consolidation fluxnet hybrid

examples/exp_fluxnet_hybrid/Project.toml

@@ -1,9 +1,11 @@
 [deps]
 CMAEvolutionStrategy = "8d3b24bd-414e-49e0-94fb-163cc3a3e411"
+Distributed = "8ba89e20-285c-5b6f-9357-94700520ee1b"
 ForwardDiff = "f6369f11-7733-5829-9624-2563aa707210"
 GCMAES = "4aa9d100-eb0f-11e8-15f1-25748831eb3b"
-GLMakie = "e9467ef8-e4e7-5192-8a1a-b1aee30e663a"
+MLUtils = "f1d291b0-491e-4a28-83b9-f70985020b54"
 ProgressMeter = "92933f4c-e287-5a05-a399-4b506db050ca"
+Revise = "295af30f-e4ad-537b-8983-00126c2a3abe"
 Sindbad = "6686e6de-2010-4bf8-9178-b2bcc470766e"
 SindbadData = "772b809f-5329-4bfd-a2e9-cc4714ce84b1"
 SindbadML = "eb1d8004-2d42-4eef-b369-ed0645c101e8"
@@ -12,14 +14,26 @@ SindbadOptimization = "64fff8bd-0c13-402c-aa96-b97d0bc3655c"
 SindbadSetup = "2b7f2987-8a1e-48b4-8a02-cc9e7c4eeb1c"
 SindbadTEM = "f6108451-10cb-42fa-b0c1-67671cf08f15"
 SindbadUtils = "a2ad09c8-73f9-4387-ada0-5c5a8c58f8ab"
+SlurmClusterManager = "c82cd089-7bf7-41d7-976b-6b5d413cbe0a"
 StatsBase = "2913bbd2-ae8a-5f71-8c99-4fb6c76f3a91"
 
-[sources]
-Sindbad = {path = "../.."}
-SindbadData = {path = "../../lib/SindbadData"}
-SindbadML = {path = "../../lib/SindbadML"}
-SindbadMetrics = {path = "../../lib/SindbadMetrics"}
-SindbadOptimization = {path = "../../lib/SindbadOptimization"}
-SindbadSetup = {path = "../../lib/SindbadSetup"}
-SindbadTEM = {path = "../../lib/SindbadTEM"}
-SindbadUtils = {path = "../../lib/SindbadUtils"}
+[sources.Sindbad]
+path = "../../"
+
+[sources.SindbadData]
+path = "../../lib/SindbadData"
+
+[sources.SindbadML]
+path = "../../lib/SindbadML"
+
+[sources.SindbadMetrics]
+path = "../../lib/SindbadMetrics"
+
+[sources.SindbadSetup]
+path = "../../lib/SindbadSetup"
+
+[sources.SindbadTEM]
+path = "../../lib/SindbadTEM"
+
+[sources.SindbadUtils]
+path = "../../lib/SindbadUtils"

examples/exp_fluxnet_hybrid/analysis/Project.toml

@@ -3,7 +3,6 @@ ArchGDAL = "c9ce4bd3-c3d5-55b8-8973-c0e20141b8c3"
 CairoMakie = "13f3f980-e62b-5c42-98c6-ff1f3baf88f0"
 DimensionalData = "0703355e-b756-11e9-17c0-8b28908087d0"
 Flux = "587475ba-b771-5e3f-ad9e-33799f191a9c"
-GLMakie = "e9467ef8-e4e7-5192-8a1a-b1aee30e663a"
 GeoMakie = "db073c08-6b98-4ee5-b6a4-5efafb3259c6"
 JLD2 = "033835bb-8acc-5ee8-8aae-3f567f8a3819"
 LaTeXStrings = "b964fa9f-0449-5b57-a5c2-d3ea65f4040f"
@@ -13,5 +12,7 @@ NetCDF = "30363a11-5582-574a-97bb-aa9a979735b9"
 Proj = "c94c279d-25a6-4763-9509-64d165bea63e"
 Rasters = "a3a2b9e3-a471-40c9-b274-f788e487c689"
 Statistics = "10745b16-79ce-11e8-11f9-7d13ad32a3b2"
+SwarmMakie = "0b1c068e-6a84-4e66-8136-5c95cafa83ed"
+TypedTables = "9d95f2ec-7b3d-5a63-8d20-e2491e220bb9"
 YAXArrays = "c21b50f5-aa40-41ea-b809-c0f5e47bfa5c"
 Zarr = "0a941bbe-ad1d-11e8-39d9-ab76183a1d99"

