add note

add Olivetti pca note

_freeze/category/dimension-reduction/2-olivetti-face/execute-results/html.json

@@ -0,0 +1,11 @@
+{
+  "hash": "b3e9fbdefd68e585282e16606555fadb",
+  "result": {
+    "markdown": "---\ntitle: \"2-olivetti-face-project\"\ncode-fold: true\n---\n\n## 1. load package\n\n::: {.cell execution_count=1}\n``` {.julia .cell-code}\nusing MLJ,DataFrames,CSV,Random\ninclude(\"./olivetti-face-code/1-dataprocessing.jl\")\n```\n\n::: {.cell-output .cell-output-display execution_count=2}\n```\nload_olivetti_faces (generic function with 1 method)\n```\n:::\n:::\n\n\n## 2. load data\n\n::: {.cell execution_count=2}\n``` {.julia .cell-code}\n(Xtrain, Xtest), (ytrain, ytest)=load_olivetti_faces()\n```\n\n::: {.cell-output .cell-output-display execution_count=3}\n\n::: {.ansi-escaped-output}\n```{=html}\n<pre>((<span class=\"ansi-bold\">320×4096 DataFrame</span>\n<span class=\"ansi-bold\"> Row </span>│<span class=\"ansi-bold\"> x1        </span><span class=\"ansi-bold\"> x2       </span><span class=\"ansi-bold\"> x3        </span><span class=\"ansi-bold\"> x4       </span><span class=\"ansi-bold\"> x5       </span><span class=\"ansi-bold\"> x6       </span><span class=\"ansi-bold\"> x7       </span> ⋯\n     │<span class=\"ansi-bright-black-fg\"> Float64   </span><span class=\"ansi-bright-black-fg\"> Float64  </span><span class=\"ansi-bright-black-fg\"> Float64   </span><span class=\"ansi-bright-black-fg\"> Float64  </span><span class=\"ansi-bright-black-fg\"> Float64  </span><span class=\"ansi-bright-black-fg\"> Float64  </span><span class=\"ansi-bright-black-fg\"> Float64  </span> ⋯\n─────┼──────────────────────────────────────────────────────────────────────────\n   1 │ 0.595041   0.640496  0.615702   0.644628  0.68595   0.72314   0.731405  ⋯\n   2 │ 0.103306   0.219008  0.177686   0.219008  0.392562  0.57438   0.669422\n   3 │ 0.289256   0.338843  0.417355   0.504132  0.553719  0.561983  0.582645\n   4 │ 0.528926   0.797521  0.826446   0.822314  0.822314  0.818182  0.805785\n   5 │ 0.161157   0.202479  0.268595   0.334711  0.384298  0.392562  0.396694  ⋯\n   6 │ 0.169422   0.293388  0.561983   0.677686  0.727273  0.756198  0.768595\n   7 │ 0.136364   0.177686  0.235537   0.289256  0.334711  0.363636  0.396694\n   8 │ 0.768595   0.756198  0.743802   0.743802  0.752066  0.747934  0.735537\n   9 │ 0.144628   0.210744  0.285124   0.342975  0.392562  0.404959  0.409091  ⋯\n  10 │ 0.566116   0.595041  0.603306   0.619835  0.636364  0.640496  0.661157\n  11 │ 0.524793   0.53719   0.578512   0.628099  0.669422  0.690083  0.68595\n  ⋮  │     ⋮         ⋮          ⋮         ⋮         ⋮         ⋮         ⋮      ⋱\n 311 │ 0.181818   0.338843  0.355372   0.404959  0.438017  0.458678  0.471074\n 312 │ 0.454545   0.528926  0.644628   0.747934  0.780992  0.780992  0.801653  ⋯\n 313 │ 0.553719   0.607438  0.636364   0.64876   0.652893  0.64876   0.673554\n 314 │ 0.0909091  0.136364  0.177686   0.231405  0.363636  0.504132  0.541322\n 315 │ 0.252066   0.252066  0.252066   0.252066  0.256198  0.239669  0.235537\n 316 │ 0.326446   0.483471  0.524793   0.599174  0.665289  0.702479  0.702479  ⋯\n 317 │ 0.53719    0.57438   0.553719   0.615702  0.38843   0.487603  0.690083\n 318 │ 0.128099   0.18595   0.247934   0.31405   0.38843   0.46281   0.520661\n 319 │ 0.586777   0.702479  0.731405   0.731405  0.743802  0.772727  0.793388\n 320 │ 0.136364   0.107438  0.0909091  