Merge branch 'filesize-per-step' into 'master'

Filesize per step

See merge request !5

Project.toml

@@ -8,6 +8,8 @@ Glob = "c27321d9-0574-5035-807b-f59d2c89b15c"
 DataFrames = "a93c6f00-e57d-5684-b7b6-d8193f3e46c0"
 GLM = "38e38edf-8417-5370-95a0-9cbb8c7f171a"
 Plots = "91a5bcdd-55d7-5caf-9e0b-520d859cae80"
+CSV = "336ed68f-0bac-5ca0-87d4-7b16caf5d00b"
+DataStructures = "864edb3b-99cc-5e75-8d2d-829cb0a9cfe8"
 
 [compat]
 julia = "1"
@@ -20,6 +22,3 @@ JuliaFormatter = "98e50ef6-434e-11e9-1051-2b60c6c9e899"
 [targets]
 test = ["Test"]
 
-
-
-

examples/ExampleAmrexCastro.jl

@@ -6,17 +6,20 @@ import Exio
 function main()
     #### Configuration section ######
     ## exio_init requires three inputs and returns a handler of type exioH  
-  
+
     # only supported: AmrexCastro
     app::String = "AmrexCastro"
-    
+
     # location where several run outputs are located they must have a common prefix 
-    outputPrefix::String = "/home/wgodoy/workspace/Castro/Exec/hydro_tests/Sedov/case"
-    
+    # Local
+    # outputPrefix::String = "/home/wgodoy/workspace/Castro/Exec/hydro_tests/Sedov/case"
+    # Summit location
+    outputPrefix::String = "/gpfs/alpine/proj-shared/csc310/wgodoy/amrex/Castro/Exec/hydro_tests/Sedov/case"
+
     # file name inside outputs directories
     logFile::String = "run.log"
     #################################
-    
+
     # init Exio (can create as many entities as needed)
     exio = Exio.exio_init(app, outputPrefix, logFile)
     # create a linear model based on provided runs (the larger, the better)

plot_size.csv

+plots_cells,plots_size,plots_size_L0,plots_size_L1,plots_size_L2,caseID,caseDir
+4096,472612.0,461951.0,0.0,0.0,case12,plt_3d_00000
+8192,945273.0,923914.0,0.0,0.0,case12,plt_3d_00002
+12288,1.417935e6,1.385877e6,0.0,0.0,case12,plt_3d_00004
+16384,1.890595e6,1.84784e6,0.0,0.0,case12,plt_3d_00006
+20480,2.363255e6,2.309803e6,0.0,0.0,case12,plt_3d_00008
+24576,2.835917e6,2.771766e6,0.0,0.0,case12,plt_3d_00010

scripts/requirements.jl

@@ -11,9 +11,13 @@ Pkg.add("DataFrames")
 Pkg.add("GLM")
 Pkg.add("Plots")
 Pkg.add("PackageCompiler")
+Pkg.add("Flux")
+Pkg.add("CSV")
+Pkg.add("DataStructures")
 
 
 using PackageCompiler
-create_sysimage([:Glob, :Plots, :GLM, :DataFrames], sysimage_path = "jexio_deps.so")
+create_sysimage([:Glob, :Plots, :GLM, :DataFrames, :Flux, :CSV, :DataStructures], 
+                 sysimage_path = "jexio_deps.so")
 
 exit()

src/extractor/Amrex.jl

@@ -11,7 +11,14 @@ mutable struct Amrex <: AbstractAmrex
     Amrex(outputPrefix::String, runlogFile::String) = new(
         "Amrex",
         outputPrefix,
-        ["max_step", "amr.check_int", "amr.plot_int", "amr.n_cell", "amr.max_level", "caseID"],
+        [
+            "max_step",
+            "amr.check_int",
+            "amr.plot_int",
+            "amr.n_cell",
+            "amr.max_level",
+            "caseID",
+        ],
         ["plots_size", "checkpoints_size"],
         runlogFile,
     )
@@ -109,8 +116,8 @@ function _get_independent_variables(extractor::AbstractAmrex, parameters)::Dict{
     maxLevel = get(parameters, "amr.max_level", "")
     independentVariables["amr.max_level"] = maxLevel == "" ? 1 : parse(Int32, maxLevel)
 
