# Logistic Regression
Given `x`, we want `y^ = P(y=1 | x)`
- If `x` is a picture, we want `y^` to tell us if it is a cat (probability value between 0 and 1)
- Input parameters: `w` (an `nx` dimensional vector) and `b`, a real number
- Those will be kept separate
-
- Output: `y^ = sigmoid(wTx + b)`, where `WTx + b` is `z`
- ![[20230822171939 function.png]]
- Where `w^Tx` is the `X` matrix transposed to `W` (the input of the training example) plus b
* Given { (x^1, y^1), ... (x^m, y^m) }, we want `^y^(i) ~= y^(y)`. That is, after training, we want to find the same answer as the training set.
To train a logistic regression model, we need to define a loss/error function: `L`
* It is the function we use to measure the performance of our regression
* It needs to help gradient descent (avoid local maximas)
* For the loss function, a squared error function will create trouble as it has local maxima and can thus cause issues for gradient descent
* What we use:
* ![[20230822173030 function.png]]