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

 
Core Concepts of Machine Learning
Today, there are several kinds of ML, but the notion of ML is mainly based
on three components "representation", "evaluation", and "optimization".
Here are some of the standard concepts that apply to all of them:
Representation
Machine learning models can not directly hear, see, or sense
input examples. Data representation is therefore needed to provide a helpful
vantage point for the model in the main data attributes. The choice of
significant characteristics that best represent data is very essential to train a
machine learning model effectively. “Representation” simply refers to the
act of “representing” data points to a computer in a language that it
understands using a set of classifiers. A classifier may be defined as "a
model that inputs a vector of discrete and/or ongoing function values and
outputs a single discrete value called “class". To learn from the represented
data, a model must have the desired classifier in the training data set or
"hypothesis space" that you want the models to be trained on. The data
features used to represent the input are very critical to the machine learning
system. Any "classifier" that is external to the hypothesis space cannot be
learned by the model. For developing a required machine learning model,

---

data characteristics are so essential that it can easily be the difference
between successful and unsuccessful machine learning projects.
A training data set with several independent “features” which are well
linked to the "class" can make learning much easier for the machine. On the
other side, it may not be easy for the machine to learn from the class with
complex functions. This often requires the processing of the raw data so
that the desired features for the ML model can be built from it. The method
of deriving features from raw data set tends to be the ML project's most
time-consuming and laborious component. It is also considered to be the
most creative and interesting part of the project where intuition and "trial
and error" play just as important a role as the technical requirements. The
ML process is not a "one-shot" process of developing and executing a
training data set, but an iterative process requiring analysis of the post-
execution output, followed by modification of the training data set. Domain
specificity is another reason why the training dataset requires
comprehensive-time and effort. Training data set to produce predictions
based on consumer behavior analysis for an e-commerce platform will be
very distinct from the training data set needed to create a self-driving car.
Nevertheless, in the industrial sectors, the core machine learning
mechanism stays the same. No wonder, there is a lot of research going on to
automate the process of feature engineering.