2021-1ν•™κΈ°, λŒ€ν•™μ—μ„œ β€˜ν†΅κ³„μ  λ°μ΄ν„°λ§ˆμ΄λ‹β€™ μˆ˜μ—…μ„ λ“£κ³  κ³΅λΆ€ν•œ λ°”λ₯Ό μ •λ¦¬ν•œ κΈ€μž…λ‹ˆλ‹€. 지적은 μ–Έμ œλ‚˜ ν™˜μ˜μž…λ‹ˆλ‹€ :)

2 minute read

2021-1ν•™κΈ°, λŒ€ν•™μ—μ„œ β€˜ν†΅κ³„μ  λ°μ΄ν„°λ§ˆμ΄λ‹β€™ μˆ˜μ—…μ„ λ“£κ³  κ³΅λΆ€ν•œ λ°”λ₯Ό μ •λ¦¬ν•œ κΈ€μž…λ‹ˆλ‹€. 지적은 μ–Έμ œλ‚˜ ν™˜μ˜μž…λ‹ˆλ‹€ :)

이 ν¬μŠ€νŠΈλŠ” Regression Splineκ³Ό μ΄μ–΄μ§€λŠ” λ‚΄μš©μž…λ‹ˆλ‹€ 😊

Non-parameteric Logistic Regression

본래 <Binary Logistic Regreeion>은 μ•„λž˜μ™€ 같이 λͺ¨λΈλ§ν•œλ‹€.

\[\log \frac{P(Y = 1 \mid X=x)}{P(Y = 0 \mid X=x)} = \beta^T x\]

μœ„μ˜ 식을 λ‹€μ‹œ 잘 μ •λ¦¬ν•˜λ©΄ μ•„λž˜μ™€ κ°™λ‹€.

\[P(Y = 1 \mid X = x) = \frac{e^{\beta^T x}}{1 + e^{\beta^T x}}\]

<Non-parametric (binary) logistic regression>은 μœ„μ˜ μ‹μ—μ„œ $\beta^T x$λ₯Ό $f(x)$둜 λŒ€μ²΄ν•œλ‹€!!

\[P(Y = 1 \mid X = x) = \frac{e^{f(x)}}{1 + e^{f(x)}}\]

μ΄λ•Œ, $f(x)$λŠ” ν˜„μž¬ λͺ¨λ₯΄λŠ” μƒνƒœλ‘œ μš°λ¦¬κ°€ estimation ν•΄μ•Ό ν•˜λŠ” λŒ€μƒμ΄λ‹€!!

μ •κ·œ μˆ˜μ—…μ—μ„œλŠ” $f(\cdot)$λ₯Ό μΆ”μ •ν•˜λŠ” λ°©μ‹μœΌλ‘œ μ•„λž˜μ˜ β€œpenalized log-likelihood function”을 Maximize ν•˜λŠ” 것을 μ œμ‹œν•œλ‹€.

\[\ell_\lambda (f) = \sum^n_{i=1} \left[ y_i f(x_i) - \log (1 + e^{f(x_i)}) \right] - \frac{\lambda}{2} \int \left\{ f''(t) \right\}^2 \; dt\]

λ³΅μž‘ν•˜κ²Œ μƒκ°ν•˜κΈ° λ³΄λ‹€λŠ” <smoothing spline>κ³Ό λΉ„μŠ·ν•œ ν˜•νƒœλΌκ³  인식 ν•΄λ‘μž!


Multi-dimensional Splines

μ§€κΈˆκΉŒμ§€ μ‚΄νŽ΄λ³Έ <Spline Method>λŠ” λͺ¨λ‘ 1-dimensional spline modelμ΄μ—ˆλ‹€. ν•˜μ§€λ§Œ, λ§Žμ€ 경우 feature의 μƒν˜Έμž‘μš©μ„ κ³ λ €ν•˜λŠ” multi-dimensionalν•œ 접근을 ν•„μš”λ‘œ ν•œλ‹€.

<Multi-dimensional Spline>은 μ•„λž˜μ™€ 같이 λͺ¨λΈλ§ ν•œλ‹€.

\[f(X) = \sum^{M_1}_{i=1} \sum^{M_2}_{j=1} \; \theta_{ij} \cdot g_{ij} (X)\]

where $g_{ij}(X)$ is the tensor product of basis function, defined by

\[g_{ij}(X) = h_{1i} (X_1) \cdot h_{2j} (X_2)\]

즉, β€œmulti-dimensional spline”은 두 basis spline을 κ³±ν•œ 것을 basis function으둜 μ‚ΌλŠ”λ‹€λŠ” 말이닀!

μœ„μ™€ 같은 λ°©μ‹μœΌλ‘œ μ ‘κ·Όν•˜λ©΄, 2-dim 뿐만 μ•„λ‹ˆλΌ d-dimκΉŒμ§€λ„ μ‰½κ²Œ generalization ν•  수 μžˆλ‹€.

κ·ΈλŸ¬λ‚˜ input variable의 수 $d$κ°€ μ¦κ°€ν•œλ‹€λ©΄, multi-dimensional model이 ν•„μš”λ‘œ ν•˜λŠ” basis function은 exponentialν•˜κ²Œ μ¦κ°€ν•œλ‹€. 이것은 κ³„μ‚°λŸ‰ 뿐만 μ•„λ‹ˆλΌ curse of dimensionality λ“±μ˜ 문제λ₯Ό λ™λ°˜ν•œλ‹€.

이런 문제λ₯Ό ν•΄κ²°ν•˜κΈ° μœ„ν•œ λŒ€μ•ˆμœΌλ‘œ 1991λ…„, <MARS; Multi-variate Adaptive Regression Spline>κ°€ μ œμ‹œλ˜μ—ˆλ‹€.

또, μ •κ·œ κ³Όμ •μ˜ λ§ˆμ§€λ§‰ μ¦ˆμŒμ— λ‹€λ£° <Additive Model> μ—­μ‹œ 이런 multi-dimensional model의 문제λ₯Ό ν•΄κ²°ν•˜λŠ” λŒ€μ•ˆμ΄ λœλ‹€.


μ΄μ–΄μ§€λŠ” ν¬μŠ€νŠΈμ—μ„œλŠ” KNN 기반의 non-parametric method에 λŒ€ν•΄ μ‚΄νŽ΄λ³΄κ² λ‹€.

πŸ‘‰ KNN & kernel method