1- Binary Classification:
Hypothesis space is depending on the how the inputs are arranged in 2D space. The goal is to separate two classes, so the simple hypothesis space is the class of lines (different lines that we choose the best one among them). This is the preferred hypothesis for us but it can also be a circle (one class inside the circle and another class is outside the circle) or n-dimensional lines.
In order to ease the process of mathematical calculations, sometimes we prefer to move the input data to the space that let us separate the data with line.
😎2- Multi-Class Classification:
Multi-Class classification is just like the binary classification with this difference that instead of two classes we have p classes. Therefore, the hypothesis space again can be similar, a set of lines or curved lines.
Note:
we can have one single global nonlinear model or we can extend the binary classification to multi-class classification by employing one nonlinear model per two-class sub-problems.
3- Regression:
In Regression, the goal is to find a function that goes through the data. So depends on the shape of input data, it can be a line (linear regression) or polynomial. The trade-off between complexity and flexibility of our model is the factor to choose the degree of polynomial and so our hypothesis space.
😇