-    regrid = get(parameters, "amr.regrid_int", "")
-    independentVariables["amr.regrid_int"] = regrid == "" ? 1 : parse(Int32, regrid)
+    regrid = prod(parse.(Int64, split(get(parameters, "amr.regrid_int", ["0"]))))
+    independentVariables["amr.regrid_int"] = regrid
 
     return independentVariables
 

src/extractor/AmrexCastro.jl

@@ -3,6 +3,9 @@ include("../helper/helperSystem.jl")
 
 import DataFrames
 import GLM
+import Plots
+import CSV
+import DataStructures
 
 struct AmrexCastro <: AbstractAmrex
     app::String
@@ -44,7 +47,8 @@ function run_linear_models(extractor::AmrexCastro)
 
     for key in keys(extractor.outputs)
         if (key == "plots_size")
-            _run_linear_model_plots_size(extractor, X)
+            #_run_linear_model_plots_size(extractor, X)
+            _run_linear_models_plots_size_step(extractor, X)
         end
     end
 end
@@ -89,7 +93,7 @@ function _get_linear_model_X(extractor::AmrexCastro)::DataFrames.DataFrame
         # get outputDir name
         independentVariables["caseID"] = helper_get_relative_path(outputDir)
 
-        if !isXInit
+        if isXInit == false
             X = DataFrames.DataFrame(independentVariables)
             isXInit = true
         else
@@ -102,63 +106,213 @@ function _get_linear_model_X(extractor::AmrexCastro)::DataFrames.DataFrame
 end
 
 
-function _run_linear_model_plots_size(extractor::AmrexCastro, X::DataFrames.DataFrame)
-
-    outputDirs = helper_get_prefix_directories(extractor.outputPrefix)
-
-    # single column DataFrame Y in the linear model
-    plotsSizesData = DataFrames.DataFrame(plots_sizes = Int64[])
+function _run_linear_models_plots_size_step(extractor::AmrexCastro, X::DataFrames.DataFrame) 
 
+  function _push_datasize!(directory::String, data::Array{Int64,1})
+    if size(data,1) == 0
+      push!(data, helper_get_directory_size(directory) )
+    else
+      push!(data, last(data) + helper_get_directory_size(directory) )
+    end
+  end
+  
+  function _find_max_nlevels(outputDirs::Array{String})::Int64
+    
+    max_nlevels::Int64 = 0
+    
     for outputDir in outputDirs
+      inputFile::String = _get_input_file(extractor, outputDir)
+      parameters = _input_parser(extractor, inputFile)
+      rootPlotName = get(parameters, "amr.plot_file", "")
 
-        # TODO refactor this later
-        inputFile::String = _get_input_file(extractor, outputDir)
-        parameters = _input_parser(extractor, inputFile)
-        rootPlotName = get(parameters, "amr.plot_file", "")
-
-        # find all directories with rootPlotName and get its size
-        plotFileDirs = helper_get_prefix_directories(string(outputDir, "/", rootPlotName))
-
-        plotsSize::Int64 = 0
-        for plotFileDir in plotFileDirs
-            sizeDir::Int64 = helper_get_directory_size(plotFileDir)
-            plotsSize += sizeDir
+      # find all directories with rootPlotName and get its size
+      # root of _plt directories 
+      plotFileDirs = helper_get_prefix_directories(string(outputDir, "/", rootPlotName))
+      
+      for plotFileDir in plotFileDirs
+        levelDirs = helper_get_prefix_directories(string(plotFileDir,"/Level_"))
+        nlevels = size(levelDirs,1)  
+      
+        if max_nlevels < nlevels
+          max_nlevels = nlevels
         end
-
-        push!(plotsSizesData, [plotsSize])
+      end
     end
+    
+    return max_nlevels
+  end
+  
+  
+  function _find_max_nranks(outputDirs::Array{String})::Int64
+    
+    max_nranks::Int64 = 0
+    for outputDir in outputDirs
+      
+      inputFile::String = _get_input_file(extractor, outputDir)
+      parameters = _input_parser(extractor, inputFile)
+      rootPlotName = get(parameters, "amr.plot_file", "")
 