0.115702  0.115702  0.119835  0.181818  ⋯\n<span class=\"ansi-cyan-fg\">                                               4089 columns and 299 rows omitted</span>, <span class=\"ansi-bold\">80×4096 DataFrame</span>\n<span class=\"ansi-bold\"> Row </span>│<span class=\"ansi-bold\"> x1       </span><span class=\"ansi-bold\"> x2       </span><span class=\"ansi-bold\"> x3       </span><span class=\"ansi-bold\"> x4       </span><span class=\"ansi-bold\"> x5       </span><span class=\"ansi-bold\"> x6       </span><span class=\"ansi-bold\"> x7       </span><span class=\"ansi-bold\"> x</span> ⋯\n     │<span class=\"ansi-bright-black-fg\"> Float64  </span><span class=\"ansi-bright-black-fg\"> Float64  </span><span class=\"ansi-bright-black-fg\"> Float64  </span><span class=\"ansi-bright-black-fg\"> Float64  </span><span class=\"ansi-bright-black-fg\"> Float64  </span><span class=\"ansi-bright-black-fg\"> Float64  </span><span class=\"ansi-bright-black-fg\"> Float64  </span><span class=\"ansi-bright-black-fg\"> F</span> ⋯\n─────┼──────────────────────────────────────────────────────────────────────────\n   1 │ 0.219008  0.235537  0.252066  0.326446  0.392562  0.553719  0.714876  0 ⋯\n   2 │ 0.194215  0.268595  0.367769  0.487603  0.545455  0.561983  0.578512  0\n   3 │ 0.210744  0.206612  0.194215  0.181818  0.219008  0.239669  0.256198  0\n   4 │ 0.392562  0.475207  0.661157  0.590909  0.471074  0.545455  0.673554  0\n   5 │ 0.719008  0.727273  0.72314   0.714876  0.72314   0.731405  0.739669  0 ⋯\n   6 │ 0.429752  0.458678  0.549587  0.623967  0.673554  0.714876  0.72314   0\n   7 │ 0.289256  0.157025  0.14876   0.190083  0.169422  0.194215  0.404959  0\n   8 │ 0.628099  0.665289  0.68595   0.694215  0.719008  0.731405  0.752066  0\n   9 │ 0.479339  0.549587  0.628099  0.690083  0.677686  0.652893  0.640496  0 ⋯\n  10 │ 0.677686  0.677686  0.681818  0.706612  0.731405  0.739669  0.756198  0\n  11 │ 0.123967  0.132231  0.11157   0.11157   0.119835  0.136364  0.136364  0\n  ⋮  │    ⋮         ⋮         ⋮         ⋮         ⋮         ⋮         ⋮        ⋱\n  71 │ 0.123967  0.128099  0.115702  0.136364  0.115702  0.107438  0.115702  0\n  72 │ 0.384298  0.22314   0.22314   0.305785  0.429752  0.508265  0.557851  0 ⋯\n  73 │ 0.409091  0.590909  0.657025  0.681818  0.694215  0.731405  0.760331  0\n  74 │ 0.487603  0.330578  0.252066  0.38843   0.785124  0.789256  0.780992  0\n  75 │ 0.285124  0.285124  0.272727  0.214876  0.169422  0.165289  0.264463  0\n  76 │ 0.136364  0.132231  0.123967  0.119835  0.11157   0.128099  0.132231  0 ⋯\n  77 │ 0.243802  0.243802  0.247934  0.247934  0.252066  0.256198  0.256198  0\n  78 │ 0.603306  0.586777  0.541322  0.603306  0.603306  0.607438  0.64876   0\n  79 │ 0.772727  0.764463  0.752066  0.764463  0.785124  0.793388  0.797521  0\n  80 │ 0.475207  0.491736  0.5       0.512397  0.524793  0.528926  0.545455  0 ⋯\n<span class=\"ansi-cyan-fg\">                                                4089 columns and 59 rows omitted</span>), (CategoricalArrays.CategoricalValue{Int64, UInt32}[27, 3, 39, 10, 23, 24, 23, 30, 23, 14  …  21, 26, 5, 35, 28, 9, 31, 0, 15, 37], CategoricalArrays.CategoricalValue{Int64, UInt32}[6, 35, 6, 31, 5, 4, 3, 2, 19, 4  …  32, 29, 22, 36, 12, 32, 16, 14, 3, 20]))</pre>\n```\n:::\n\n:::\n:::\n\n\n",
+    "supporting": [
+      "2-olivetti-face_files/figure-html"
+    ],
+    "filters": [],
+    "includes": {}
+  }
+}