examples/exp_fluxnet_hybrid/analysis/fold_analysis.jl

@@ -1,32 +1,42 @@
+ENV["JULIA_NUM_PRECOMPILE_TASKS"] = "1"
 using CairoMakie
 using SwarmMakie
 using Flux
 using Statistics
 using TypedTables
 using JLD2
 
-path_ptmp = "/ptmp/lalonso/HybridOutput/"
-
+path_ptmp = "/ptmp/lalonso/HybridOutputALL/"
+exp_name = "HyALL_ALL"
+_nfold = 5
+nlayers = 3
+n_neurons = 32
+bs= 32
+nepochs=500
 experiment = "$(exp_name)_fold_$(_nfold)_nlayers_$(nlayers)_n_neurons_$(n_neurons)_batch_size_$(bs)/checkpoint"
+# experiment = "$(exp_name)_kσ_1.0_fold_$(_nfold)_nlayers_$(nlayers)_n_neurons_$(n_neurons)_$(nepochs)epochs_batch_size_$(bs)/checkpoint"
 checkpoint_path = joinpath(path_ptmp, experiment)
-losses = JLD2.load(joinpath(checkpoint_path, "checkpoint_epoch_500.jld2"))
+losses = JLD2.load(joinpath(checkpoint_path, "checkpoint_epoch_100.jld2"))
+# "HyALL_ALL_kσ_1.0_fold_5_nlayers_3_n_neurons_32_500epochs_batch_size_32"
+# "HyALL_ALL_kσ_1.0_fold_5_nlayers_3_n_neurons_32_batch_size_32_500epochs"
 
 losses["loss_split_testing"]
 
 # Load the training history
-function load_losses(exp_name, _nfold, nlayers, n_neurons, bs, nepochs)
-    path_ptmp = "/ptmp/lalonso/HybridOutput/"
+function load_losses(exp_name, _nfold, nlayers, n_neurons, bs, nepochs, load_nepochs)
+    path_ptmp = "/ptmp/lalonso/HybridOutputALL/"
     if exp_name == "HyFixK_PFT"
         bs = bs * 123
     end
     experiment = "$(exp_name)_fold_$(_nfold)_nlayers_$(nlayers)_n_neurons_$(n_neurons)_batch_size_$(bs)/checkpoint"
+    # experiment = "$(exp_name)_kσ_1.0_fold_$(_nfold)_nlayers_$(nlayers)_n_neurons_$(n_neurons)_$(nepochs)epochs_batch_size_$(bs)/checkpoint"
     checkpoint_path = joinpath(path_ptmp, experiment)
 
     μtrain = Float32[]
     μval = Float32[]
     μtest = Float32[]
 
-    for epoch in 1:nepochs
+    for epoch in 1:load_nepochs
         losses = JLD2.load(joinpath(checkpoint_path, "checkpoint_epoch_$epoch.jld2"))
         push!(μtrain, mean(losses["loss_training"]))
         push!(μval, mean(losses["loss_validation"]))
@@ -36,6 +46,8 @@ function load_losses(exp_name, _nfold, nlayers, n_neurons, bs, nepochs)
     return μtrain, μval, μtest
 end
 
+mkpath(joinpath(@__DIR__, "figs"))
+
 function plot_training_history(μtrain, μval, μtest)
     with_theme(theme_light()) do
         fig = Figure(; size = (600, 400))
@@ -45,62 +57,72 @@ function plot_training_history(μtrain, μval, μtest)
         lines!(ax, μtest, color = :olive, linewidth = 1.25, label = "test")
         # ylims!(ax, 3, 4)
         axislegend(ax, position = :rt)
-        save("history_losses_5.png", fig)
+        save(joinpath(@__DIR__, "figs/history_losses_5.png"), fig)
     end
 end
 