-    println(plotsSizesData)
+      # find all directories with rootPlotName and get its size
+      # root of _plt directories 
+      plotFileDirs = helper_get_prefix_directories(string(outputDir, "/", rootPlotName))
+      
+      for plotFileDir in plotFileDirs
+      
+        levelDirs = helper_get_prefix_directories(string(plotFileDir,"/Level_"))
+      
+        for levelDir in levelDirs
+          rankDirs = helper_get_prefix_directories(string(levelDir,"/Cell_D_"))
+          nranks = size(rankDirs,1)
+        
+          if max_nranks < nranks
+            max_nranks = nranks
+          end
+        end
+      end
+    end  
+    return max_nranks
+  end
+    
+  # output directories
+  outputDirs = helper_get_prefix_directories(extractor.outputPrefix)
+  # input with total_cells * output_number
+  Xd::Array{Int64,1} = []
+  # output cumulative data size, per timestep, level, rank
+  Yd_Timesteps::Array{Int64,1} = []
+
+  Yd_Levels::DataStructures.SortedDict{ String,Array{Int64,1} } = Dict()
+  Yd_Ranks::DataStructures.SortedDict{ String,Array{Int64,1} } = Dict()
+  
+  nlevels::Int64 =  _find_max_nlevels(outputDirs)
+  nranks::Int64 = _find_max_nranks(outputDirs)
+  
+  for level in (1:nlevels)   
+    levelKey::String = string("Level_",level-1)
+    push!(Yd_Levels, levelKey => [])
+    println("Yd_Levels", Yd_Levels, " ", level)
     
-    # Prepare the X independent variables in the linear model
-    XNames::Array{String} = extractor.outputs["plots_size"]
+    for rank in (1:nranks)
+      rankKey::String = string(levelKey,"/Cell_D_", lpad(rank-1,5,"0") )
+      push!(Yd_Ranks, rankKey => [])
+    end
+    
+  end
+  
+  println("Yd_Ranks: ", Yd_Ranks)
+    
+  caseIDs::Array{String,1} = []
+  caseDirs::Array{String,1} = []
       
-   col1 = Symbol("amr.nplot_files")
-   col2 = Symbol("amr.ncells")
-   
-   insert!(plotsSizesData, 2 ,X[!,col1] .* X[!,col2], :new_data) 
-   #plotsSizesData[!,col1] = 
+  for (index,outputDir) in enumerate(outputDirs)
     
-                                    
-
-    #for XName in XNames
-        #columnName = Symbol(XName)
-        # this syntax [!, symbol] does not make a copy, use [:, symbol] for copies
-        #plotsSizesData[!, columnName] = X[!, columnName]
-    #end
-
-    # formula from https://discourse.julialang.org/t/glm-jl-with-unknown-column-names/20692/5
-    response = Symbol(names(plotsSizesData)[1])
-    predictors = Symbol.(names(plotsSizesData)[2:end])
-
-    println(plotsSizesData)
-    f = @eval(GLM.@formula($response ~ (+)( $(predictors...))))
+    inputFile::String = _get_input_file(extractor, outputDir)
+    parameters = _input_parser(extractor, inputFile)
+    rootPlotName = get(parameters, "amr.plot_file", "")
 
+    # find all directories with rootPlotName and get its size
+    # root of _plt directories 
+    plotFileDirs = helper_get_prefix_directories(string(outputDir, "/", rootPlotName))
 
