Machine Learning: 2 Books in 1: Machine Learning for Beginners, Machine Learning Mathematics. An Introduction Guide to Understand Data Science Through the Business Application

Prediction and Inference
If there are several inputs "X" easily accessible, but the output "B"
production is unknown, "f" is often treated as a black box, provided that it
generates accurate predictions for "Y". This is called "prediction". There are
circumstances in which we need to understand how "Y" is influenced as
"X" changes. We want to estimate "f" in this scenario, but our objective is
not simply to generate predictions for "Y". In this situation, we want to
establish and better understand the connection between "Y" and "X". Now
"f" is not regarded as a black box since we have to understand the
underlying process of the system. This is called "inference". In everyday
life, various issues can be categorized into the setting of "predictions", the
setting of "inferences", or a "hybrid" of the two.
Parametric and Non-parametric Techniques
The “parametric technique” can be defined as an evaluation of “f” by
calculating the set parameters (finite summary of the data) while

---

establishing an assumption about the functional form of “f”. The
mathematical equation of this technique is “f(X) = β0 + β1X1 + β2X2 + . . .
+ βpXp”. The “parametric models” tend to have a finite number of
parameters which is independent of the size of the data set. This is also
known as “model-based learning”. For example, “k-Gaussian models” are
driven by parametric technique.
On the other hand, “non-parametric technique” generates an estimation of
“f” on the basis of its closeness to the data points, without making any
assumptions on the functional form of “f”. The “non-parametric models”
tend to have a varying number of parameters which grown proportionally
with the size of the data set. This is also known as “memory-based
learning”. For example, “kernel density models” are driven by a non-
parametric technique.