 # ? load n-fold history
 nepochs=500
 _nfold = 5
-nlayers= 2
+nlayers= 3
 n_neurons = 32
 bs = 32
-exp_name = "HyFixK_ALL"
+# exp_name = "HyFixK_ALL"
+exp_name = "HyALL_ALL"
 
-μtrain, μval, μtest = load_losses(exp_name, _nfold, nlayers, n_neurons, bs, 500)
+μtrain, μval, μtest = load_losses(exp_name, _nfold, nlayers, n_neurons, bs, 500, 210)
 plot_training_history(μtrain, μval, μtest)
 
 
-function plot_training_history_folds(exp_name, nlayers, n_neurons, bs, nepochs)
-    with_theme(theme_light()) do
-        fig = Figure(; size = (400, 900))
-        axs = [Axis(fig[row, 1]) for row in 1:5]
-        for (_nfold, ax) in enumerate(axs)
-            μtrain, μval, μtest = load_losses(exp_name, _nfold, nlayers, n_neurons, bs, nepochs)
+function plot_training_history_folds(exp_name, nlayers, n_neurons, bs, nepochs; xpos = 185)
+    with_theme(theme_latexfonts()) do
+        fig = Figure(; size = (1200, 400), fontsize=24)
+        axs = [Axis(fig[row, col]; xgridstyle=:dash, ygridstyle=:dash, xlabel = "epoch", ylabel="loss")
+            for row in 1:2 for col in 1:3]
+        for (_nfold, ax) in enumerate(axs[1:5])
+            μtrain, μval, μtest = load_losses(exp_name, _nfold, nlayers, n_neurons, bs, 500, nepochs)
 
             lines!(ax, μtrain, color = :dodgerblue, linewidth = 1.25, label = "training")
             lines!(ax, μval, color = :orangered, linewidth = 1.25, label = "validation")
             lines!(ax, μtest, color = :olive, linewidth = 1.25, label = "test")
-            ax.title = "fold $(_nfold)"
+            # ax.title = "fold $(_nfold)"
+            text!(ax, [xpos], [5.5], text="fold $(_nfold)", color = :grey25)
         end
-        axislegend(axs[1], position = :rt, nbanks=3)
-        hidexdecorations!.(axs[1:4], ticks=false, grid=false)
+        Legend(fig[2, 3], axs[1], tellwidth=false, tellheight=false, halign=0,
+            framewidth=0.25, patchcolor = (:white, 0.25) )
+        # axislegend(axs[], position = :ct, nbanks=3, framewidth=0.25, patchcolor = (:white, 0.25))
+        hidexdecorations!.(axs[1:3], ticks=false, grid=false)
+        hidespines!(axs[end])
+        hidedecorations!(axs[end])
         linkaxes!.(axs)
-        limits!.(axs, 0, nepochs, 2.7, 4.5)
+        limits!.(axs, 0, nepochs, 2.7, 6)
         rowgap!(fig.layout, 0)
-        save("$(exp_name)_nlayers_$(nlayers)_n_neurons_$(n_neurons)_bs_$(bs)_history.png", fig)
+        hidespines!.(axs)
+        save("$(exp_name)_nlayers_$(nlayers)_n_neurons_$(n_neurons)_bs_$(bs)_history.pdf", fig)
     end
 end
 
 # ? load n-fold history
-nepochs=500
-_nfold = 5
-nlayers= 2
-n_neurons = 32
-bs = 32
-exp_name = "HyFixK_ALL" # HyFixK_PFT
-plot_training_history_folds(exp_name, nlayers, n_neurons, bs, nepochs)
+# nepochs=500
+# _nfold = 5
+# nlayers= 2
+# n_neurons = 32
+# bs = 32
+# exp_name = "HyFixK_ALL" # HyFixK_PFT
+# plot_training_history_folds(exp_name, nlayers, n_neurons, bs, nepochs)
 