-    ols = GLM.lm(f, plotsSizesData)
-    println("I/O linear model formula:")
-    println(ols)
-
-    #import Plots
-    #display(Plots.plot(X, Y))
+    counter::Int32 = 1
+    
+    # extract ncells
+    ncells = X[!,"amr.ncells"][index]
+    
+    # caseID
+    caseID = helper_get_relative_path(outputDir)
+    
+    for (plotIndex,plotFileDir) in enumerate(plotFileDirs)
+      # Xd
+      push!( Xd, counter*ncells)
+      
+      # Yd timesteps cumulative
+      if plotIndex == 1
+         push!(Yd_Timesteps, helper_get_directory_size(plotFileDir) )
+      else
+         push!(Yd_Timesteps, last(Yd_Timesteps) + helper_get_directory_size(plotFileDir) )
+      end
+      
+      levelDirs = helper_get_prefix_directories(string(plotFileDir,"/Level_"))
+      currentLevels = helper_get_relative_path.(levelDirs)
+      
+      #println("Current levels: ", currentLevels)
+      
+      # Levels
+      for (levelKey,levelSizeValues) in Yd_Levels
+
+        levelDir = string(plotFileDir,"/",levelKey)
+        foundArray = findall( x -> x == levelKey, currentLevels)
+        
+        if size(foundArray,1) == 1
+           # bug _push_datasize!(levelDir, Yd_Levels[levelKey])
+           if plotIndex == 1
+             push!( Yd_Levels[levelKey], helper_get_directory_size(levelDir) )
+           else
+             push!( Yd_Levels[levelKey], last(Yd_Levels[levelKey]) + helper_get_directory_size(levelDir) )
+           end
+        else
+           push!(Yd_Levels[levelKey],0)
+        end
+        
+        # Level Rank data
+        # add ranks 
+        rankFullFiles = helper_get_prefix_directories(string(levelDir,"/Cell_D_"))
+        rankFiles = helper_get_relative_path.(rankFullFiles)
+           
+        for rankID in 1:nranks
+            
+            rankKey::String = string(levelKey,"/Cell_D_", lpad(rankID-1,5,"0") )
+            # check if file exist
+            checkFileName =  string(levelDir,"/Cell_D_", lpad(rankID-1,5,"0") )
+            
+            if isfile( checkFileName )
+              rankFileSize = filesize(checkFileName)
+              # bug push!(Yd_Ranks[rankKey], rankFileSize)
+              
+              if plotIndex == 1
+                push!(Yd_Ranks[rankKey], rankFileSize)  
+                # push!( Yd_Levels[levelKey], helper_get_directory_size(levelDir) )
+              else
+                push!( Yd_Ranks[rankKey], last(Yd_Ranks[rankKey]) + rankFileSize )
+              end
+              
+            else
+              push!(Yd_Ranks[rankKey],0)  
+            end
+        end
+        
+      end
+        
+      # caseID
+      push!(caseIDs, caseID )
+      
+      # caseDir
+      push!(caseDirs, helper_get_relative_path(plotFileDir) )
+      
+      counter = counter + 1
+    end
+    
+  end
+
+  df::DataFrames.DataFrame = 
+      DataFrames.DataFrame( plots_cells = Xd, 
+                            plots_size = Yd_Timesteps)
+                            
+  for (key,Yd_Level) in Yd_Levels
+    df[!,key] = Yd_Level
+  end
+  
+  for (key,Yd_Rank) in Yd_Ranks
+    df[!,key] = Yd_Rank
+  end
+  
+  df[!,"caseID"] = caseIDs
+  df[!,"caseDir"] = caseDirs
+
+  #println(df)
+  
+  CSV.write("plot_size.csv",  df, header=true)
+  #display(Plots.scatter(Xd, Yd))
+  #readline()
 end
+

src/helper/helperFlux.jl

+
+
+import Flux
+import Statistics
+
+using Plots;
+pyplot();
+using LaTeXStrings
+
+
+function FluxTrain(nepocs::Int32, X::Array{Float64,1}, Y::Array{Float64,1})
+
+    W = [500.0]
+    b = [10000.0]
+
+    model = Flux.Dense(W, b)
+    parameters = Flux.params(model)
+    println("Initial parameter: ", parameters)
+
+    # set up data
+    Xd = reduce(hcat, X)
+    Yd = reduce(hcat, Y)
+    data = [(Xd, Yd)]
+    println("Data", data)
+
+    # optimizer = Flux.Descent(1)
+    # optimizer = Flux.ADAM()
+    optimizer = Flux.ADAM(1, (0.99, 0.999))
+
+    loss(x, y) = Statistics.mean((model(x) .- y) .^ 2)
+
+    Yd_0 = model(Xd)
+    println("Initial solution: ", Yd_0)
+
+    plot(X, Y, st = :scatter, label = L"y")
+    Yd_nE = model(Xd)
+    for iter = 1:nepocs
+        Flux.train!(loss, parameters, data, optimizer)
+    end
+
+    println("Module: ", model)
+    println("Parameters_nE: ", parameters)
+    println("Final solution: ", Yd_nE)
+    Yd_nE = model(Xd)
+
+
+    plot!(Xd', Yd_0', lc = :green, label = L"y_0")
+    plot!(Xd', Yd_nE', lc = :red, label = L"y_{n_\mathrm{E}}")
+    plot!(xlabel = L"x", ylabel = L"y", title = "Data for linear model")
+    
+
+end
+
+
+function test()
+
+    x::Array{Float64,1} = [256, 4096, 512, 512, 8192]
+    y::Array{Float64,1} = [435867, 2959963, 1475489, 1485569, 4518030]
+
+    nepocs::Int32 = 100000
+
+    FluxTrain(nepocs, x, y)
+
+end
+
+
+test()

