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

1. 
Problem Definition
In this stage, the business problem that needs to be resolved using a
machine learning model will be identified and documented with all
pertinent details.
2. 
Data Ingestion
The first stage of any machine learning workflow is to channel input data
into a database server. The most important thing to remember is that the
data is ingested raw and with no modification to enable us to have an
invariable record of the original dataset. Data may be supplied from a
variety of sources which can be acquired either through request or
transmitted from other systems.
“NoSQL document databases" are best suited to store huge amount of
defined and labeled as well as unorganized raw data, that are quickly
evolving as they do not need to adhere to a predefined scheme.
It even provides a "distributed, scalable and replicated data storage".
“Offline”
Data will flow in the "offline" layer to the raw data storage through an
"Ingestion Service", which is a "composite orchestration service that is
capable of encapsulating the data sourcing and persistence". A repository
model is used internally to communicate with a data service that will
interact with the data storage in exchange. When you save the data in the
database, a unique batch Id will be given to the dataset, which allows for

---

the effective query of the data as well as end-to-end tracking and
monitoring of the data.
To be computationally efficient, the ingestion of the data is distributed into
two folds.
The first one is a specific pipeline for every dataset so that
each of the datasets can be processed individually and
simultaneously. 
The second aspect is that within each pipeline data can be divided
to make the best of a variety of server cores, processors and
perhaps even the entire server.
Distributing the prepping of data across several vertical and horizontal
pipelines will reduce the total time required to perform the tasks.
The "ingestion service" would run at regular intervals based on a predefined
schedule (one or more times a day) or upon encountering a trigger. A
subject will decouple producers (data source) from processors, which would
be the data pipeline for this example, so when the source data is collected,
the "producer system" will send a notification to the "broker" and
subsequently the "embedded notification service" will respond by
inducing ingestion of the data. The "notification service" would also inform
the "broker" that the processing of the original dataset was completed with
success and now the dataset is being stored in the database.
“Online”

---

The "Online Ingestion Service" makes up the entrance to the "streaming
architecture" of the online layer, as it would decouple and manage the data
flow from the source to the processing and storage components by offering
consistent, high performance, low latency functionalities. It also works as
an enterprise-level "Data Bus". Data would be stored on a long-term "Raw
Data Storage", which also serves as a mediating layer to the subsequent
online streaming service for further processing in real-time. For instance,
such techniques that are utilized in this case may be "Apache Kafka
(pub/sub messaging system)" and "Apache Flume (data collection to the
long-term database)". A variety of other similar techniques are available
and can be selectively applied based on the technology stack of the
business.