Learn a Line
Train a neuron to learn line equation
To learn neural networks (aka deep learning), one must start with the smallest unit - a neuron. In this project, you’ll learn the basics of a neuron by training it on a straight line equation: y = mx + c
Preface:
It’s good to have a machine learning background to get started with deep learning, but it is not necessary! All you need to know to get started at the moment is that you send in some input numbers and out comes some numbers. The training part is to iteratively make sure that we get the right outputs.
Alright, so when it comes to learning a straight line, your inputs would be x values [x1, x2, x3, ...]
and the outputs should be [y1, y2, y3, ...].
I used “should be” because the neural networks may predict erroneous outputs [y1 + e1, y2 + e2, y3 + e3, ...].
So the goal of training is to minimize the error, that is, e1 = 0, e2 = 0, e3 = 0 ….
Now, all you need to know is m and c to predict the correct y values.
In deep learning terms, m and c are called weights or parameters.
Here’s the task we are trying to solve:
Given a bunch of (x, y) pairs on a line [(x1, y1), (x2, y2), ...],
we must find the parameters m and c.
You must’ve learnt in your school how to find the slope m of a straight line using 2 points
(x1, y1) and (x2, y2) using the formula m = (x2 - x1) / (y2 - y1).
And how would you find the y intercept? - by substituting x = 0 in the line equation y = mx + c.
Cool, instead of using these formulas, let's use a neuron!