category/dimension-reduction/2-olivetti-face-intro.qmd

@@ -0,0 +1,113 @@
+---
+title: "2-olivetti-face-project"
+code-fold: true
+---
+
+
+## olivetti  face recognition 的项目
+
+## 1. dataprocessing.jl 
+
+读取 csv 文件, 返回 返回 olivetti face  训练数据,测试数据和标签
+`load_olivetti_faces` 函数返回训练,测试数据集和对应标签
+```julia
+(Xtrain, Xtest), (ytrain, ytest)=load_olivetti_faces()
+```
+
+## 2. train&save-model.jl
+
+导入数据,返回一个高阶函数`make_model`
+`make_model` 函数首先接受训练数据集, 然后等待 `dim` 需要缩减到的维度参数
+
+最终训练的模型保存的对应的 jlso 文件中:`JLSO.save("$(pwd())/models/of-model-$(dim)pcs.jlso",:pca=>mach)`
+
+
+```julia
+function  make_model(Xtr)
+    return (dim)->begin
+        model = PCA(maxoutdim=dim)
+        mach = machine(model, Xtr) |> fit!
+        try
+            JLSO.save("$(pwd())/models/of-model-$(dim)pcs.jlso",:pca=>mach)
+            @info "$(dim) dimension pca model saved"
+        catch e
+            @warn "$(e) has problem"
+        end
+    end
+end
+
+make_ol_model=make_model(Xtrain)
+#make_ol_model.([10,20,150])
+```
+
+
+## 3-transform-reconstruct-methods
+在 `MLJ.jl`的 pca 方法和 `MultiVariate.jl` 的方法稍有不同, 使用的是 `transform`函数,而不是`project`方法 
+
+主要函数为`transform_to_pcadata1`,为高阶函数
+输入参数`dim` 为需要缩减到的维度
+函数内部调用第2 步获得的模型
+返回一个函数 等待 df 参数
+内部调用`transform`函数对数据做降维处理
+返回数据
+
+```julia
+function transform_to_pcadata(dim::Int)
+    mach = JLSO.load("$(pwd())/models/of-model-$(dim)pcs.jlso")[:pca]
+    return (imgs::DataFrame)->begin
+        @info "$dim pca proceeding..."
+        pcaX = transform(mach, imgs)     # 降维数据
+        # 返回降维数据
+        return pcaX
+    end
+end
+```
+
+
+`transform_to_pcadata2` 与`transform_to_pcadata1`  一样是高阶函数
+但是参数输入的顺序不同, `transform_to_pcadata2`中先输入dataframe, 然后等待维度参数
+```julia
+    function transform_to_pcadata2(imgs::DataFrame)
+
+        return (dim::Int)->begin
+            @info "$dim pca proceeding..."
+            mach = JLSO.load("$(pwd())/models/of-model-$(dim)pcs.jlso")[:pca]
+            pcaX = transform(mach, imgs)     # 降维数据
+            # 返回降维数据
+            return pcaX
+        end
+    end
+```
+
+### 重建数据方法
+从低维度数据恢复原始维度数据, 在`MLJ`中使用的方法是`inverse_transform`
+`reconstruct_data` 方法首先从数据 dataframe 中获取维度`column`数据
+从存储模型中调用训练模型
+执行重建变换
+
+```julia
+    """
+        reconstruct_data(imgs::DataFrame)
+    从降维数据重建图片
+    TBW
+    """
+    function reconstruct_data(imgs::DataFrame)
+
+            _,cols=size(imgs)
+            @info "imgs  reconstructing from $(cols) dimension" 
+            mach = JLSO.load("$(pwd())/models/of-model-$(cols)pcs.jlso")[:pca]
+            Xr = inverse_transform(mach, imgs)  # 重建近似数据
+            return Xr
+    end
+```
+
+## 维度缩减到 1d 的数据
+
+以 faces 为例,如果缩减到一个维度, 获得的数据是所有图片共用的最大元素,也就是所有人面部共用的特征
+
+
+
+
+
+
+