src/helper/helperPlot.jl

+
+import CSV
+import DataFrames
+using Plots
+
+function plot_CSV_3col( filename::String )
+  
+  df = CSV.File(filename) |> DataFrames.DataFrame
+  
+  x = df[!,"plots_cells"]
+  y = df[!,"plots_size"]
+    
+  z = df[!,"caseID"]
+  
+  fig1 = Plots.scatter( x, y, groups = z, xaxis=:log, yaxis=:log, 
+                      legendfontsize=7,
+                      legend = :outertopleft, 
+                      # [shapes], size, alpha
+                      marker = ([:x :+ :star4 :vline :square :circle], 7, 1.0),
+                      markercolor = [:blue :green :red :orange :pink],
+                      xlabel = "cumulative output ncells ( output_counter x ncells )", 
+                      ylabel = "cumulative output data size (bytes)", 
+                      title = "Amrex Castro hydro_test Sedov 2d.cyl_in_cartcoords output size",
+                      size=(1200,850),
+                      reuse = false
+                      ) 
+  Plots.gui(fig1)
+  #Plots.png(fig1)
+  readline()
+
+  yL0 = df[!,3]
+  yL1 = df[!,4]
+  yL2 = df[!,5]  
+  fig2 = Plots.scatter( x, yL0, groups = z, label="L0",
+                          marker = ([:x], 10, 1.0),
+                          markercolor = [:blue :green :red :orange :pink],
+                          legendfontsize=9,
+                          legend = :outertopleft,
+                          xlabel = "cumulative output ncells ( output_counter x ncells )", 
+                          ylabel = "cumulative output data size (bytes)", 
+                          title = "Amrex Castro hydro_test Sedov 2d.cyl_in_cartcoords output size per Level",
+                          size=(1200,850),
+                          reuse = true
+                          )
+  Plots.scatter!( x, yL1, groups = z, label= "L1", 
+                          marker = ([:+], 10, 1.0),
+                          markercolor = [:blue :green :red :orange :pink],
+                           )
+                            
+  Plots.scatter!( x, yL2, groups = z, label="L2", 
+                          marker = ([:star4], 10, 1.0),
+                          markercolor = [:blue :green :red :orange :pink],
+                          )
+  Plots.gui(fig2)
+  #Plots.png(fig2)
+  readline()
+  
+end
+
+
+
+function main()
+
+  # modify this to the file path
+  filename = "/home/wgodoy/hdd_home/work/proxy_io/AmrexCastro/plot_size.csv"
+  # filename = "/home/wgodoy/hdd_home/work/proxy_io/AmrexCastro/plot_size_case4.csv"
+  plot_CSV_3col(filename)
+end
+
+main()
+

src/helper/helperSystem.jl

@@ -57,6 +57,10 @@ end
 function helper_get_directory_size(directory::String)::Int64
 
     size::Int64 = 0
+    if !isdir(directory)
+      return size
+    end
+    
     for (root, dirs, files) in walkdir(directory)
         size += sum(map(filesize, joinpath.(root, files)))
     end

test/data/AmrexCastro/gravity_tests/DustCollapse/case12/chk_3d_00000/CPUtime

+0.10744606400000001
\ No newline at end of file

test/data/AmrexCastro/gravity_tests/DustCollapse/case12/chk_3d_00000/CastroHeader

+Checkpoint version: 7

test/data/AmrexCastro/gravity_tests/DustCollapse/case12/chk_3d_00000/FabArrayHeaders.txt

+Level_0/SD_0_New_MF
+Level_0/SD_1_New_MF
+Level_0/SD_3_New_MF

test/data/AmrexCastro/gravity_tests/DustCollapse/case12/chk_3d_00000/Header

+CheckPointVersion_1.0
+3
+0
+0
+0