-exp_name = "HyFixK_PFT" #
-plot_training_history_folds(exp_name, nlayers, n_neurons, bs, nepochs)
+# exp_name = "HyFixK_PFT" #
+# plot_training_history_folds(exp_name, nlayers, n_neurons, bs, nepochs)
 
 exp_name = "HyALL_ALL"
-nepochs = 150
-plot_training_history_folds(exp_name, nlayers, n_neurons, bs, nepochs)
+nepochs = 210
+plot_training_history_folds(exp_name, 2, n_neurons, bs, nepochs; xpos = 170)
+plot_training_history_folds(exp_name, 3, n_neurons, bs, 180; xpos = 170)
 
-exp_name = "HyALL_PFT"
-nepochs = 100
-plot_training_history_folds(exp_name, 3, n_neurons, bs, nepochs)
+
+# exp_name = "HyALL_PFT"
+# nepochs = 100
+# plot_training_history_folds(exp_name, 3, n_neurons, bs, nepochs)
 
 # do per variable
 
@@ -147,8 +169,8 @@ split_tested = collect_folds()
 function plot_box_split(split_m)
     _constraints = ["gpp", "nee", "reco", "transpiration", "evapotranspiration", "agb", "ndvi"]
     _colors = ["#4CAF50", "#1565C0", "#D32F2F", "#00ACC1", "#00897B", "#8D6E63", "#CDDC39"]
-    with_theme() do
-        fig = Figure(; size = (1200, 300))
+    with_theme(theme_latexfonts()) do
+        fig = Figure(; size = (1200, 300), fontsize=24)
         axs = [Axis(fig[1, i]) for i in 1:7]
         for (i, ax) in enumerate(axs)
             tmp_vec = filter(!isnan, split_m[:,i])
@@ -165,7 +187,7 @@ function plot_box_split(split_m)
         # hidexdecorations!.(axs, grid=false, label=false, ticklabels=false, ticks=true)
         hidexdecorations!.(axs, grid=false, label=false, ticklabels=true)
         hidespines!.(axs)
-        save("$(exp_name)_nlayers_$(nlayers)_n_neurons_$(n_neurons)_bs_$(bs)_box_test.png", fig)
+        save("$(exp_name)_nlayers_$(nlayers)_n_neurons_$(n_neurons)_bs_$(bs)_box_test.pdf", fig)
     end
 end
 
@@ -175,8 +197,8 @@ plot_box_split(split_tested)
 function plot_violin_split(split_m)
     _constraints = ["gpp", "nee", "reco", "transpiration", "evapotranspiration", "agb", "ndvi"]
     _colors = ["#4CAF50", "#1565C0", "#D32F2F", "#00ACC1", "#00897B", "#8D6E63", "#CDDC39"]
-    with_theme() do
-        fig = Figure(; size = (1200, 300))
+    with_theme(theme_latexfonts()) do
+        fig = Figure(; size = (1200, 300), fontsize=24)
         axs = [Axis(fig[1, i]) for i in 1:7]
         for (i, ax) in enumerate(axs)
             tmp_vec = filter(!isnan, split_m[:,i])
@@ -195,19 +217,50 @@ function plot_violin_split(split_m)
         # hidexdecorations!.(axs, grid=false, label=false, ticklabels=false, ticks=true)
         hidexdecorations!.(axs, grid=false, label=false, ticklabels=true)
         hidespines!.(axs)
-        save("$(exp_name)_nlayers_$(nlayers)_n_neurons_$(n_neurons)_bs_$(bs)_violin_test.png", fig)
+        save("$(exp_name)_nlayers_$(nlayers)_n_neurons_$(n_neurons)_bs_$(bs)_violin_test.pdf", fig)
     end
 end
 
 plot_violin_split(split_tested)
 
