2-svm-diabetes-classfication

介绍

• 参考博客文章:diagnose-diabetes-with-svm

• SVM(支持向量机)通过引入 kernelfunction,使得模型的分类灵活性大大增强,可以解决更多问题.在julia中可以通过在LIBSVM.jl 引入 kernel function 实现, 参见 文档: Support Vector Machine

• MLJ.jl 通过包装接口也提供相似功能

• 响应变量需要转换类型 to_ScienceType(d)=coerce(d,:Outcome=> Multiclass)

1. load package

Code

include("../utils.jl")
import MLJ: fit!, predict
using CSV,DataFrames,Random
using MLJ
using Plots
using KernelFunctions

2. process data

Code

 df=load_csv("diabetes")
 to_ScienceType(d)=coerce(d,:Outcome=> Multiclass)
 df=to_ScienceType(df)
 first(df,5)|>display
 y, X =  unpack(df, ==(:Outcome), rng=123);
 (Xtrain, Xtest), (ytrain, ytest)  = partition((X, y), 0.7, multi=true,  rng=123)
display(schema(X))

5×9 DataFrame

Row	Pregnancies	Glucose	BloodPressure	SkinThickness	Insulin	BMI	DiabetesPedigreeFunction	Age	Outcome
	Int64	Int64	Int64	Int64	Int64	Float64	Float64	Int64	Cat…
1	6	148	72	35	0	33.6	0.627	50	1
2	1	85	66	29	0	26.6	0.351	31	0
3	8	183	64	0	0	23.3	0.672	32	1
4	1	89	66	23	94	28.1	0.167	21	0
5	0	137	40	35	168	43.1	2.288	33	1

┌──────────────────────────┬────────────┬─────────┐
│ names                    │ scitypes   │ types   │
├──────────────────────────┼────────────┼─────────┤
│ Pregnancies              │ Count      │ Int64   │
│ Glucose                  │ Count      │ Int64   │
│ BloodPressure            │ Count      │ Int64   │
│ SkinThickness            │ Count      │ Int64   │
│ Insulin                  │ Count      │ Int64   │
│ BMI                      │ Continuous │ Float64 │
│ DiabetesPedigreeFunction │ Continuous │ Float64 │
│ Age                      │ Count      │ Int64   │
└──────────────────────────┴────────────┴─────────┘

3. MLJ workflow

3.1 defin model

Code

SVC = @load SVC pkg=LIBSVM
#define kernel function,evaulate  kernelfunctions methods
kernels=[PolynomialKernel(; degree=2, c=1), 
         SqExponentialKernel(),
         NeuralNetworkKernel(),
         LinearKernel(;c=1.0)
]

svc_mdls = [SVC(;kernel=k) for  k in kernels]
svcs = [machine(model, Xtrain, ytrain;scitype_check_level=0) for model in svc_mdls]
[fit!(svc) for svc in svcs]

[ Info: For silent loading, specify `verbosity=0`. 
[ Info: Training machine(SVC(kernel = Polynomial Kernel (c = 1, degree = 2), …), …).

WARNING: reaching max number of iterations
[ Info: Training machine(SVC(kernel = Squared Exponential Kernel (metric = Distances.Euclidean(0.0)), …), …).
[ Info: Training machine(SVC(kernel = Neural Network Kernel, …), …).
[ Info: Training machine(SVC(kernel = Linear Kernel (c = 1.0), …), …).

WARNING: reaching max number of iterations

import MLJLIBSVMInterface ✔

4-element Vector{Machine{MLJLIBSVMInterface.SVC, true}}:
 machine(SVC(kernel = Polynomial Kernel (c = 1, degree = 2), …), …)
 machine(SVC(kernel = Squared Exponential Kernel (metric = Distances.Euclidean(0.0)), …), …)
 machine(SVC(kernel = Neural Network Kernel, …), …)
 machine(SVC(kernel = Linear Kernel (c = 1.0), …), …)

3.2 predict test

Code

for (idx, str) in enumerate(["Polynomial ","Gaussian","NeuralNetwork","Linear"])
    local yhat=predict(svcs[idx],Xtest)
    local acc=accuracy(yhat,ytest) 
    @info "$(str) kernel predict accuracy"=>acc   
end

[ Info: "Polynomial  kernel predict accuracy" => 0.47391304347826085
[ Info: "Gaussian kernel predict accuracy" => 0.6434782608695652
[ Info: "NeuralNetwork kernel predict accuracy" => 0.6478260869565218
[ Info: "Linear kernel predict accuracy" => 0.7782608695652173