## General
is an algorithm for binary classification, output either is or isn't
![[CleanShot 2024-06-05 at
[email protected]|300]]
- model is $\sigma(w^Tx + b) = \hat{y}$
![[CleanShot 2024-06-05 at
[email protected]|300]]
- $\sigma(z) = \frac{1}{1 + e^{-z}}$
- Model: Given $x$, we want $\hat{y} = P(y=1|x)$, $x \in \mathbb{R}^{n_x}$
- Parameters: $w \in \mathbb{R}^{n_x}$, $b \in \mathbb{R}$
- $n_x$ is number of input features
- [[Linear Regression]] is poor at binary classification, since we need a classification between 0 & 1 for output, not an infinite ranging positive value
- so we use logistic regression
## Measuring Success
[[Loss Function]]:
- measures how close our prediction $\hat{y}$ is to the real labeled output $y$
- we use the [[Binary Cross Entropy Loss|Binary Cross Entropy Loss Function]]
- $L(\hat{y},y) = -(y\,log(\hat{y})) + (1 - y)log(1 - \hat{y}))$
- mean squared error doesn't work that well, leads to non-convex optimization problem, so we reach bad local minima
[[Cost Function]]:
$J(w, b) = - \frac{1}{m} \sum^{m}_{i=1} y^{(i)}log(\hat{y}^{(i)}) + (1 - y^{(i)})log(1 - \hat{y}^{(i)})$