+function plot_beeswarm_split(split_m)
+    _constraints = ["gpp", "nee", "reco", "transpiration", "evapotranspiration", "agb", "ndvi"]
+    _colors = ["#4CAF50", "#1565C0", "#D32F2F", "#00ACC1", "#00897B", "#8D6E63", "#CDDC39"]
+    with_theme(theme_latexfonts()) do
+        fig = Figure(; size = (1200, 300), fontsize=24)
+        axs = [Axis(fig[1, i]) for i in 1:7]
+        for (i, ax) in enumerate(axs)
+            tmp_vec = filter(!isnan, split_m[:,i])
+
+            beeswarm!(ax, fill(1, length(tmp_vec)), tmp_vec; color = (_colors[i], 0.65), markersize = 6)
+            # boxplot!(ax, fill(1, length(tmp_vec)), tmp_vec; width=0.35, strokecolor = :white,
+            #     strokewidth=1.5, whiskercolor=:white, mediancolor=:white,
+            #     color = :transparent)
+            ax.title= rich("$(_constraints[i])", color=:grey15, font=:bold)
+            ax.yticks=[0, 0.25, 0.5, 0.75, 1.0]
+        end
+        axs[1].ylabel = "Loss"
+        ylims!.(axs, 0, 1)
+        xlims!.(axs, 0.5, 1.5)
+        hideydecorations!.(axs[2:end], ticks=true, grid=false)
+        # hidexdecorations!.(axs, grid=false, label=false, ticklabels=false, ticks=true)
+        hidexdecorations!.(axs, grid=false, label=false, ticklabels=true)
+        hidespines!.(axs)
+        save("$(exp_name)_nlayers_$(nlayers)_n_neurons_$(n_neurons)_bs_$(bs)_beeswarm_test.pdf", fig)
+    end
+end
 
 # ! Collect all experiments
 # "HyFixK_ALL", "HyFixK_PFT", "HyALL_ALL", "HyALL_PFT"
-split_tested_HyALL_ALL = collect_folds("HyALL_ALL", 400, 32, 2, 32)
-split_tested_HyALL_PFT = collect_folds("HyALL_PFT", 300,  32, 2, 32)
-split_tested_HyFixK_ALL = collect_folds("HyFixK_ALL", 500, 32, 2, 32)
-split_tested_HyFixK_PFT = collect_folds("HyFixK_PFT", 500, 32, 2, 32)
+split_tested_HyALL_ALL = collect_folds("HyALL_ALL", 200, 32, 2, 32)
+# split_tested_HyALL_PFT = collect_folds("HyALL_PFT", 300,  32, 2, 32)
+# split_tested_HyFixK_ALL = collect_folds("HyFixK_ALL", 500, 32, 2, 32)
+# split_tested_HyFixK_PFT = collect_folds("HyFixK_PFT", 500, 32, 2, 32)
+plot_box_split(split_tested_HyALL_ALL)
+plot_violin_split(split_tested_HyALL_ALL)
+plot_beeswarm_split(split_tested_HyALL_ALL)
+
+
 
 function plot_box_split_all(split_m1, split_m2, split_m3, split_m4)
     _constraints = ["gpp", "nee", "reco", "transpiration", "evapotranspiration", "agb", "ndvi"]

examples/exp_fluxnet_hybrid/exp_fluxnet_hybrid.jl

@@ -1,3 +1,14 @@
+# activate project's environment and develop the package
+using Pkg
+Pkg.activate("examples/exp_fluxnet_hybrid")
+Pkg.develop(path=pwd())
+Pkg.instantiate()
+
+# get data
+# $examples/data> scp -r lalonso@ruthenia:/Net/Groups/BGI/work_5/scratch/lalonso/CovariatesFLUXNET_3.zarr .
+# @examples/data> scp -r lalonso@ruthenia:/Net/Groups/BGI/work_4/scratch/lalonso/FLUXNET_v2023_12_1D.zarr .
+
+# start using the package
 using SindbadData
 using SindbadData.DimensionalData
 using SindbadData.AxisKeys
@@ -7,7 +18,7 @@ using SindbadML
 using SindbadML.JLD2
 using ProgressMeter
 using SindbadOptimization
-include("load_covariates.jl")
+# include("load_covariates.jl")
 
 # load folds # $nfold $nlayer $neuron $batchsize
 _nfold = 5 #Base.parse(Int, ARGS[1])