Whenever you are reading a machine learning paper, you may encounter this kind of words: “our model is non-parametric…” and wondering what on earth is non-parametric? What is its advantage over parametric model?
For parametric model, data is assumed to fit a normal distribution with unknown parameters μ (population mean) and σ² (population variance), which are then estimated using the sample mean and sample variance. The sample mean and sample variance are those you are going to get when you do sampling from population data. To use parametric model, one should ensure each element of data points relates to the prediction, and there’s almost no outlier — so that it can be assumed under normal distribution.
For non-parametric model, it does not assume that data is drawn from a normal distribution. The model structure depends on data. Such model’s number and nature of parameters are flexible and not fixed in advance. As its characteristic of data-driving and being able to deal with the data made up from mixing distributions, the non-parametric model are widely-used in more realistic scenario.