category/dimension-reduction/3-olivetti-face.qmd

@@ -0,0 +1,93 @@
+---
+title: "3-olivetti-face"
+author: math4mad
+code-fold: true
+---
+
+## 1. load package
+```{julia}
+include("./olivetti-face-code/1-dataprocessing.jl")
+
+import MLJ: transform, inverse_transform
+using MLJ,DataFrames,CSV,Random,JLSO,GLMakie
+Random.seed!(4545343)
+
+```
+
+## 2. import data
+```{julia}
+(Xtrain, Xtest), (ytrain, ytest)=load_olivetti_faces()
+```
+
+## 3. train&save model
+```{julia}
+    PCA = @load PCA pkg=MultivariateStats
+    function  make_model(Xtr)
+    return (dim)->begin
+        model = PCA(maxoutdim=dim)
+        mach = machine(model, Xtr) |> fit!
+        try
+            JLSO.save("./olivetti-face-code/models/of-model-$(dim)pcs.jlso",:pca=>mach)
+            @info "$(dim) dimension pca model saved"
+        catch e
+            @warn "$(e) has problem"
+        end
+    end
+end
+make_ol_model=make_model(Xtrain)
+make_ol_model.([1,2,3,100])
+
+```
+
+## 4. imgs project to  low dimension feature space
+第三行是降维到 100的图片, 最后一行是原始图片
+```{julia}
+include("./olivetti-face-code/3-transform-reconstruct-methods.jl")
+cat=ytrain|>Array|>levels
+rows,cols=size(Xtrain)
+
+pick20=rand(1:rows,20)
+pickXtrain=Xtrain[pick20,:]
+pickytrain=ytrain[pick20]
+
+pcaData=transform_to_2d(pickXtrain)
+reconstructImgs=reconstruct_data(pcaData)
+
+pcaData3=transform_to_3d(pickXtrain)
+reconstructImgs3=reconstruct_data(pcaData3)
+
+transform_to_100d=transform_to_pcadata1(100)
+pcaData100=transform_to_100d(pickXtrain)
+reconstructImgs100=reconstruct_data(pcaData100)
+
+df=vcat(reconstructImgs,reconstructImgs3,reconstructImgs100,pickXtrain)
+```
+plot reconstruct imgs
+```{julia}
+function  plot_img(df)
+    
+    fig=Figure(resolution=(130*20,130*4))
+    
+    for i in 0:3
+        for j in 1:20
+            idx=i*20+j
+            ax=Axis(fig[i+1,j],yreversed=true)
+            img=df[idx,:]|>Array|>d->reshape(d,w,h)
+            image!(ax,img)
+            hidespines!(ax)
+            hidedecorations!(ax)
+        end
+    end
+
+    fig
+    #save("./imgs/reconstruct-of-face.png",fig)
+end
+
+
+plot_img(df)
+```
+
+
+
+
+

category/dimension-reduction/olivetti-face-code/1-dataprocessing.jl

@@ -0,0 +1,29 @@
+
+
+using MLJ,DataFrames,CSV,Random
+
+const w=64
+const h=64
+const length=4096
+const str="olivetti_faces"
+
+
+
+
+of=olivetti_faces=CSV.File("./olivetti-face-code/olivetti_faces.csv") |> DataFrame
+coerce!(of,:label=>Multiclass)
+label,X=  unpack(of, ==(:label), rng=123);
+
+
+"""
+    load_olivetti_faces()
+    返回 olivetti face  训练数据,测试数据和标签
+    return (Xtrain, Xtest), (ytrain, ytest)
+    train:test=0.8
+"""
+function load_olivetti_faces()
+    (Xtrain, Xtest), (ytrain, ytest)  = partition((X, label), 0.8, multi=true,  rng=123)
+    return (Xtrain, Xtest), (ytrain, ytest)
+end
+
+return  load_olivetti_faces

category/dimension-reduction/olivetti-face-code/2-train&save-model.jl

@@ -0,0 +1,39 @@
+
+using MLJ,DataFrames,CSV,Random,JLSO
+
+include("1-dataprocessing.jl")
+
+
+(Xtrain, Xtest), (ytrain, ytest)=load_olivetti_faces()
+#df=vcat(Xtrain,Xtest)
+
+
+PCA = @load PCA pkg=MultivariateStats
+
+"""
+    make_model(Xtr)
+    接收训练数据, 返回函数等待输入降维维度
+
+ Arguments
+- Xtr  训练数据集
+ 
+return 新函数 arguments 为 降维维度 dim
+
+
+
+"""
+function  make_model(Xtr)
+    return (dim)->begin
+        model = PCA(maxoutdim=dim)
+        mach = machine(model, Xtr) |> fit!
+        try
+            JLSO.save("$(pwd())/models/of-model-$(dim)pcs.jlso",:pca=>mach)
+            @info "$(dim) dimension pca model saved"
+        catch e
+            @warn "$(e) has problem"
+        end
+    end
+end
+
+#make_ol_model=make_model(Xtrain)
+#make_ol_model.([10,20,150])