 20
+
+    r = sample(rng, model, alg, nperiteration; chain_type=MCMCChains.Chains, save_state=true, progress=false)
+    put!(chn, Array(r)[end,1:2])
+    for i in (nperiteration + 1):nperiteration:nsamples
+        r = Turing.Inference.resume(r, nperiteration, save_state=true, progress=false)
+        put!(chn, Array(r)[end,1:2])
+    end
     return "success"
 end
 end
diff --git a/run.jl b/run.jl
@@ -3,7 +3,7 @@
 using TuringDash
 
 # using .PrecompileFuncs
-# PrecompileFuncs.precompile_optim()
+# PrecompileFuncs.precompile_turing()
 
 app = TuringDash.make_app()