PDF is defined by a Probability Distribution curve. PDF can be thought of like, given a sample data points, the PDF squashes the points to get a model (curve or distribution curve) that can be used to define the total samples. We can select a pdf train it on our dataset. find the pdf parameters and then use the model to predict future events. So PDF is a function p(x) that returns probability of x and uses parameters we obtained from the population.
We have freedom to choose what to use for p(x). There are multiple density functions you can choose from.
- Bernoulli's Distribution
- Poission's Distribution
- Normal Distribution
- Bionomial Distribution
- Negative Binomial Distribution
Which distributions to use ? Well, it depends on what kind of datasets we have.