Lets enter into the world of Machine learning! This is introductory post. You may skip this one to dive into direct technical part. Although, I would recommend you to scan through this one as well so that it can get you clear, in what directions lessons would proceed.
A machine learning involves tools or algorithms which are data driven. Their primary work is to guess based on past/training data provided to them. Unlike conventional algorithms, their output is data driven.
For example, you can have a price prediction model for apartments. Train them with past trends and prices. Next you ask the algorithm what might be price in next five years! Or you could train a system with 100 thousands of spam emails and then filter out new messages based on analysis done by machine.
In short, we train machine with huge test data and then ask for the result to input that we don’t know.
Trust Issues. How do we decide the correctness then?
Accuracy score: Accuracy score is value with which we determine the correctness of any machine learning algorithm.
Its ratio of total values predicted correct to the total input values. So to calculate the accuracy score, we set aside a portion of training set (whose input and output we know!) lets say 10% of it. We train model with 90% of data and ask for predictions of rest 10%. Next we match with actual answers we have. More the accuracy more the reliable model is.
Note: Throughout the series we shall use python as coding language. So you may encounter the snippets in python while explaining the stuff.
This is first and foremost parameter to gauge how good your model is.
Two broad Categories
Machine learning involves two broad categories: Supervised and Unsupervised learning. Initially in this blog series we shall dive into supervised learning and then explore the unsupervised learning.
Supervised learning is one where we provide a model with set of input and the output related to it ( the training model) and then later the machine refers this training set to predict the value for input asked.
On contrary, if we only provide a machine with set of inputs, and let machine figure out all the relations, features and behavior, falls under Unsupervised learning.
Hmm.. Supervised learning first. So Where and How do we start then?
The first stepping stone in supervised learning is to gain knowledge about Linear Regression.