User talk:Dsevern

Newton-Raphson Method (Lecture: Oct 11, 2011)

Previously we had derivated the log likelihood function for the logistic function.

[math]\displaystyle{ \begin{align} {\ell(\mathbf{\beta\,})} & {} = \sum_{i=1}^n \left(y_i {\mathbf{\beta\,}^T \mathbf{x_i}} - \ln({1+e^{\mathbf{\beta\,}^T \mathbf{x_i}}})\right) \end{align} }[/math]

Our goal is to find the [math]\displaystyle{ \beta\, }[/math] that maximizes [math]\displaystyle{ {\ell(\mathbf{\beta\,})} }[/math]. We use calculus to do this ie solve [math]\displaystyle{ {\frac{\partial \ell}{\partial \mathbf{\beta\,}}}=0 }[/math]. To do this we use the famous numerical method of Newton-Raphson. This is an iterative method were we calculate the first & second derivative at each iteration.

The first derivative is typically called the score vector.

[math]\displaystyle{ \begin{align} S(\beta\,) {}= {\frac{\partial \ell}{ \partial \mathbf{\beta\,}}}&{} = \sum_{i=1}^n \left(y_i \mathbf{x_i} - \frac{e^{\mathbf{\beta\,}^T \mathbf{x}}}{1+e^{\mathbf{\beta\,}^T \mathbf{x}}} \mathbf{x_i} \right) \\[8pt] \end{align} }[/math]

The negative of the second derivative is typically called the information matrix.

[math]\displaystyle{ \begin{align} I(\beta\,) {}= -{\frac{\partial \ell}{\partial \mathbf {\beta\,} \partial \mathbf{\beta\,}^T}}&{} = \sum_{i=1}^n \left(\mathbf{x_i}\mathbf{x_i}^T (\frac{e^{\mathbf{\beta\,}^T \mathbf{x}}}{1+e^{\mathbf{\beta\,}^T \mathbf{x}}})(\frac{1}{1+e^{\mathbf{\beta\,}^T \mathbf{x}}}) \right) \\[8pt] \end{align} }[/math]

We then use the following update formula to calcalute continually better estimates of the optimal [math]\displaystyle{ \beta\, }[/math]. It is not typically important what you use as your initial estimate [math]\displaystyle{ \beta\,^{(1)} }[/math] is.

[math]\displaystyle{ \beta\,^{(r+1)} {}= \beta\,^{(r)} + I^{-1}(\beta\,^{(r)} )S(\beta\,^{(r)} ) }[/math]

Matrix Notation

let [math]\displaystyle{ \mathbf{y} }[/math] be a (n x 1) vector of all class labels. This is called the response in other contexts.

let [math]\displaystyle{ \mathbb{X} }[/math] be a (n x (d+1)) matrix of all your features. Each row represents a data point. Each column represents a feature/covariate.

let [math]\displaystyle{ \mathbf{p}^{(r)} }[/math] be a (n x 1) vector with values [math]\displaystyle{ P(\mathbf{x_i} |\beta\,^{(r)} ) }[/math]

let [math]\displaystyle{ \mathbb{W}^{(r)} }[/math] be a (n x n) diagonal matrix with [math]\displaystyle{ \mathbb{W}_{ii}^{(r)} {}= P(\mathbf{x_i} |\beta\,^{(r)} )(1 - P(\mathbf{x_i} |\beta\,^{(r)} )) }[/math]

we can rewrite our score vector, information matrix & update equation in terms of this new matrix notation.

[math]\displaystyle{ \begin{align} S(\beta\,^{(r)}) {}= {\frac{\partial \ell}{ \partial \mathbf{\beta\,}}}&{} = \mathbb{X}^T(\mathbf{y} - \mathbf{p}^{(r)})\end{align} }[/math]

[math]\displaystyle{ \begin{align} I(\beta\,^{(r)}) {}= -{\frac{\partial \ell}{\partial \mathbf {\beta\,} \partial \mathbf{\beta\,}^T}}&{} = \mathbb{X}^T\mathbb{W}^{(r)}\mathbb{X} \end{align} }[/math]

[math]\displaystyle{ \beta\,^{(r+1)} {}= \beta\,^{(r)} + I^{-1}(\beta\,^{(r)} )S(\beta\,^{(r)} ) {}= \beta\,^{(r)} + (\mathbb{X}^T\mathbb{W}^{(r)}\mathbb{X})^{-1}\mathbb{X}^T(\mathbf{y} - \mathbf{p}^{(r)}) }[/math]

Iteratively Re-weighted Least Squares

If we reorganize this updating formula we can see it is really a iteratively solving a least squares problem each time with a new weighting.

[math]\displaystyle{ \beta\,^{(r+1)} {}= \beta\,^{(r)} + (\mathbb{X}^T\mathbb{W}^{(r)}\mathbb{X})^{-1}(\mathbb{X}^T\mathbb{W}^{(r)}\mathbb{X}\beta\,^{(r)} + \mathbb{X}^T(\mathbf{y} - \mathbf{p}^{(r)})) }[/math]

[math]\displaystyle{ \beta\,^{(r+1)} {}= \beta\,^{(r)} + (\mathbb{X}^T\mathbb{W}^{(r)}\mathbb{X})^{-1}\mathbb{X}^T\mathbb{W}^{(r)}\mathbf(z)^{(r)} }[/math]

where [math]\displaystyle{ \mathbf{z}^{(r)} = \mathbb{X}\beta\,^{(r)} + (\mathbb{W}^{(r)})^{-1}(\mathbf{y}-\mathbf{p}^{(r)}) }[/math]

Fisher Scoring Method

Fisher Scoring is a method very similiar to Newton-Raphson. It uses the expected Information Matrix as opposed to the observed information matrix. This distinction simplifies the problem and in perticular the computational complexity. To learn more about this method & logistic regression in general you can take Stat431/831 at the University of Waterloo.