Splines Method (2)
2021-1νκΈ°, λνμμ βν΅κ³μ λ°μ΄ν°λ§μ΄λβ μμ μ λ£κ³ 곡λΆν λ°λ₯Ό μ 리ν κΈμ λλ€. μ§μ μ μΈμ λ νμμ λλ€ :)
μ΄ ν¬μ€νΈλ Regression Splineκ³Ό μ΄μ΄μ§λ λ΄μ©μ λλ€ π
Non-parameteric Logistic Regression
λ³Έλ <Binary Logistic Regreeion>μ μλμ κ°μ΄ λͺ¨λΈλ§νλ€.
\[\log \frac{P(Y = 1 \mid X=x)}{P(Y = 0 \mid X=x)} = \beta^T x\]μμ μμ λ€μ μ μ 리νλ©΄ μλμ κ°λ€.
\[P(Y = 1 \mid X = x) = \frac{e^{\beta^T x}}{1 + e^{\beta^T x}}\]<Non-parametric (binary) logistic regression>μ μμ μμμ $\beta^T x$λ₯Ό $f(x)$λ‘ λ체νλ€!!
\[P(Y = 1 \mid X = x) = \frac{e^{f(x)}}{1 + e^{f(x)}}\]μ΄λ, $f(x)$λ νμ¬ λͺ¨λ₯΄λ μνλ‘ μ°λ¦¬κ° estimation ν΄μΌ νλ λμμ΄λ€!!
μ κ· μμ μμλ $f(\cdot)$λ₯Ό μΆμ νλ λ°©μμΌλ‘ μλμ βpenalized log-likelihood functionβμ Maximize νλ κ²μ μ μνλ€.
\[\ell_\lambda (f) = \sum^n_{i=1} \left[ y_i f(x_i) - \log (1 + e^{f(x_i)}) \right] - \frac{\lambda}{2} \int \left\{ f''(t) \right\}^2 \; dt\]볡μ‘νκ² μκ°νκΈ° 보λ€λ <smoothing spline>κ³Ό λΉμ·ν ννλΌκ³ μΈμ ν΄λμ!
Multi-dimensional Splines
μ§κΈκΉμ§ μ΄ν΄λ³Έ <Spline Method>λ λͺ¨λ 1-dimensional spline modelμ΄μλ€. νμ§λ§, λ§μ κ²½μ° featureμ μνΈμμ©μ κ³ λ €νλ multi-dimensionalν μ κ·Όμ νμλ‘ νλ€.
<Multi-dimensional Spline>μ μλμ κ°μ΄ λͺ¨λΈλ§ νλ€.
\[f(X) = \sum^{M_1}_{i=1} \sum^{M_2}_{j=1} \; \theta_{ij} \cdot g_{ij} (X)\]where $g_{ij}(X)$ is the tensor product of basis function, defined by
\[g_{ij}(X) = h_{1i} (X_1) \cdot h_{2j} (X_2)\]μ¦, βmulti-dimensional splineβμ λ basis splineμ κ³±ν κ²μ basis functionμΌλ‘ μΌλλ€λ λ§μ΄λ€!

μμ κ°μ λ°©μμΌλ‘ μ κ·Όνλ©΄, 2-dim λΏλ§ μλλΌ d-dimκΉμ§λ μ½κ² generalization ν μ μλ€.
κ·Έλ¬λ input variableμ μ $d$κ° μ¦κ°νλ€λ©΄, multi-dimensional modelμ΄ νμλ‘ νλ basis functionμ exponentialνκ² μ¦κ°νλ€. μ΄κ²μ κ³μ°λ λΏλ§ μλλΌ curse of dimensionality λ±μ λ¬Έμ λ₯Ό λλ°νλ€.
μ΄λ° λ¬Έμ λ₯Ό ν΄κ²°νκΈ° μν λμμΌλ‘ 1991λ , <MARS; Multi-variate Adaptive Regression Spline>κ° μ μλμλ€.
λ, μ κ· κ³Όμ μ λ§μ§λ§ μ¦μμ λ€λ£° <Additive Model> μμ μ΄λ° multi-dimensional modelμ λ¬Έμ λ₯Ό ν΄κ²°νλ λμμ΄ λλ€.
μ΄μ΄μ§λ ν¬μ€νΈμμλ KNN κΈ°λ°μ non-parametric methodμ λν΄ μ΄ν΄λ³΄κ² λ€.
π KNN & kernel method